Enterprise resource planning (ERP) has long served as the operational backbone of modern businesses. It connects financial management, procurement, inventory, sales, manufacturing, projects, and other core processes within a structured system of record.

Artificial intelligence (AI) is beginning to expand that role.

Instead of using ERP primarily to record what has already happened, businesses can increasingly use AI to interpret enterprise data, identify patterns, predict potential outcomes, generate insights, recommend actions, and support increasingly automated workflows.

This is creating a broader evolution:

SYSTEM OF RECORD → SYSTEM OF INSIGHT → AI-ENABLED ERP → SYSTEM OF ACTION

The progression does not mean that every ERP system has become autonomous, nor that AI replaces the ERP foundation. Rather, AI is extending what businesses can do with the data, processes, and business logic already contained within their enterprise systems.

For executives, the question therefore goes beyond whether an ERP platform has AI features.

The more important question is:

Is your ERP environment ready to create measurable business value from AI?

That question matters because the value of AI depends on more than the AI technology itself. Reliable data, connected architecture, well-defined processes, security, governance, and organisational readiness all influence whether AI can be applied effectively at scale.

This article examines how AI is transforming ERP, where businesses can apply it today, how AI agents are changing automation and ERP implementation, what an AI-ready ERP foundation looks like, and how organisations can evaluate the business case before investing.

Key Takeaway

• AI does not make ERP less important. It can make the quality of the ERP foundation more important.
• AI extends ERP beyond transaction processing. It can help businesses move from recording historical activity toward prediction, insight, recommendations, and increasingly automated action.
• AI agents introduce a new layer of automation. They can assist with or, within controlled boundaries, execute multi-step business workflows based on defined permissions, business rules, and human oversight.
• The ERP foundation still matters. Data quality, system integration, process consistency, architecture, security, and governance can directly influence the value AI can deliver.
• AI adoption should start with business problems, not AI features. Organisations should identify where AI can address measurable operational challenges and create quantifiable business value.
• AI does not automatically require ERP replacement. Depending on the existing environment, businesses may improve, integrate, modernise, or eventually replace their ERP based on their actual requirements and constraints.
• The objective is measurable business value. The goal is to create an ERP environment that can turn enterprise data into better insight, better decisions, and better business outcomes.

---

What Is AI in ERP?

Artificial intelligence (AI) in ERP refers to the integration of AI capabilities with enterprise resource planning systems, business data, and operational workflows to help organisations analyse information, identify patterns, generate insights, predict potential outcomes, recommend actions, and automate selected activities.

Traditional ERP systems are primarily designed to record, process, and manage structured business transactions. AI extends those capabilities by enabling software to interpret information and support activities that traditionally required more manual analysis or human judgement.

Depending on the technology and business use case, AI in ERP can support:

• Data analysis — examining large volumes of financial, operational, customer, and supply-chain data.
• Pattern recognition — identifying anomalies, trends, relationships, or unusual activity.
• Prediction and forecasting — estimating potential outcomes such as demand, cash requirements, or operational risks.
• Natural-language interaction — allowing users to ask questions about business information using conversational language.
• Content generation — producing summaries, explanations, or other business content from available enterprise information.
• Recommendations — helping users identify potential actions based on patterns, rules, and contextual information.
• Workflow automation — reducing manual work in defined business processes.
• Agent-based execution — enabling AI agents to perform or coordinate multiple steps within authorised workflows, subject to appropriate controls.

The exact capabilities vary between ERP platforms and AI implementations. Gartner’s recent research describes ERP vendors as increasingly introducing AI-driven and agentic capabilities that change how ERP processes are executed and managed, while also highlighting the need for appropriate roles, governance, and control.

Traditional ERP vs. AI-Enabled ERP

The difference is not that traditional ERP is suddenly obsolete. Rather, AI adds another layer of intelligence to the transactional and operational foundation already provided by ERP.

Traditional ERP	AI-Enabled ERP
Records transactions	Interprets business data
Provides reports	Generates contextual insights
Uses predefined rules for automation	Can use AI to identify patterns and recommend actions
Users initiate most workflows	AI can assist with workflow initiation and execution
Primarily describes historical activity	Can support predictive analysis and forecasting
Users interact primarily through structured screens and reports	Users can increasingly interact through natural-language interfaces
Executes defined business processes	Can assist with more adaptive, multi-step workflows

Modern AI-enabled ERP therefore should not be understood simply as “ERP with a chatbot.” The more significant change is the ability to connect intelligence with the operational context of the business.

For example, an AI capability could analyse financial or operational records, identify an unusual pattern, explain what may have caused it, and recommend a next step. In more advanced implementations, an AI agent may be able to carry out defined actions within an authorised workflow.

The underlying ERP remains important because the AI needs reliable business context to produce useful results. Gartner identifies data quality and integration complexity among the challenges organisations face when adopting embedded AI in ERP.

What AI Adds to ERP

The evolution can be understood as a progression:

RECORD → UNDERSTAND → PREDICT → RECOMMEND → ACT

Record
ERP captures transactions and operational events such as invoices, purchases, sales orders, inventory movements, production activity, project costs, and financial entries.

Understand
AI can analyse those records to identify relationships, patterns, anomalies, and context that may not be immediately visible through conventional reporting.

Predict
Machine learning and other predictive techniques can use historical and current information to support forecasting and risk identification.

Recommend
AI can turn analysis into recommendations, helping users determine where attention or intervention may be required.

Act
With appropriate permissions and controls, automation and AI agents can increasingly support or execute defined actions within business workflows.

This progression does not mean that every ERP system will become fully autonomous. The level of AI involvement depends on the use case, technology, business process, risk level, and governance model.

The direction of travel is nevertheless becoming clearer: ERP is moving beyond simply recording what happened toward helping organisations understand what is happening, anticipate what may happen next, and determine what actions could be taken.

Gartner’s 2026 research describes this broader evolution through AI-driven automation, adaptive analytics, AI-driven planning and forecasting, and agentic capabilities in cloud ERP.

What AI Does Not Change

Despite these new capabilities, AI does not eliminate the fundamental responsibilities of an ERP system.

ERP remains the operational foundation for areas such as:

• Core business transactions
• Financial records
• Master data
• Business rules
• Process control
• Operational records
• Auditability
• Access and permissions

This distinction is important.

AI can analyse an invoice, but the ERP still needs to maintain the underlying financial transaction.

AI can identify an inventory trend, but the ERP remains responsible for inventory records and related transactions.

AI can recommend an action, but the organisation still needs defined business rules, permissions, controls, and accountability around whether and how that action is executed.

In other words, AI does not replace the ERP’s role as the structured operational foundation of the business. It adds intelligence around that foundation.

This is also why the quality of the underlying ERP environment matters. McKinsey’s 2026 analysis of AI and ERP emphasises that data, application architecture, and the system-of-record foundation remain critical even as AI and agentic systems increasingly operate on top of enterprise applications.

A Simple AI + ERP Architecture

At a high level, AI-enabled ERP can be understood through the following flow:

Business Data → ERP Core → AI Capabilities → Insights / Recommendations → Business Actions

The ERP core provides the operational context: transactions, master data, business processes, rules, and records.

The AI layer can apply different capabilities depending on the use case, including predictive AI, generative AI, intelligent automation, or AI agents.

The resulting insights or recommendations can then be presented to users or connected to controlled workflows.

Finally, business actions can be taken by employees, automated processes, or AI agents operating within defined permissions and governance.

For example:

Sales Orders + Inventory + Customer Data → ERP → AI Analysis → Demand / Fulfilment Insight → Recommended Action → Controlled Workflow

The architecture can vary significantly between ERP platforms. Some AI capabilities are embedded directly within the ERP, while others may operate through connected services and integrations. What matters is not simply where the AI resides, but whether it can access the appropriate business context securely and produce outputs that can be used within real operational workflows.

Modern ERP vendors are increasingly embedding AI directly into these workflows. For example, Acumatica describes its AI capabilities as operating across financial, operational, customer, project, and inventory workflows, including anomaly detection, document recognition, conversational analysis, automation, and AI agents.

The important distinction is therefore not simply ERP versus AI.

It is ERP as the business foundation, with AI extending what organisations can understand, predict, recommend, and automate from that foundation.

---

Why AI Is Changing the Role of ERP

For decades, ERP has primarily been the system businesses rely on to record transactions, manage processes, maintain financial and operational data, and provide a common source of information across the organisation.

AI is changing what businesses can do with that foundation.

The shift is not simply from traditional ERP to “AI-powered ERP.” It is a broader change in how organisations interact with enterprise information and how decisions and workflows are supported.

The evolution can be viewed as:

SYSTEM OF RECORD → SYSTEM OF INSIGHT → AI-ENABLED ERP → SYSTEM OF ACTION

This does not mean that every ERP system is becoming autonomous. The ERP foundation remains important for maintaining business records, enforcing rules, and providing the operational context that AI needs. McKinsey’s 2026 analysis similarly argues that even as AI agents increasingly mediate enterprise interactions, the underlying ERP data, application logic, and system-of-record capabilities remain important for reliability, auditability, compliance, and consistency.

From System of Record to System of Insight

Traditional ERP is designed primarily to capture and process business activity.

A sales order becomes a transaction.
A supplier invoice becomes an accounts-payable record.
A production order becomes an operational record.
A project cost becomes part of project accounting.
A customer payment becomes a financial transaction.

These records are essential, but recording information is only the beginning.

AI can analyse information across those records to identify relationships, patterns, anomalies, and trends that may require attention.

For example, an ERP environment may contain information about:

• sales orders and customer demand;
• inventory levels and movements;
• supplier performance;
• purchasing activity;
• production schedules;
• project costs;
• accounts receivable and payable;
• cash movements; and
• historical financial performance.

AI can analyse these data points in combination rather than requiring users to examine each report independently.

The result is a shift from asking:

“What happened?”

toward questions such as:

“What is changing?”
“Why is it changing?”
“What could happen next?”
“What requires attention?”

This is one reason data quality and integration have become central considerations in AI-enabled ERP. Gartner identifies integration and data quality among the key challenges ERP leaders need to address when activating embedded AI.

The ERP therefore remains the system of record, while AI can increasingly operate as a layer of intelligence around that record.

From Reporting to Prediction

Traditional reporting is primarily retrospective. It helps organisations understand what has already happened.

AI can extend this capability toward prediction and scenario analysis.

Depending on the quality, availability, and relevance of the underlying data, AI can support areas such as:

Business Area	Traditional Approach	AI-Enabled Capability
Demand	Review historical sales	Identify demand patterns and support forecasting
Cash Flow	Review historical cash movements	Identify trends and support cash-flow forecasting
Inventory	Monitor current stock levels	Identify demand patterns and potential inventory risks
Production	Review schedules and capacity	Analyse constraints and support production planning
Project Cost	Compare actual vs. budget	Identify cost trends and potential overruns
Supply Chain	Monitor supplier and logistics data	Identify patterns that may indicate disruption or risk

The important distinction is that AI does not guarantee a correct prediction.

Forecasting remains dependent on the quality and relevance of the available data, the model or analytical approach used, the business context, and the conditions under which the prediction is applied.

The value therefore comes from improving the organisation’s ability to identify signals earlier and respond with better information.

Gartner’s 2026 research on cloud ERP describes AI-driven planning and forecasting as an emerging capability that can support scenario modelling, predictive analytics, and risk management. At the same time, Gartner notes that data quality, integration complexity, and skills gaps remain barriers to adoption.

From Automation to Orchestration

Another important change is the progression from rule-based automation toward increasingly intelligent workflow orchestration.

The distinction can be illustrated simply:

Traditional automation

IF X → DO Y

A predefined condition triggers a predefined action.

For example:

If an invoice exceeds a specified threshold, route it to the appropriate approval workflow.

This approach remains valuable because predictable, repeatable processes can often be automated efficiently with clear business rules.

AI-assisted workflows introduce another layer:

ANALYSE → RECOMMEND

Instead of only responding to a predefined condition, AI can analyse information and help determine what may require attention.

For example:

Analyse purchasing and supplier information → identify an unusual price change → explain the pattern → recommend that a buyer review the supplier.

The next progression is the AI agent:

UNDERSTAND → REASON → EXECUTE

Within defined permissions and controls, an AI agent can potentially interpret a business objective, use relevant enterprise information, determine the appropriate sequence of steps, and execute authorised actions.

For example:

Detect a potential supply issue → review relevant purchase and inventory information → identify affected orders → recommend or initiate defined actions → escalate exceptions to a human.

This is what makes agentic AI particularly relevant to ERP. The opportunity is not simply to generate an answer but to connect intelligence with the workflow in which the business decision needs to be made.

McKinsey describes this development as AI agents operating on top of traditional enterprise applications to automate decisions and orchestrate processes, while emphasising that the underlying ERP foundation remains important.

However, orchestration does not mean unrestricted autonomy.

The level of authority given to an AI system should depend on the risk and nature of the workflow. Financial transactions, supplier commitments, customer communications, production changes, and other consequential actions may require approval thresholds, role-based permissions, exception handling, monitoring, and audit trails.

The practical objective is therefore not:

“Let AI do everything.”

It is:

“Apply the appropriate level of intelligence and automation to the appropriate business process, with appropriate controls.”

Why the ERP Foundation Becomes More Important

As AI becomes more capable, it may be tempting to think that the underlying ERP matters less.

The opposite can be true.

AI needs reliable information and business context to operate effectively. If critical information is incomplete, inconsistent, fragmented across systems, or difficult to access, AI may have limited context from which to generate useful insights or recommendations.

The foundation can be viewed through six dimensions:

Dimension	Why It Matters for AI
Data	AI needs accurate, relevant, accessible, and governed business data
Architecture	AI needs systems and interfaces that can exchange information reliably
Processes	AI needs clearly defined workflows, rules, and business context
Security	AI access must respect roles, permissions, privacy, and security requirements
People	Employees need the skills and confidence to work with AI-enabled processes
Governance	Organisations need accountability, monitoring, controls, and oversight

This is not simply a theoretical concern.

Gartner has identified poor data quality as a recurring barrier to advanced analytics and AI deployment and has emphasised the need for trusted, reusable, AI-ready data.

McKinsey’s ERP research makes a similar point from an architecture perspective: AI initiatives can struggle to scale when the underlying end-to-end processes, data, people, and technology capabilities are not prepared to support them.

This leads to an important strategic principle:

AI does not eliminate ERP complexity. It increases the importance of having a reliable enterprise foundation.

An organisation with fragmented systems, inconsistent master data, heavily customised processes, or weak governance may still be able to deploy individual AI tools. The question is whether those tools can produce reliable, scalable, and measurable value across the business.

ERP Becomes the Context Layer for AI

The future role of ERP is therefore not necessarily smaller because AI is becoming more capable.

It can become broader.

ERP provides the structured business context in which AI can operate:

Business Data + Business Rules + Business Processes + AI

Together, these components can enable a progression from:

Record → Understand → Predict → Recommend → Act

The interaction between users and ERP may also change. Instead of navigating multiple screens to retrieve information, users may increasingly interact with AI capabilities that interpret their intent and retrieve or act on relevant enterprise information.

But the underlying system still needs to maintain the authoritative records, business rules, permissions, and auditability required to operate the business.

McKinsey describes this as a potential shift from screen-based transactions toward AI agents that mediate, decide, and execute, while retaining the ERP backbone for data consistency, business rules, auditability, and compliance.

The strategic question for organisations is therefore no longer simply whether their ERP has AI features.

It is whether their ERP environment provides the data, architecture, processes, security, people, and governance required to turn AI capabilities into useful business outcomes.

That question becomes even more important as AI moves from isolated assistants toward embedded intelligence and agentic workflows.

---

How Does AI Work Inside an ERP System?

AI does not operate in isolation from an ERP system.

For AI to produce useful business outputs, it needs access to relevant data, sufficient business context, appropriate AI capabilities, and a controlled way to turn its outputs into business actions.

At a simplified level, the architecture can be understood through four layers:

BUSINESS DATA → ERP CORE → AI CAPABILITIES → INSIGHTS, RECOMMENDATIONS & ACTIONS

Each layer has a different role.

The data provides the information.
The ERP provides the business context.
AI provides intelligence.
The action layer connects that intelligence back to the business process.

This distinction is important because an AI model can generate an answer without necessarily understanding how a particular organisation operates. The ERP environment provides the context that helps connect AI capabilities to actual transactions, processes, rules, and business relationships.

McKinsey’s 2026 research on ERP and AI similarly describes a model in which AI agents operate on top of enterprise applications while relying on the underlying ERP data, application logic, and system-of-record capabilities.

The AI + ERP Architecture

A simplified AI-enabled ERP architecture looks like this:

1. Business Data
   ↓
2. ERP Core
   ↓
3. AI Capabilities
   ↓
4. Insights, Recommendations & Actions

The layers are connected rather than independent.

For example, consider an inventory planning scenario:

Inventory & Sales Data → ERP Inventory Context → AI Forecasting → Demand Insight → Planning Recommendation → Approved Inventory Action

The AI is not simply analysing an isolated spreadsheet. It is working with business information that has meaning within the ERP environment.

This is also why data quality and integration are fundamental to AI-enabled ERP. Gartner identifies integration and data quality as important considerations when organisations activate embedded AI in ERP, while its 2026 research specifically highlights ERP data readiness as a factor in AI outcomes.

Business Data Layer

The first layer is the information AI can access and analyse.

ERP environments can contain several types of business data.

Transactional Data

Transactional data represents business events as they occur.

Examples include:

• Sales orders
• Purchase orders
• Invoices
• Payments
• Inventory movements
• Production orders
• Timesheets
• Project transactions
• Expense claims

This data provides a record of what has happened within the business.

Master Data

Master data provides the entities and definitions used across business processes.

Examples include:

• Customers
• Suppliers
• Products and items
• Employees
• Warehouses
• Accounts
• Projects
• Locations

If customer, supplier, product, or other master data is inconsistent, AI may have difficulty interpreting relationships correctly.

Historical Data

Historical information allows AI capabilities to analyse patterns over time.

Examples include:

• Previous sales
• Historical purchasing behaviour
• Past inventory levels
• Previous production performance
• Historical project costs
• Prior payment behaviour

Historical data can support forecasting and pattern recognition, but more historical data does not automatically mean better predictions. Relevance, quality, consistency, and context matter.

Operational Data

Operational data describes the current state of business activities.

For example:

• Current inventory availability
• Open orders
• Production schedules
• Outstanding receivables
• Supplier status
• Project progress
• Current resource utilisation

This information can help AI interpret what is happening now rather than relying only on historical patterns.

External and Contextual Data

Some AI use cases may also require information outside the ERP.

Depending on the business problem, this might include:

• Market information
• Weather data
• Commodity prices
• Logistics information
• External customer information
• Industry data
• Regulatory information

The relevant data depends on the use case. Not every AI capability needs external data.

The objective is not to connect every possible data source to AI. It is to provide the right data for the specific business problem.

ERP Core: Where Business Context Lives

The second layer is the ERP core.

This is where business information is connected to the processes and relationships that give it meaning.

An ERP does more than store individual transactions. It connects information across business functions and applies the structures, rules, and workflows used to operate the organisation.

For example, an ERP can establish relationships between:

Customer → Sales Order → Inventory → Shipment → Invoice → Payment

Or:

Supplier → Purchase Order → Receipt → Invoice → Payment

Or:

Project → Budget → Resources → Costs → Billing → Profitability

Those relationships provide context that an AI system can use when analysing business activity.

The ERP can also provide:

• Business rules
• Process definitions
• Transaction status
• User roles and permissions
• Organisational structures
• Approval workflows
• Financial dimensions
• Product and customer relationships
• Operational dependencies

This context is critical.

Consider an AI system that identifies an unusually large purchase order. On its own, the number may appear abnormal.

But within the ERP context, the system may know that:

• the purchase is linked to a major project;
• the item is normally purchased in large quantities;
• the supplier has an existing contract;
• the purchase has already been approved; or
• the order is associated with a planned production requirement.

The same transaction can therefore have a completely different meaning depending on the business context available to the AI.

McKinsey’s 2026 research describes a similar principle through the importance of ERP data, application logic, business rules, and a shared business ontology in enabling AI agents to operate consistently across enterprise processes.

AI Capabilities Layer

The third layer is where different forms of AI can be applied to the business context.

Not every AI capability performs the same function.

Predictive AI

Predictive AI analyses patterns in available data to support forecasts, estimates, and risk identification.

Examples include:

• Demand forecasting
• Cash-flow forecasting
• Inventory forecasting
• Customer payment prediction
• Production forecasting
• Risk detection

The objective is generally to estimate what may happen or identify signals that require attention.

Generative AI

Generative AI can create or transform content based on information and instructions.

Within an ERP environment, this may support activities such as:

• Summarising financial information
• Explaining business trends
• Generating management summaries
• Drafting communications
• Answering questions about business information
• Converting complex information into natural-language explanations

Generative AI is therefore particularly relevant to how users interact with ERP information.

Intelligent Automation

Intelligent automation combines automation with analytical or AI capabilities.

Instead of simply following a fixed rule, an intelligent workflow may use AI to interpret information before determining the next step.

For example:

Document → AI Interpretation → Validation → ERP Workflow → Human Approval

This can be useful when incoming information varies in format or requires some level of interpretation.

AI Agents

AI agents introduce another level of capability.

An agent can be designed to pursue a defined objective by interpreting information, using available tools or systems, deciding between permitted steps, and executing authorised actions.

Within ERP, this can potentially connect multiple activities into a single workflow.

For example:

Business Request → Retrieve ERP Data → Analyse → Determine Next Step → Execute Permitted Action → Report Result

The important distinction is that an agent is not simply generating text. It can potentially interact with business systems and participate in a workflow.

However, the level of autonomy should be appropriate to the risk of the process. Permissions, approval requirements, monitoring, exception handling, and auditability remain important.

Insights, Recommendations and Actions

The fourth layer is where AI outputs become useful to the business.

An AI capability may produce several types of output.

Insight

AI identifies something that may be relevant.

“Inventory for Product A has declined significantly over the past four weeks.”

Recommendation

AI goes one step further and suggests a potential response.

“Based on current demand and lead time, consider reviewing the replenishment quantity for Product A.”

Alert

AI identifies a condition that may require attention.

“Three customer orders may be affected by the projected inventory shortage.”

Workflow

The insight or recommendation can be connected to a defined business process.

“Create a replenishment review task and route it to the inventory planner.”

Approved Action

Where the workflow and permissions allow it, an AI-enabled process may execute a defined action.

“Update the replenishment proposal and submit it for approval.”

This creates a progression:

INSIGHT → RECOMMENDATION → WORKFLOW → ACTION

Not every AI use case should progress to the final stage.

For low-risk activities, automation may be appropriate. For financial, legal, customer, production, or other consequential decisions, human review may remain necessary.

The goal is therefore not maximum automation.

The goal is appropriate automation with appropriate controls.

Why Context Matters

One of the most important principles in AI-enabled ERP is that AI output is only as useful as the business context available to the system.

An AI model may produce an answer that is technically plausible but operationally inappropriate if it does not have access to the information required to understand the situation.

For example, an AI system may identify that a customer has an overdue invoice.

That alone does not determine what the business should do.

The customer may also:

• have an approved payment arrangement;
• be involved in an active dispute;
• have a credit note pending;
• have multiple entities within the group;
• be a strategically important account; or
• have another transaction that changes the financial context.

Without the appropriate business context, an AI-generated recommendation could be reasonable in isolation but inappropriate for the actual situation.

This is why AI readiness cannot be reduced to selecting an AI model or enabling an AI feature.

The AI needs access to the right information, in the right structure, with the right permissions and business context.

McKinsey’s 2026 research on AI data readiness similarly emphasises that scaling AI requires structured and unstructured data to be treated as a governed, reusable foundation that systems can interpret and trust.

From Data to Business Action

The complete flow can therefore be represented as:

BUSINESS DATA
↓
ERP CORE
Transactions • Master Data • Processes • Rules • Relationships
↓
AI CAPABILITIES
Predictive AI • Generative AI • Intelligent Automation • AI Agents
↓
OUTPUT
Insight • Recommendation • Alert
↓
BUSINESS PROCESS
Workflow • Approval • Exception Handling
↓
ACTION
Human Action • Automated Action • Controlled Agent Action
↓
MEASUREMENT
Outcome • Performance • Business Value

This final step—measurement—is important.

AI should not simply produce more outputs. The organisation needs to determine whether those outputs improve the underlying business process.

For example:

AI Forecast → Better Inventory Planning → Lower Stock Risk → Measurable Operational Outcome

Or:

AI-Assisted Reconciliation → Faster Exception Handling → Shorter Close Cycle → Measurable Finance Outcome

The architecture therefore connects technology to business value:

DATA → CONTEXT → INTELLIGENCE → DECISION → ACTION → MEASUREMENT

That is the fundamental role of AI inside an ERP environment.

AI provides the intelligence layer, but the ERP provides much of the business context and operational foundation required to turn that intelligence into controlled business activity.

---

What Types of AI Are Used in ERP?

AI in ERP is not a single technology.

Different AI capabilities solve different types of business problems. Some are designed to identify patterns and make predictions. Others generate or explain information, automate the interpretation of documents and data, or coordinate multi-step workflows.

For ERP, the important question is therefore not simply:

“Does the ERP have AI?”

A more useful question is:

“Which type of AI is appropriate for the business problem we are trying to solve?”

Current ERP developments increasingly combine several AI capabilities. Gartner’s September 2026 research on composite AI in ERP notes that many ERP use cases are fundamentally problems of prediction, classification, optimisation, and automation rather than content generation alone.

Four broad categories are particularly relevant to modern ERP environments:

1. Predictive AI
2. Generative AI
3. Intelligent Automation
4. AI Agents

These capabilities can overlap. A single ERP workflow may use more than one.

Predictive AI

Predictive AI uses historical and current data to identify patterns, estimate potential outcomes, detect anomalies, and support forecasting.

In ERP, predictive AI is particularly relevant when the business needs to answer:

“What is likely to happen next?”

Common applications include:

• Demand forecasting
• Cash-flow forecasting
• Inventory planning
• Risk detection
• Anomaly detection
• Customer payment prediction
• Production forecasting
• Predictive maintenance
• Sales forecasting

For example, an ERP system may analyse historical sales, current orders, inventory levels, lead times, and other relevant information to help identify potential future demand.

Predictive AI can also identify unusual activity.

A finance system might detect a transaction pattern that differs significantly from historical behaviour. An inventory system might identify an unusual movement or demand pattern. A manufacturing environment might detect signals associated with equipment performance.

The output is generally not a guaranteed prediction.

Instead, predictive AI provides an estimate, signal, or probability that can help people and processes respond earlier.

Gartner identifies AI-driven planning and forecasting, predictive analytics, and anomaly detection among emerging capabilities in cloud ERP finance applications.

The value of predictive AI therefore depends heavily on the quality and relevance of the data being analysed.

Generative AI

Generative AI is designed to create new content or responses based on information and instructions.

In ERP, its value is often less about making the underlying transaction and more about making enterprise information easier to understand, access, and use.

Common applications include:

• Management summaries
• Natural-language explanations
• Report generation
• Drafting business communications
• Conversational interaction
• Summarising financial or operational information
• Answering questions about enterprise data
• Generating contextual explanations of business activity

For example, instead of manually reviewing several financial reports, a user might ask:

“What changed in our operating expenses this month?”

A generative AI capability could interpret the request, retrieve relevant information, and present a natural-language explanation based on the available business context.

This changes the way users interact with ERP.

Instead of learning where a particular report or field is located, users can increasingly interact with enterprise information through natural language.

However, generative AI should not be confused with factual certainty.

A fluent response does not automatically mean that the underlying information is correct. ERP implementations therefore need appropriate grounding, permissions, validation, and controls when generative AI is used with business data.

Gartner identifies conversational analytics and natural-language interfaces as emerging capabilities in cloud ERP, alongside predictive and agentic capabilities.

Intelligent Automation

Intelligent automation combines traditional workflow automation with AI or machine-learning capabilities.

Traditional automation generally follows predefined rules:

IF X → DO Y

Intelligent automation can introduce an additional interpretation or decision layer:

INTERPRET → CLASSIFY → VALIDATE → ROUTE → ACT

This is particularly useful when business information is not always presented in exactly the same structure.

Common ERP applications include:

• Document classification
• Data extraction
• Invoice processing
• Order processing
• Transaction matching
• Exception identification
• Repetitive data processing
• Workflow routing
• Reconciliation support

Consider supplier invoices.

A traditional process may require employees to manually read an invoice, identify relevant information, enter the data, and send it through an approval workflow.

An AI-enabled process can potentially:

Read → Extract → Classify → Validate → Match → Route

The ERP remains responsible for the underlying transaction and workflow. AI helps interpret the incoming information and reduce manual processing.

This distinction matters because intelligent automation does not necessarily require an autonomous AI agent.

Many processes can be improved through a combination of AI-assisted interpretation and conventional business rules.

Gartner’s current ERP research identifies intelligent process automation as a major area where agentic AI, machine learning, and process orchestration are being combined to streamline activities such as reconciliation and collections.

AI Agents

AI agents represent a further evolution from AI that primarily analyses or generates information toward AI that can participate in multi-step workflows.

An AI agent can be designed to:

1. Understand a defined objective
2. Access relevant information
3. Analyse the available context
4. Determine the next permitted step
5. Use authorised tools or systems
6. Execute an action
7. Report the result or escalate an exception

The conceptual progression is:

ASSIST → RECOMMEND → EXECUTE WITH APPROVAL → EXECUTE WITHIN DEFINED CONTROLS

For example, an AI agent supporting accounts receivable could potentially review customer payment information, identify accounts requiring attention, prepare an appropriate next step, and initiate an authorised workflow.

A manufacturing-related agent could potentially analyse production information, identify a scheduling issue, evaluate permitted alternatives, and route a recommendation or approved action to the appropriate workflow.

The important distinction is that an AI agent can connect multiple steps rather than simply returning an answer.

McKinsey describes this broader shift as AI agents operating on top of traditional enterprise applications to automate decisions and orchestrate processes end to end.

However, agentic AI does not mean unrestricted autonomy.

The level of authority given to an agent should reflect the risk of the process.

A useful principle is:

The greater the business consequence of an action, the stronger the controls around that action should be.

For example, an AI system might be permitted to:

• summarise a report without approval;
• recommend a purchase adjustment for human review;
• prepare a financial reconciliation for approval;
• execute a low-risk workflow within predefined limits; but
• require explicit human approval before a high-impact financial transaction.

Gartner’s 2026 research specifically highlights governance as an important consideration as AI agents become embedded across ERP environments.

Choosing the Right Level of AI

The most advanced form of AI is not automatically the most appropriate.

A business should select the level of AI based on the nature of the problem, the potential value, the quality of available data, the complexity of the workflow, and the consequences of an incorrect action.

A practical model is:

Level	AI Role	Typical Application
Assist	Helps the user understand information	Summaries, explanations, search
Recommend	Suggests a potential decision or action	Forecasts, anomaly alerts, recommendations
Execute with Approval	Prepares or performs an action subject to human approval	Reconciliation, purchasing, workflow actions
Execute Within Defined Controls	Performs authorised actions automatically within predefined boundaries	Low-risk, repeatable business workflows

This creates a principle of appropriate autonomy.

The objective is not to maximise the amount of autonomy given to AI.

The objective is to determine where autonomy creates value without introducing unacceptable operational, financial, security, compliance, or governance risk.

How the Four AI Capabilities Work Together

In practice, these categories are not isolated.

A single ERP workflow can combine multiple capabilities.

For example:

Predictive AI
identifies a potential inventory shortage.

↓

Generative AI
explains the likely drivers of the shortage.

↓

Intelligent Automation
routes the issue to the appropriate planning workflow.

↓

AI Agent
within defined permissions, gathers the relevant information and initiates an approved replenishment process.

This can create a broader flow:

PREDICT → EXPLAIN → RECOMMEND → ORCHESTRATE → ACT

The appropriate combination depends on the business process.

An organisation does not necessarily need an AI agent to improve every ERP workflow. In many cases, predictive models, classification, anomaly detection, or generative assistance may provide the required value with lower complexity and lower operational risk.

This is consistent with Gartner’s current view of composite AI in ERP: organisations should evaluate multiple forms of AI according to business value, cost, risk, and governance rather than treating generative or agentic AI as the answer to every ERP problem.

What This Means for ERP Buyers

When evaluating AI capabilities in an ERP platform, businesses should look beyond the presence of an “AI” label.

Important questions include:

• What business problem does the capability address?
• What type of AI is actually being used?
• What enterprise data does it require?
• Is the capability embedded in the ERP workflow or operating separately?
• Can users understand why a recommendation was generated?
• What actions can the AI take?
• What permissions and approval controls are available?
• How are exceptions handled?
• Can activity be monitored and audited?
• How is performance measured over time?

The answers matter because AI capability should ultimately be evaluated in the context of the business process it is intended to improve.

Modern ERP platforms are increasingly combining these approaches. For example, Acumatica’s 2026 R1 release describes AI Assistant and AI Studio capabilities alongside AI-powered anomaly detection, reporting, AI-assisted workflows, and AI agents.

The broader direction is therefore not simply toward “more AI.”

It is toward using the right type of intelligence at the right point in the business process, with the right level of autonomy and control.

That distinction becomes increasingly important as organisations move from individual AI assistants toward AI-enabled workflows and, eventually, more agentic ERP environments.

---

AI in ERP: Key Business Use Cases

The value of AI in ERP becomes clearer when it is connected to specific business processes.

Rather than treating AI as a general-purpose technology, organisations can evaluate where it can improve a particular workflow, reduce manual effort, identify risks earlier, support better decisions, or help employees act on business information more effectively.

Common opportunities span finance, procurement, inventory, supply chain, manufacturing, projects, sales, customer management, and executive reporting.

The appropriate technology will vary by use case. Some problems are better suited to predictive AI or machine learning, while others may benefit from generative AI, intelligent automation, or AI agents. Gartner’s 2026 research on composite AI in ERP specifically notes that many ERP problems are fundamentally prediction, classification, optimisation, and automation problems rather than generative-AI problems alone.

A useful way to evaluate any ERP AI use case is:

DATA → CONTEXT → INTELLIGENCE → DECISION → ACTION → MEASUREMENT

The technology matters, but the business outcome matters more.

Key AI Use Cases Across ERP

	AI Use Case	What AI Can Do	Business Objective
Finance	Reconciliation	Identify discrepancies, matching issues, and exceptions	Support a faster financial close
Finance	Cash forecasting	Analyse historical and current patterns	Improve liquidity planning
Procurement	Invoice processing	Extract and classify information from documents	Reduce manual processing
Procurement	Supplier analysis	Identify purchasing patterns and exceptions	Improve supplier oversight
Sales	Order processing	Interpret incoming order information	Accelerate order entry
Inventory	Demand forecasting	Analyse demand patterns	Improve inventory planning
Supply Chain	Risk detection	Identify patterns that may indicate disruption	Enable earlier intervention
Manufacturing	Production planning	Analyse demand, capacity, and constraints	Support production scheduling
Manufacturing	Predictive maintenance	Detect patterns associated with equipment issues	Reduce unplanned disruption
Projects	Cost forecasting	Identify cost trends and potential overruns	Improve project control
CRM	Customer insights	Analyse customer and transaction information	Improve prioritisation
Management	AI reporting	Generate summaries and explanations	Accelerate management analysis
Operations	AI agents	Assist with multi-step workflows	Increase process efficiency

These use cases are not equally mature, and not every organisation should implement all of them. The appropriate starting point depends on the quality of available data, process maturity, technical feasibility, risk, and potential business value.

Finance

Finance is one of the areas where AI and automation can have a direct relationship with transaction processing, reconciliation, forecasting, controls, and management reporting.

Gartner identifies intelligent process automation, anomaly detection, adaptive analytics, and AI-driven planning and forecasting as important emerging themes in cloud ERP finance applications.

Reconciliation and Exception Detection

Financial reconciliation can involve comparing large volumes of transactions and identifying items that do not match expected records.

AI can assist by:

• identifying potential discrepancies;
• recognising matching patterns;
• prioritising exceptions;
• highlighting unusual transactions; and
• helping finance teams focus on items that require investigation.

The objective is not simply to automate reconciliation. It is to help finance teams spend less time searching for exceptions and more time resolving the exceptions that matter.

Cash-Flow Forecasting

AI can analyse historical cash movements alongside current financial information to support cash-flow forecasting.

Depending on the available data and model, this may help finance teams identify:

• expected cash requirements;
• changes in payment behaviour;
• emerging liquidity pressures;
• potential timing differences; and
• alternative scenarios.

Forecasting remains an estimate rather than a guarantee. The quality of the output depends on the quality, relevance, and completeness of the underlying financial information.

Financial Reporting and Analysis

Generative AI can also change how finance teams interact with ERP data.

Instead of manually reviewing multiple reports to identify a variance, a user may ask:

“What changed in operating expenses this month, and which areas require attention?”

AI can potentially summarise relevant information and explain observed patterns in natural language, subject to the data and context available to it.

This represents a shift from simply producing reports toward helping users interpret them.

Procurement

Procurement processes generate significant amounts of structured and unstructured information, making them another area where AI can support operational efficiency.

Invoice Processing

AI can help interpret supplier invoices by extracting relevant information and classifying the document before it enters an ERP workflow.

A typical flow can be:

Invoice → AI Extraction → Classification → Validation → ERP Record → Approval

This can reduce repetitive data entry while retaining the ERP as the system where the resulting transaction is managed.

For example, Acumatica describes AI-powered document recognition that can read vendor invoices, extract key details, and create bills for review and approval.

Supplier Analysis

AI can analyse purchasing and supplier information to help identify:

• unusual price movements;
• purchasing patterns;
• supplier performance trends;
• potential exceptions; and
• areas requiring procurement review.

The objective is not to replace procurement judgement but to surface information that may otherwise require significant manual analysis.

Exception Identification

AI can help identify transactions that deviate from expected patterns.

For example:

Purchase Order → Historical Pattern → AI Analysis → Unusual Price / Quantity → Procurement Review

This can help procurement teams focus attention where the data indicates a potential issue rather than reviewing every transaction with the same level of scrutiny.

Inventory and Supply Chain

Inventory and supply chain processes are particularly dependent on timing, demand patterns, supplier performance, and operational constraints.

AI can help organisations move from simply monitoring current conditions toward identifying potential future issues.

Demand Forecasting

AI can analyse historical demand, current orders, inventory information, and other relevant variables to support demand forecasting.

The objective is to provide planners with a more informed view of potential demand rather than relying solely on historical averages or manual spreadsheet analysis.

Inventory Planning

AI can help identify relationships between:

Demand → Inventory → Lead Time → Supply → Replenishment

This can support decisions around inventory levels, replenishment, and potential shortages.

The objective is not to maximise inventory reduction at all costs. It is to balance inventory availability, working capital, service levels, and operational requirements.

Supply Chain Risk Detection

AI can analyse supply and operational information to identify patterns that may indicate potential disruption.

For example:

Supplier Performance + Open Orders + Inventory Position → AI Analysis → Potential Supply Risk → Planner Review

This can give supply chain teams an opportunity to investigate a potential issue before it becomes an operational disruption.

Gartner’s 2026 research identifies AI use cases across supply chain performance optimisation and cost efficiency, while separate Gartner research assesses AI opportunities across customer fulfilment and order management.

Manufacturing

Manufacturing combines production planning, inventory, labour, machinery, quality, scheduling, and cost information. This creates multiple areas where AI can support decision-making and operational processes.

Production Planning and Scheduling

AI can analyse factors such as:

• demand;
• available capacity;
• material availability;
• production constraints;
• existing orders; and
• scheduling requirements.

The resulting analysis can support production planners in evaluating potential schedules and identifying constraints.

The goal is not necessarily to remove planners from the process. It is to give them better information for evaluating complex planning decisions.

Quality

AI can help identify patterns in production and quality data that may indicate potential quality issues.

Depending on the manufacturing environment, this could include analysing:

• production records;
• inspection information;
• defect patterns;
• machine data; and
• historical quality results.

The application will vary significantly by manufacturing process and available data.

Predictive Maintenance

AI can analyse equipment and operational data to identify patterns associated with potential equipment issues.

A simplified workflow is:

Equipment Data → Pattern Detection → Potential Failure Signal → Maintenance Review → Planned Action

This can shift maintenance from responding only after equipment fails toward identifying potential issues earlier.

Gartner’s 2026 research continues to identify manufacturing as an area with substantial AI use-case opportunities, including use cases assessed according to business value and technical feasibility.

Modern ERP platforms are also beginning to connect AI capabilities more directly to manufacturing workflows. Acumatica’s 2026 R1 release, for example, describes AI-powered insights alongside manufacturing capabilities such as production planning and real-time operational visibility.

Projects

Project-based businesses need visibility into costs, resources, progress, billing, and profitability.

AI can support project teams by analysing information across these areas.

Cost Forecasting

AI can identify cost patterns that may indicate potential budget pressure.

For example:

Actual Costs + Committed Costs + Project Progress → AI Analysis → Cost Trend → Project Review

This can help project managers investigate potential overruns earlier rather than waiting until the project is substantially off budget.

Resource Analysis

AI can analyse project schedules, workloads, skills, and resource utilisation to help identify potential allocation issues.

Depending on the data available, this may support:

• resource planning;
• workload analysis;
• schedule evaluation; and
• capacity planning.

Project Performance

AI can bring together project financial and operational information to help users understand:

• where projects are performing differently from plan;
• which costs are changing;
• where resources are being utilised differently; and
• which projects may require management attention.

The objective is to turn project information into earlier visibility rather than simply producing another project report.

Sales and CRM

Sales and CRM processes contain large volumes of customer, order, interaction, and opportunity data.

AI can help sales teams interpret this information and prioritise their attention.

Order Processing

AI can assist with interpreting incoming order information and translating it into structured ERP or order-management workflows.

For example:

Customer Order → AI Interpretation → Data Validation → ERP Order → Fulfilment Workflow

This can be particularly useful when order information arrives through documents, email, or other less-structured formats.

Gartner identifies customer fulfilment and order management as areas with AI opportunities spanning agentic AI, generative AI, and machine-learning-based applications.

Customer Insights

AI can analyse customer information and transaction history to identify patterns such as:

• changes in purchasing behaviour;
• product preferences;
• unusual activity;
• service issues; and
• changes in engagement.

The objective is to give sales and account teams more context when deciding where to focus.

Opportunity Prioritisation

AI can help sales teams analyse opportunities using available customer and pipeline information.

Rather than treating every opportunity identically, AI can help surface patterns that may warrant closer attention.

However, recommendations should support—not replace—the judgement of sales professionals, particularly when customer relationships involve information that may not be captured in structured ERP or CRM data.

Management Reporting and Executive Decision Support

Executives increasingly need to move quickly from raw business information to an understanding of what requires attention.

AI can help shorten this interpretation process.

AI-Generated Summaries

Instead of manually consolidating information from multiple reports, AI can potentially generate concise summaries based on available business data.

For example:

Revenue increased, but margin declined due to changes in product mix and higher fulfilment costs.

The value is not the sentence itself.

The value is the ability to move from raw information toward an explanation that can be investigated.

Explanation of Trends

AI can help users explore questions such as:

• Why did costs increase?
• Which business units changed most?
• What is driving an inventory movement?
• Which projects are showing cost pressure?
• What changed compared with the previous period?

This can make ERP information more accessible to non-technical users.

Executive Decision Support

The broader opportunity is to make enterprise information easier to interrogate.

A CFO, COO, or business leader may be able to ask a question in natural language and receive an answer based on relevant enterprise information, rather than navigating several dashboards and reports.

However, executive decision support still requires appropriate controls. AI-generated explanations should be grounded in authoritative business data, and consequential decisions should remain subject to human judgement and organisational governance.

Gartner identifies conversational analytics and context-sensitive dashboards among emerging AI capabilities in cloud ERP finance applications.

AI Agents and Multi-Step ERP Workflows

AI agents introduce a different type of opportunity because they can potentially coordinate multiple steps rather than simply provide information.

Consider a simplified financial workflow:

Identify Exception → Review Transactions → Analyse Supporting Information → Prepare Reconciliation → Route for Approval → Record Outcome

A traditional ERP workflow may require users to perform several of these steps manually.

An AI-enabled workflow could potentially assist across multiple stages, with the agent operating within defined permissions and escalating decisions that require human approval.

The same principle can apply to other processes:

Supply Chain

Detect Risk → Review Orders → Check Inventory → Identify Affected Transactions → Recommend Action

Customer Service

Review Customer History → Summarise Issue → Identify Relevant Information → Prepare Response → Route for Approval

Finance

Identify Exception → Gather Records → Analyse Difference → Prepare Resolution → Escalate or Execute Within Controls

The important distinction is that the agent connects multiple activities into a workflow.

It should not be interpreted as unrestricted autonomous decision-making.

Permissions, business rules, approval thresholds, monitoring, exception handling, and auditability remain essential—particularly when an AI system can affect financial records, customer commitments, supplier relationships, production activity, or other consequential business processes.

Acumatica’s current AI direction illustrates this progression, describing capabilities that span AI assistance, automation, and orchestration, including AI-powered workflows and agents that can work across business processes.

The Common Pattern Behind AI Use Cases

Although the applications differ across ERP functions, many successful AI use cases follow the same fundamental pattern:

DATA → CONTEXT → INTELLIGENCE → DECISION → ACTION → MEASUREMENT

Data
The system provides relevant business information.

↓

Context
ERP connects that information to business processes, rules, relationships, and current conditions.

↓

Intelligence
AI identifies patterns, predicts outcomes, generates explanations, or determines potential next steps.

↓

Decision
A user or controlled workflow determines what should happen.

↓

Action
The organisation performs the appropriate business action.

↓

Measurement
The result is measured against the original business objective.

This final step is critical.

An AI use case should not be considered successful simply because the technology produces an output.

The organisation should be able to ask:

• Did the process become faster?
• Did errors or exceptions decrease?
• Did forecast accuracy improve?
• Did employees spend less time on repetitive work?
• Did decision-makers receive information earlier?
• Did the business reduce avoidable cost or risk?
• Did the workflow produce a measurable operational outcome?

This is the difference between deploying AI and creating business value with AI.

The most useful AI use cases in ERP are therefore not necessarily the most sophisticated ones. They are the ones where a clearly defined business problem, suitable data, appropriate AI capability, controlled workflow, and measurable outcome come together.

AI should be applied where it can improve the business process—not simply where AI technology is available.

---

How AI Is Transforming Different ERP Functions

AI does not transform every ERP function in the same way.

The impact depends on the decisions each function makes, the data it manages, the processes it controls, and the level of automation that is appropriate for the business.

A useful way to understand the change is to compare the traditional role of each function with the capabilities AI can add.

ERP Function	Traditional Focus	With AI	Primary Shift
Finance	Transaction processing and reporting	Analysis, reconciliation, forecasting, exception management	From recording financial activity to interpreting and anticipating it
Procurement	Manual processing and approvals	Extraction, classification, supplier intelligence	From processing transactions to identifying purchasing insights
Supply Chain	Historical analysis and reactive planning	Predictive planning and risk detection	From reacting to disruption to anticipating it
Manufacturing	Scheduled production planning	Constraint-aware planning and predictive insights	From fixed planning to more adaptive decision support
Sales & CRM	Customer and opportunity records	Prioritisation and customer intelligence	From recording interactions to identifying where attention is needed
Project Operations	Retrospective project reporting	Forward-looking cost and performance analysis	From reporting project status to anticipating project risk
Executive Management	Dashboards and periodic reports	Conversational analysis and decision support	From consuming reports to interacting with business intelligence

The objective is not to replace the functional expertise of finance, procurement, operations, sales, or management teams.

Instead, AI can change where people spend their attention.

Routine information processing can increasingly be assisted by technology, while employees can focus more on exceptions, decisions, relationships, and activities that require business judgement.

Finance

Traditional ERP finance functions are built around accurate transaction processing, accounting controls, reconciliation, reporting, and financial management.

AI can extend these capabilities by helping finance teams interpret large volumes of financial information and identify issues that require attention.

The shift can be represented as:

TRANSACTION PROCESSING → ANALYSIS → EXCEPTION MANAGEMENT → FORECASTING

AI can support finance through:

• automated or assisted reconciliation;
• anomaly detection;
• cash-flow forecasting;
• variance analysis;
• financial reporting summaries;
• natural-language analysis; and
• exception prioritisation.

For example, instead of reviewing every transaction with the same level of attention, AI can help identify transactions or patterns that appear unusual and direct finance professionals toward the exceptions that warrant investigation.

This does not eliminate financial controls.

The ERP remains responsible for maintaining financial records, applying accounting rules, managing permissions, and supporting auditability. AI adds an analytical and assistance layer around those processes.

Gartner’s 2026 research identifies embedded AI assistants, machine learning, generative AI, and AI agents as technologies expected to reshape cloud ERP finance applications. Gartner also highlights applications including intelligent process automation, adaptive analytics, AI-driven planning and forecasting, and anomaly detection.

The result is a potential shift from:

“What was recorded?”

toward:

“What changed, why did it change, and what requires attention?”

Procurement

Procurement has traditionally relied on structured workflows for requisitions, purchase orders, supplier management, approvals, receiving, and invoice processing.

AI can introduce greater intelligence into these processes.

The shift can be represented as:

MANUAL PROCESSING → AI-ASSISTED INTERPRETATION → SUPPLIER INTELLIGENCE

AI can assist with:

• extracting information from supplier documents;
• classifying purchasing information;
• matching invoices and transactions;
• identifying unusual prices or quantities;
• analysing supplier performance;
• identifying purchasing patterns; and
• prioritising procurement exceptions.

Consider a supplier invoice.

A conventional workflow may require an employee to inspect the document, extract the relevant information, enter it into the ERP, and initiate the appropriate workflow.

An AI-enabled workflow can potentially:

READ → EXTRACT → CLASSIFY → VALIDATE → MATCH → ROUTE

The ERP still maintains the resulting transaction and approval process.

The difference is that AI can reduce the amount of manual interpretation required before the transaction enters that process.

The same principle can extend to supplier intelligence. Instead of simply storing supplier transactions, AI can analyse purchasing history, pricing, delivery patterns, and other available information to identify trends or exceptions.

The objective is therefore not simply faster procurement.

It is better visibility into procurement decisions and exceptions.

Supply Chain

Supply chain management is particularly sensitive to uncertainty.

Demand changes. Suppliers experience delays. Inventory positions move. Lead times vary. Logistics conditions change.

Traditional ERP provides visibility into the current and historical state of these activities.

AI can extend that capability toward prediction and early risk detection.

The shift can be represented as:

HISTORICAL ANALYSIS → PREDICTIVE PLANNING → RISK DETECTION

AI can support supply chain teams by analysing:

• demand patterns;
• inventory positions;
• supplier performance;
• open orders;
• lead times;
• fulfilment activity; and
• other relevant operational information.

For example:

Demand + Inventory + Lead Time + Supplier Data → AI Analysis → Potential Risk → Planner Review

This can help planners identify potential problems earlier.

The objective is not to predict every disruption perfectly.

It is to improve the organisation’s ability to identify signals, evaluate scenarios, and respond before a potential issue becomes a larger operational problem.

Acumatica’s 2026 R1 release, for example, describes AI-powered capabilities alongside supply-chain improvements designed to help manufacturers, distributors, and retailers respond to demand changes, improve fulfilment, and strengthen profitability control.

This illustrates an important principle of AI-enabled supply chain management:

The value of prediction comes from giving the business more time and information to respond.

Manufacturing

Manufacturing ERP connects production planning with materials, inventory, capacity, orders, costs, and operational execution.

Traditional planning can rely heavily on predefined schedules, available capacity, and planner judgement.

AI can add another layer of analysis.

The shift can be represented as:

SCHEDULED PLANNING → CONSTRAINT ANALYSIS → PREDICTIVE INSIGHT

AI can help analyse relationships between:

• customer demand;
• production orders;
• material availability;
• machine or work-centre capacity;
• production constraints;
• historical production performance; and
• inventory requirements.

This can support planners in evaluating potential production scenarios.

For example:

Demand → Material Availability → Capacity → Constraints → AI Analysis → Planning Recommendation

The objective is not necessarily to remove the production planner from the process.

Manufacturing decisions can involve commercial priorities, quality considerations, customer commitments, workforce constraints, and operational knowledge that may not be completely represented in structured ERP data.

AI can therefore act as a decision-support layer that helps planners evaluate more information and identify potential constraints earlier.

AI can also support predictive maintenance and quality analysis where appropriate operational data is available.

Acumatica’s 2026 product direction includes AI capabilities across manufacturing and distribution, while its 2026 R1 release describes AI-powered workflows and tools designed to improve operational responsiveness.

The broader transformation is from fixed visibility into production activity toward more forward-looking operational intelligence.

Sales and CRM

Sales and CRM systems traditionally focus on maintaining customer records, managing leads and opportunities, tracking interactions, and supporting the sales pipeline.

AI can make that information more actionable.

The shift can be represented as:

CUSTOMER RECORDS → CUSTOMER INTELLIGENCE → PRIORITISATION

AI can analyse available customer and transaction information to help identify:

• changes in purchasing behaviour;
• customer trends;
• opportunity patterns;
• potential risks;
• cross-sell or upsell signals;
• inactive or changing accounts; and
• opportunities requiring attention.

For sales teams, this can reduce the need to manually review every customer or opportunity with the same level of attention.

Instead, AI can help surface patterns that warrant investigation.

For example:

Customer History + Orders + Interactions → AI Analysis → Customer Signal → Sales Action

The salesperson still provides the relationship knowledge and judgement.

AI provides another layer of analytical context.

This distinction is important because customer relationships are not completely represented by structured ERP or CRM data. A model may identify a statistical pattern, but a sales professional may know about a strategic relationship, contract negotiation, market event, or customer concern that is not captured in the system.

AI should therefore support customer intelligence rather than be treated as a substitute for customer judgement.

Project-Based Operations

Project-driven organisations need to manage budgets, resources, costs, schedules, billing, and profitability throughout the project lifecycle.

Traditional project reporting is often retrospective.

It tells management:

“Here is where the project stands.”

AI can help move the conversation toward:

“Where is the project heading?”

The shift can be represented as:

RETROSPECTIVE REPORTING → FORWARD-LOOKING ANALYSIS

AI can analyse:

• actual project costs;
• committed costs;
• project budgets;
• resource utilisation;
• project progress;
• billing information;
• historical project patterns; and
• other available operational data.

For example:

Budget + Actual Cost + Committed Cost + Progress → AI Analysis → Cost Trend → Project Review

This can help project managers identify potential cost pressure or resource issues earlier.

The objective is not to automatically determine whether a project will succeed or fail.

Rather, AI can help surface signals that allow project managers to investigate potential problems while there is still time to respond.

This can be particularly relevant to construction, professional services, engineering, and other project-based businesses where project profitability depends on the interaction between time, resources, scope, costs, and billing.

Executive Management

Executive teams often have access to extensive ERP dashboards and reports.

The challenge is not necessarily a lack of information.

It is the amount of information that must be interpreted before a decision can be made.

AI can change the interaction model.

The shift can be represented as:

DASHBOARDS & REPORTS → CONVERSATIONAL ANALYSIS → DECISION SUPPORT

Instead of navigating multiple reports, an executive might ask:

“What changed in our gross margin this quarter?”

Then:

“Which business units contributed most to the change?”

Then:

“What operational factors appear to be driving it?”

Then:

“Which areas should management investigate?”

The potential value is not simply that AI can produce a natural-language response.

It is that users can interact with enterprise information in a more direct way.

This can make ERP information more accessible to decision-makers who may not know the structure of every underlying report or database.

However, executive decision support requires a particularly strong emphasis on data integrity and transparency.

A concise AI-generated explanation is only useful if the underlying information is authoritative, current, and relevant to the question being asked.

Gartner’s 2026 research identifies conversational analytics and context-sensitive dashboards among emerging AI capabilities in cloud ERP finance applications.

The Functional Transformation Is Connected

These changes should not be viewed as isolated improvements to individual ERP modules.

One of the fundamental strengths of ERP is the connection between business functions.

Consider a manufacturing company.

A change in customer demand can affect:

Sales → Demand Planning → Inventory → Procurement → Production → Finance

AI can potentially analyse these relationships across the enterprise rather than treating each function as an isolated system.

For example:

Customer Demand Signal
↓
Demand Forecast
↓
Inventory Requirement
↓
Procurement / Production Planning
↓
Operational Execution
↓
Financial Impact

This is where integrated ERP becomes particularly important.

The more AI moves toward cross-functional decision support and workflow orchestration, the more important the connections between business processes become.

McKinsey’s 2026 research describes AI agents as increasingly capable of operating across enterprise processes while emphasising the continued importance of ERP data, application logic, and business rules as the foundation for reliable and auditable operations.

From Functional Automation to Enterprise Intelligence

The long-term opportunity is therefore larger than automating individual tasks.

The progression can be viewed as:

FUNCTIONAL DATA
↓
FUNCTIONAL INSIGHT
↓
CROSS-FUNCTIONAL CONTEXT
↓
COORDINATED DECISION SUPPORT
↓
CONTROLLED BUSINESS ACTION

Finance can understand financial exceptions.

Procurement can identify supplier patterns.

Supply chain can detect potential risks.

Manufacturing can evaluate production constraints.

Sales can identify customer signals.

Project teams can identify cost trends.

Executives can interrogate business performance.

But the greatest value can emerge when these capabilities operate against a connected enterprise context.

That is the strategic significance of AI-enabled ERP.

The objective is not to create a collection of disconnected AI features across individual modules.

It is to create an environment where data, processes, intelligence, and decision-making are connected across the organisation.

This is also why AI readiness is ultimately an enterprise architecture and operating-model question—not simply a software feature question.

The next challenge is understanding how this transformation differs across industries, because the value, data requirements, processes, and risks of AI in ERP are not identical for a manufacturer, construction company, distributor, or professional services organisation.

---

AI in ERP Across Industries

The impact of AI in ERP depends heavily on the business model.

A manufacturer, construction company, distributor, and professional services firm may all use the same underlying AI technologies, but the data they generate, the decisions they make, and the workflows they need to improve are fundamentally different.

This means there is no single “AI for ERP” use case that applies equally to every industry.

A more useful way to evaluate the opportunity is:

BUSINESS MODEL → DATA → ERP PROCESS → AI USE CASE → BUSINESS OUTCOME

The business model determines what the organisation needs to optimise.
The data provides the context.
The ERP connects that information to operational processes.
AI identifies patterns, supports decisions, or automates appropriate activities.
The outcome should be measurable against the original business objective.

This industry-specific approach is increasingly reflected in modern ERP development. Acumatica, for example, states that its 2026 R1 AI capabilities are being developed within its industry-specific applications for manufacturing, distribution, construction, and professional services rather than treating AI as a universal layer detached from industry workflows.

Manufacturing

Manufacturing businesses operate through interconnected processes involving demand, materials, production capacity, inventory, quality, labour, equipment, and profitability.

That makes manufacturing particularly dependent on the quality and timeliness of operational data.

AI can support manufacturers across several areas:

• Demand forecasting
• Production planning
• Inventory planning
• Material requirements
• Production scheduling
• Quality analysis
• Predictive maintenance
• Cost analysis
• Supply chain risk detection
• Production performance analysis

The potential transformation can be represented as:

DEMAND → PLANNING → PRODUCTION → INVENTORY → QUALITY → PROFITABILITY

AI can analyse relationships across these processes rather than treating each activity independently.

For example, a manufacturer may combine customer demand, existing orders, inventory availability, production capacity, material requirements, and historical production data to identify potential constraints.

The resulting workflow might look like:

Demand Data + Inventory + Capacity + Materials → AI Analysis → Production Insight → Planner Decision → Production Action

The objective is not necessarily to automate production planning completely.

Manufacturing decisions can involve customer commitments, quality requirements, machine availability, workforce considerations, and operational knowledge that may not be fully represented in structured data.

AI can instead provide planners with earlier signals and a broader view of the variables affecting production.

Acumatica’s 2026 R1 release provides a current example of this direction, combining AI capabilities with manufacturing functionality and improved production visibility. Its release includes AI Studio and AI Assistant capabilities alongside manufacturing improvements such as real-time shop-floor data capture, production-cost tracking, and improved visibility into in-transit inventory for planning.

For manufacturers, the strategic opportunity is therefore not simply “AI for the factory.”

It is the ability to connect demand, materials, production, inventory, and financial information so that decisions can be made with greater context.

Construction

Construction is fundamentally project-driven.

Financial performance depends on the relationship between contracts, estimates, procurement, labour, subcontractors, materials, project progress, change orders, billing, and actual costs.

This creates a different AI opportunity from manufacturing.

The central question is often:

“Are we still on track to deliver the project profitably?”

AI can support construction businesses through:

• Project cost forecasting
• Cost anomaly detection
• Budget analysis
• Change-order analysis
• Procurement insights
• Resource planning
• Project performance monitoring
• Forecasting
• Document and field information management
• Risk identification

The potential workflow can be represented as:

PROJECT DATA → COST & PROGRESS ANALYSIS → AI INSIGHT → PROJECT DECISION → ACTION

For example, an AI-enabled ERP environment could analyse actual costs, committed costs, project progress, budget information, and other available project data to identify a developing cost trend.

The purpose is not to declare that a project will overrun.

It is to give project and finance teams an earlier signal that warrants investigation.

This is particularly important because construction risks are often interconnected.

A procurement delay can affect material availability.

Material availability can affect project progress.

Project delays can affect labour and subcontractor costs.

Those changes can ultimately affect project profitability and cash flow.

An integrated ERP environment can provide the relationships between these data points, while AI can help identify patterns within them.

Acumatica’s 2026 R1 Construction Edition updates illustrate this direction, with AI-powered forecasting and anomaly detection positioned around earlier risk identification and margin protection, alongside improvements to change-order workflows, project financials, cost projections, and document management.

For construction companies, the opportunity is therefore closely connected to project visibility and early intervention.

Distribution

Distribution businesses operate at the intersection of inventory, purchasing, sales, warehousing, fulfilment, logistics, and customer demand.

Margins can depend on relatively small changes in inventory levels, purchasing costs, fulfilment efficiency, supplier performance, and order accuracy.

AI can support distributors in areas such as:

• Demand forecasting
• Inventory optimisation
• Replenishment planning
• Order management
• Warehouse operations
• Supplier analysis
• Pricing and margin analysis
• Fulfilment
• Logistics visibility
• Customer purchasing patterns

The core relationship can be represented as:

DEMAND → INVENTORY → PROCUREMENT → WAREHOUSE → FULFILMENT → CUSTOMER

AI can help identify patterns across that chain.

For example:

Customer Demand + Inventory Position + Supplier Lead Time → AI Analysis → Replenishment Insight → Purchasing Decision

Or:

Customer Order + Inventory + Warehouse Data → AI Analysis → Fulfilment Recommendation → Operational Action

This becomes particularly valuable when distributors operate across multiple warehouses, locations, sales channels, suppliers, or product categories.

Acumatica’s 2026 R1 release specifically positions AI-enabled capabilities around supply-chain execution for manufacturers, distributors, and retailers. The release includes AI Studio, AI Assistant access, warehouse improvements, inventory controls, and procurement capabilities designed to help product-driven businesses respond to demand changes and improve fulfilment and profitability control.

For distributors, AI therefore has the potential to move ERP from primarily monitoring inventory and orders toward anticipating demand, identifying exceptions, and supporting faster operational decisions.

Professional Services

Professional services businesses have a different operating model again.

Their primary assets are often people, expertise, client relationships, and project capacity rather than physical inventory.

ERP and professional services systems therefore need to connect:

PEOPLE → PROJECTS → TIME → COST → BILLING → REVENUE → PROFITABILITY

AI can support professional services organisations through:

• Project profitability analysis
• Resource utilisation
• Capacity planning
• Revenue forecasting
• Cost forecasting
• Billing analysis
• Unbilled work monitoring
• Project performance insights
• Customer and project analysis
• Management reporting

One important area is resource utilisation.

A professional services firm may need to understand:

• how much capacity is available;
• which employees are assigned to projects;
• how much time is billable;
• where non-billable time is increasing;
• which projects require additional resources; and
• how staffing decisions could affect profitability.

AI can help identify patterns across these data points.

For example:

Project Demand + Employee Capacity + Utilisation + Project Economics → AI Analysis → Resource Insight → Staffing Decision

Acumatica’s 2026 R1 Professional Services updates include employee-utilisation insights, project revenue analysis, and unbilled project revenue reporting, alongside AI-enabled workflows and improved financial visibility.

The opportunity for professional services firms is therefore less about inventory optimisation and more about understanding the relationship between people, projects, utilisation, revenue, and profitability.

Why Industry Context Matters

The examples above demonstrate why AI cannot be evaluated separately from the business model.

Consider a simple comparison:

Industry	Critical Business Context	Potential AI Focus
Manufacturing	Demand, materials, capacity, production, quality	Forecasting, planning, production insights, maintenance
Construction	Projects, costs, progress, procurement, resources	Cost forecasting, anomaly detection, project risk
Distribution	Demand, inventory, suppliers, orders, fulfilment	Forecasting, replenishment, fulfilment, supply-chain insights
Professional Services	People, projects, utilisation, billing, profitability	Resource planning, project forecasting, profitability insights

The same AI technology can therefore produce very different value depending on where it is applied.

Predictive AI in manufacturing may support production and demand planning.

The same broad class of technology in construction may support project cost forecasting.

In distribution, it may support inventory and demand planning.

In professional services, it may support resource utilisation and project profitability.

The technology is similar.

The business context is not.

From Industry Data to Industry-Specific Intelligence

This leads to a broader principle for AI-enabled ERP:

AI becomes more valuable when it understands the context in which the business operates.

An AI system that understands only isolated transactions may provide limited value.

An AI-enabled ERP environment can provide a richer context by connecting:

Industry Data + Business Rules + ERP Processes + Historical Patterns + Current Conditions

That context can then support:

AI Analysis → Business Insight → Human or Automated Decision → Measurable Outcome

This is one reason modern ERP platforms continue to develop industry-specific editions alongside AI capabilities. Acumatica currently offers industry editions for manufacturing, construction, distribution, professional services, and other sectors, with its 2026 R1 strategy explicitly combining industry depth with AI-enabled functionality.

The strategic implication is important for ERP buyers.

The question should not simply be:

“Does this ERP have AI?”

It should be:

“Can this ERP apply AI within the processes, data, and business model that matter most to our industry?”

That distinction helps move the discussion away from AI feature lists and toward practical business value.

For an organisation evaluating AI-enabled ERP, the next step is to understand how AI agents extend these capabilities beyond individual insights and recommendations into increasingly connected, multi-step business workflows.

---

How AI Agents Are Changing ERP Automation

Traditional ERP automation is built around predefined rules and workflows.

A condition occurs, the system evaluates that condition, and a predefined action follows.

AI agents introduce a different model.

Instead of executing only a predetermined sequence, an AI agent can be designed to interpret a goal, gather relevant information, determine the next permitted step, use connected systems or tools, and complete or escalate parts of a workflow.

This creates a progression:

TRADITIONAL AUTOMATION → AI-ASSISTED WORKFLOW → AI AGENT

The important distinction is not that AI agents replace automation.

Rather, they can make automation more context-aware, adaptive, and capable of coordinating multiple steps within a defined business process.

McKinsey describes this broader development as AI agents operating on top of enterprise applications to automate decisions and orchestrate processes end to end, while the underlying ERP remains important for data, business logic, reliability, auditability, and compliance.

What Is an AI Agent?

An AI agent is a software system designed to pursue a defined objective by interpreting information, reasoning about available options, using authorised tools or systems, and taking permitted actions.

Within an ERP environment, an agent may be able to interact with:

• ERP records;
• business data;
• workflows;
• business rules;
• documents;
• external systems;
• analytical tools; and
• other authorised applications.

The important difference from a conventional AI assistant is the relationship between understanding and action.

An assistant may answer:

“Which customer invoices are overdue?”

An AI agent may be designed to:

Identify overdue invoices → review relevant customer information → classify exceptions → prepare the appropriate next step → route the issue for approval or execute an authorised action.

The agent is therefore participating in a workflow rather than simply providing information.

However, the actual capabilities depend on how the agent is designed, what systems it can access, what permissions it has, and what controls the organisation establishes.

Acumatica’s current AI platform describes this direction through AI-powered workflows and agents that can coordinate processes across teams and departments, while its 2026 R1 release introduced AI agents and AI-assisted workflows for areas including customer service, sales, and operations.

Automation vs. AI-Assisted Workflow vs. AI Agent

The distinction can be simplified as follows:

Capability	Traditional Automation	AI-Assisted Workflow	AI Agent
Logic	Predefined rules	AI + rules	Context + reasoning within defined boundaries
Input	Usually structured	Structured and/or unstructured	Contextual information from authorised sources
Decision	Predetermined	AI provides analysis or recommendation	Determines permitted next steps
Action	Automatically executes predefined action	Usually assists or prepares action	Can execute defined actions
Workflow	Fixed sequence	Human-guided sequence	Can coordinate multiple permitted steps
Oversight	Rule and system monitoring	Human review	Permissions, monitoring, evaluation, and human oversight
Best suited to	Predictable, repetitive processes	Processes requiring interpretation	Multi-step processes with defined objectives and controls

Consider a simple accounts-payable process.

Traditional automation

If an invoice matches the purchase order and receipt, route it according to the approval rule.

AI-assisted workflow

Analyse the invoice and matching information → identify potential discrepancy → explain the exception → recommend the next step.

AI agent

Review the invoice → retrieve related purchase-order and receipt information → analyse the match → classify the exception → perform permitted workflow actions → escalate unresolved or high-risk cases.

The third approach involves more than adding AI to an existing automation rule.

It introduces the possibility of workflow orchestration.

From Fixed Workflows to Adaptive Workflows

Traditional automation works particularly well when the process is predictable.

For example:

IF invoice amount > threshold → require approval

The rule is clear.

But many business processes contain variations that are difficult to capture through a large collection of predefined rules.

An AI agent can potentially evaluate the context around the transaction before determining the next permitted step.

For example:

Incoming Request
↓
Understand Context
↓
Retrieve Relevant ERP Information
↓
Analyse
↓
Determine Permitted Next Step
↓
Execute / Recommend / Escalate
↓
Record Outcome

This does not mean that the agent has unlimited freedom.

The organisation can define which tools the agent can use, which records it can access, which actions it can perform, and which situations require human intervention.

The result is better understood as controlled adaptability rather than unrestricted autonomy.

Example: AI Agent in a Sales Order Workflow

Consider a business receiving a customer order through email or another unstructured channel.

A traditional process might require an employee to:

1. Read the incoming order.
2. Identify the customer.
3. Identify products and quantities.
4. Check pricing.
5. Check inventory.
6. Enter the order into the ERP.
7. Validate the information.
8. Trigger fulfilment.
9. Escalate missing or conflicting information.

An AI-enabled agent could potentially coordinate several of these steps.

The workflow could look like:

Incoming Customer Order
↓
Interpret Order Information
↓
Identify Customer and Products
↓
Validate Information Against ERP Data
↓
Check Pricing and Availability
↓
Create or Update ERP Record
↓
Trigger Downstream Workflow
↓
Escalate Exceptions to Human

The agent does not need to make every decision independently.

For example, the organisation could define rules such as:

• Orders within approved pricing parameters may proceed.
• Missing customer information requires review.
• Pricing outside an authorised range requires sales approval.
• Inventory conflicts require planner intervention.
• High-value orders require explicit approval.
• Unrecognised products cannot be automatically created.

This allows the agent to operate within the organisation’s existing control framework.

The goal is not:

“AI takes over order processing.”

The more accurate description is:

“AI coordinates authorised steps within the order-processing workflow and escalates exceptions that require human judgement.”

AI Agents and ERP Business Context

An AI agent becomes more useful when it can access the business context required to understand the workflow.

For example, an order-processing agent may need access to:

• customer records;
• product information;
• pricing rules;
• inventory availability;
• credit status;
• shipping information;
• order history; and
• approval policies.

This is where ERP architecture becomes important.

An external AI system that only receives a text description of an order may have limited context.

An agent connected to the ERP’s authorised data, business logic, permissions, and workflows can potentially operate much closer to the actual transaction layer.

Microsoft describes this principle in its 2026 Dynamics 365 agent-ready architecture: agents need governed access to business context and must operate through the same data models, rules, permissions, security guardrails, and audit trails used by the business application.

The broader principle applies beyond any individual ERP platform:

An AI agent needs business context and controlled access to act reliably within an enterprise workflow.

Governance Requirements for AI Agents

As an AI system moves from generating information toward taking action, governance becomes increasingly important.

The organisation needs to know:

What can the agent see?
What can it decide?
What can it change?
When must it ask for approval?
How is its activity monitored?
What happens when it encounters an exception?

At minimum, an agent-enabled ERP workflow should consider the following controls.

Permissions

Agents should have only the access required for the tasks they are authorised to perform.

Permissions should reflect the sensitivity and consequences of the workflow.

An agent that can read inventory information does not necessarily need permission to modify financial records.

Microsoft’s current guidance for ERP-connected agents emphasises reviewing roles, duties, privileges, permitted agent clients, and data-handling requirements.

Human Oversight

Human involvement should remain part of workflows where judgement, accountability, or business risk requires it.

Human approval may be appropriate for:

• high-value transactions;
• unusual financial activity;
• contractual commitments;
• sensitive customer communications;
• significant production changes;
• exceptions outside defined rules; and
• decisions with material financial or operational consequences.

Human oversight does not mean that every individual AI action must be manually approved.

The appropriate level of intervention should depend on the risk of the process.

Auditability

The organisation should be able to determine:

• what the agent did;
• what information it accessed;
• what decision or recommendation it produced;
• what action it took;
• when it took the action;
• which permissions were used; and
• whether a human approved or modified the outcome.

This becomes particularly important when AI participates directly in transactional workflows.

Exception Handling

AI agents will encounter situations they cannot confidently resolve.

A robust workflow therefore needs a defined path for:

Agent → Exception → Human Review → Resolution → Workflow Continuation

An agent should not be forced to complete an action simply because the workflow expects an answer.

Escalation is part of good automation design.

Monitoring

Agent performance should be monitored over time.

Relevant measures may include:

• successful workflow completion;
• exception rate;
• error rate;
• escalation frequency;
• human override rate;
• processing time;
• action accuracy; and
• business outcome.

Microsoft’s current ERP agent implementation provides an example of this approach, with activity monitoring, telemetry, explainability instrumentation, scenario-based evaluations, and precision/recall tracking for certain agent recommendations.

Business Rules

AI does not replace business rules.

Rules remain important for defining:

• approval thresholds;
• authorised actions;
• segregation of duties;
• financial controls;
• compliance requirements;
• escalation conditions; and
• prohibited actions.

AI can operate within these rules, interpret context around them, or help determine which permitted path is appropriate.

AI Agents Are Not Unrestricted Autonomy

One of the most important distinctions in discussions about agentic ERP is the difference between autonomy and authority.

An AI agent may be capable of reasoning through a workflow without being authorised to perform every action it can technically identify.

For example, an agent might determine that a supplier payment appears overdue.

It may be able to:

• retrieve the relevant supplier record;
• review invoice status;
• identify the outstanding amount;
• summarise the situation; and
• recommend a follow-up.

But that does not automatically mean it should:

• change payment terms;
• release a payment;
• contact the supplier;
• modify financial records; or
• override an approval control.

Those actions require appropriate authority.

A useful model is:

AI CAPABILITY ≠ BUSINESS AUTHORITY

The organisation determines what the agent is permitted to do.

This distinction becomes increasingly important as agents move closer to transactional systems.

Microsoft’s 2026 documentation similarly emphasises that ERP-connected agents remain subject to organisational permissions, compliance requirements, data governance, and administrative controls.

The Right Level of Autonomy

Not every ERP process requires the same level of AI autonomy.

A practical model is:

Level	AI Role	Example
Assist	Provides information or explanation	Summarise an overdue receivables report
Recommend	Suggests an action	Recommend which invoices require investigation
Execute with Approval	Performs preparation and waits for authorisation	Prepare reconciliation entries for finance approval
Execute Within Controls	Performs predefined low-risk actions	Route or classify documents within defined rules

The appropriate level depends on:

Business Value × Process Complexity × Risk × Data Quality × Control Requirements

A low-risk, repetitive workflow may be suitable for a higher level of automation.

A financially material or compliance-sensitive workflow may require significantly more human oversight.

The objective is therefore not maximum autonomy.

It is appropriate autonomy for the specific business process.

AI Agents and the Future of ERP Automation

The emergence of AI agents changes the definition of ERP automation.

Traditional automation asks:

“What rule should trigger this action?”

AI-assisted workflows ask:

“What does the available information suggest we should do?”

AI agents introduce another question:

“What objective are we trying to achieve, and which authorised steps can be taken to achieve it?”

This progression can be represented as:

RULE-BASED AUTOMATION
↓
AI-ASSISTED DECISION SUPPORT
↓
AI-ORCHESTRATED WORKFLOW
↓
CONTROLLED AGENTIC EXECUTION

The transition will not happen uniformly across every ERP process.

Some workflows will remain rule-based because rules are simpler, more predictable, and easier to govern.

Others may benefit from AI assistance.

More complex, multi-step workflows may eventually benefit from agents that can coordinate activities across applications and departments.

Acumatica’s current AI direction illustrates this progression. Its AI platform describes capabilities spanning assistance, intelligent automation, and orchestration, including agents that can classify files, summarise support conversations, and coordinate processes across teams and departments.

The strategic opportunity is therefore not to make every ERP process autonomous.

It is to determine which processes benefit from intelligence, which actions can safely be automated, and where human judgement should remain part of the workflow.

That distinction will become increasingly important as businesses move from experimenting with AI assistants toward deploying AI agents in real operational environments.

---

How AI Is Changing ERP Implementation

AI is not only changing what ERP systems can do after implementation. It is also beginning to change how ERP systems are designed, configured, tested, documented, and deployed.

Traditionally, ERP implementation has involved a sequence of highly manual activities: gathering requirements, analysing business processes, designing the target operating model, configuring the ERP, migrating data, testing the solution, preparing documentation, and training users.

AI can increasingly assist across these stages by analysing large volumes of business information, identifying patterns, generating recommendations, automating repetitive implementation tasks, and supporting consultants and business users throughout the transformation.

The implementation lifecycle can therefore be viewed as:

ANALYSE → DESIGN → CONFIGURE → TEST → DOCUMENT → TRAIN

AI does not eliminate the need for implementation expertise. Instead, it can shift human effort away from repetitive analysis and preparation toward business decisions, validation, governance, architecture, and change management.

McKinsey’s 2026 analysis suggests that AI agents have the potential to reduce the effort required to implement ERP systems by at least 50% and reduce programme duration by half. McKinsey presents this as a potential impact of emerging agentic approaches—not as a guaranteed outcome for every ERP implementation.

AI Can Change the ERP Implementation Lifecycle

The traditional ERP implementation model relies heavily on consultants and business users manually collecting information, interpreting requirements, preparing configurations, creating test scenarios, and producing documentation.

AI can augment these activities by processing information at greater scale and speed.

Implementation Stage	Traditional Approach	Potential AI Assistance
Analyse	Review requirements, documents and existing processes manually	Classify requirements, identify patterns, detect redundancy and organise information
Design	Analyse workflows and develop target processes	Analyse process variations, identify bottlenecks and support standardisation
Configure	Consultants map requirements to ERP capabilities	Assist with requirement-to-capability mapping and configuration recommendations
Test	Teams manually create and execute test scenarios	Generate test cases, analyse scenarios and identify potential defects
Document	Documentation prepared manually throughout the project	Draft process documentation, configuration summaries and user guidance
Train	Training materials developed for different user groups	Prepare role-based training content and in-system guidance

The important distinction is that AI can accelerate the work surrounding implementation, while business leaders and implementation teams remain responsible for determining whether the resulting design is appropriate.

Requirements Analysis

ERP implementations often begin with a large volume of requirements gathered from different departments.

These requirements may come from workshops, spreadsheets, process documents, interviews, existing ERP configurations, policies, reports, and other business documentation.

AI can assist by analysing this information and helping implementation teams:

• classify requirements by business function;
• identify duplicate or overlapping requirements;
• group related requirements;
• identify recurring patterns;
• highlight inconsistencies;
• organise requirements into structured categories; and
• surface areas that require further investigation.

This can reduce the amount of manual preparation required before business and implementation teams make design decisions.

The objective is not to allow AI to decide what the business should require.

Rather, AI can help teams move more quickly from a large volume of unstructured information toward a more organised understanding of the current environment.

Process Mapping and Design

Understanding how work is performed today is one of the most important parts of ERP implementation.

Businesses may have different processes across departments, business units, locations, or subsidiaries—even when those processes are intended to achieve the same outcome.

AI can assist in analysing:

• existing workflows;
• process variations;
• approval patterns;
• recurring bottlenecks;
• manual handoffs;
• process dependencies;
• exceptions; and
• opportunities for standardisation.

This creates an opportunity to move beyond simply documenting the current state.

Implementation teams can use AI-assisted analysis to compare the as-is process with the intended to-be process, identify areas of unnecessary complexity, and determine where standard ERP functionality may be sufficient.

McKinsey’s analysis describes emerging agentic approaches that can compare existing processes with target-state designs and generate recommendations, while keeping humans responsible for direction and validation.

Configuration

ERP configuration requires translating business requirements into system capabilities, workflows, parameters, roles, and rules.

AI can assist consultants and implementation teams by helping map business requirements to available ERP functionality.

For example, AI could help identify:

Business Requirement → Relevant ERP Capability → Configuration Option → Validation Requirement

This can potentially reduce repetitive research and preparation during configuration.

However, configuration decisions still require business and technical judgement.

A recommendation generated by AI does not automatically mean that the proposed configuration is appropriate. Implementation teams must still consider business rules, integration requirements, security, reporting, scalability, compliance, and future operating requirements.

Testing

Testing is another area where AI can potentially reduce manual effort.

ERP implementations involve many test scenarios across finance, procurement, sales, inventory, manufacturing, projects, customer management, integrations, and other business processes.

AI can assist with:

• generating test cases from documented requirements;
• identifying relevant test scenarios;
• analysing process dependencies;
• generating variations of test conditions;
• identifying potential defects or inconsistencies;
• summarising test results; and
• helping teams identify areas that require additional testing.

McKinsey describes agentic AI approaches that can generate, execute, and validate test cases at scale while detecting and diagnosing issues. Human teams remain responsible for quality oversight and edge cases.

This distinction is important.

AI can increase testing capacity. It does not remove the need for business validation.

A test can pass technically while still producing a business outcome that is incorrect or unacceptable.

Documentation and Training

ERP implementations generate significant amounts of documentation.

This may include:

• process documentation;
• configuration documentation;
• standard operating procedures;
• user guides;
• test documentation;
• training materials;
• role-specific instructions; and
• change-management communications.

AI can help prepare and maintain these materials from approved project information.

For example, once a target process has been validated, AI can help convert that process into role-specific guidance for finance users, procurement teams, sales teams, warehouse personnel, project managers, or executives.

This can make documentation more scalable and easier to adapt as processes change.

McKinsey identifies documentation and training preparation among the implementation activities that could become heavily automated through agentic AI approaches.

What Remains Human

Greater AI involvement does not mean that ERP implementation becomes a fully autonomous process.

Some decisions require business context, accountability, judgement, and organisational authority.

Human involvement remains particularly important for:

• Business decisions — determining what the organisation actually needs;
• Validation — confirming that AI-generated recommendations are correct;
• Architecture decisions — determining how systems, integrations, data, and security should be structured;
• Governance — defining who can approve, change, or execute processes;
• Risk management — evaluating financial, operational, regulatory, and security implications;
• Change management — helping people adopt new processes and ways of working; and
• Final approval — accepting the target design and confirming readiness for deployment.

This human role becomes particularly important because ERP is not simply a software implementation.

It changes how people work.

McKinsey notes that change management is likely to remain a major constraint in future ERP transformation programmes, even as AI reduces the effort required for many implementation activities.

From Manual Implementation to AI-Assisted Implementation

The broader transformation can be viewed as:

MANUAL IMPLEMENTATION
People analyse → People design → People configure → People test → People document → People train

↓

AI-ASSISTED IMPLEMENTATION
AI analyses → AI recommends → People validate → AI assists configuration → AI accelerates testing → AI prepares documentation → People govern adoption

The objective is not to remove people from ERP implementation.

It is to allow implementation teams to spend less time on repetitive work and more time on the decisions that determine whether the ERP transformation creates business value.

The Strategic Implication

AI could make ERP implementation faster, more data-driven, and increasingly automated.

But faster implementation alone is not the objective.

The real opportunity is to improve the quality of the transformation itself:

Better analysis → Better design → Better configuration → Better testing → Better adoption → Better business outcomes

This also changes what businesses should expect from an ERP implementation partner.

The role increasingly extends beyond configuring software. Implementation teams need to understand business processes, data, integrations, AI capabilities, governance, security, and change management—and understand where AI can safely accelerate each stage.

The organisations that benefit most from AI-assisted ERP implementation will therefore not necessarily be those that automate the most.

They will be those that combine AI-enabled productivity with strong business governance, a clean data foundation, and disciplined implementation practices.

---

What AI Cannot Fix in an ERP System

AI can make ERP systems more intelligent, more responsive, and increasingly automated.

But AI is not a substitute for a sound ERP foundation.

An organisation can deploy sophisticated AI models, AI assistants, or AI agents and still struggle to create meaningful business value if the underlying ERP environment contains inaccurate data, disconnected systems, inconsistent processes, excessive customisation, outdated architecture, weak controls, or poor user adoption.

This leads to an important principle:

AI does not remove ERP complexity. It makes the quality of the underlying ERP foundation more important.

McKinsey’s 2026 analysis similarly emphasises that even as AI agents become more capable of executing enterprise processes, the underlying ERP architecture, data, application logic, and system-of-record capabilities remain important for reliability, auditability, compliance, and consistency.

Poor Data Quality

AI depends on the quality, relevance, context, and accessibility of the data available to it.

If customer records are duplicated, product information is inconsistent, inventory balances are unreliable, supplier records are incomplete, or financial data contains unresolved errors, AI does not automatically make that information trustworthy.

It may instead process the underlying information faster and produce an answer that appears convincing but is based on flawed inputs.

The basic principle is straightforward:

Poor Data → Poor Context → Unreliable AI Output

This is particularly important as AI moves from generating information to recommending or executing actions.

An inaccurate report may create confusion.

An inaccurate AI-generated recommendation may influence a business decision.

An AI agent acting on inaccurate data can potentially create an operational error.

Gartner identifies data quality and data readiness as important considerations for organisations deploying AI, while McKinsey’s 2026 research similarly identifies data readiness as a major foundation for scaling AI.

AI can help identify anomalies, classify information, and support data-quality processes.

It cannot make unreliable source data inherently reliable.

Fragmented Systems

Modern businesses rarely operate from a single application.

They may have separate systems for:

• ERP;
• CRM;
• e-commerce;
• warehouse management;
• payroll;
• project management;
• manufacturing;
• procurement;
• spreadsheets; and
• industry-specific applications.

When these systems are poorly integrated, important business context can remain distributed across disconnected environments.

For example, an AI system analysing inventory may have access to ERP stock records but not the latest warehouse information.

An AI system supporting customer service may have customer records but lack relevant project, billing, or service history.

The result is incomplete context.

Disconnected Systems → Incomplete Context → Limited Intelligence

AI can help connect information through integration technologies and data platforms, but it does not eliminate the architectural work required to establish reliable connections between systems.

Gartner identifies integration and legacy application architecture as important challenges when organisations attempt to build AI-enabled applications and agents.

The question is therefore not simply:

“Can we add AI?”

It is:

“Can the AI access the business context it needs, from the systems that contain authoritative information?”

Inconsistent Processes

AI can identify patterns in how a business operates.

It can analyse transactions, workflows, approvals, exceptions, and historical behaviour.

But that does not automatically determine which process the organisation should follow.

Consider a procurement process in which different departments use different approval thresholds.

One business unit may require three approval levels.

Another may allow direct approval below a certain amount.

A third may rely on informal email approvals.

AI can detect these patterns.

It does not automatically establish the correct governance model.

That requires business decisions.

Organisations need to determine:

• which process is the standard;
• which exceptions are legitimate;
• who has authority to approve;
• which controls are mandatory;
• where segregation of duties applies; and
• which activities should remain human-controlled.

Only after these decisions are defined can AI reliably support or automate the process.

This is one reason standardisation remains important in AI-enabled ERP environments. Gartner has highlighted the relationship between embedded AI, process standardisation, data quality, change management, and avoiding unnecessary customisation.

Excessive Customisation

ERP customisation is not inherently wrong.

Businesses sometimes require industry-specific capabilities, specialised workflows, integrations, or differentiated processes that standard ERP functionality does not fully address.

The issue is uncontrolled or deeply invasive customisation.

Highly customised ERP environments can make upgrades, integrations, testing, maintenance, and AI adoption more complicated.

Every additional modification can introduce another dependency that needs to be understood when new capabilities are introduced.

A modern approach is therefore increasingly focused on maintaining a stable ERP core while using appropriate extension and integration mechanisms around it.

Gartner’s 2026 research recommends moving away from invasive codebase changes toward strategically governed, decoupled extensibility so organisations can maintain a more stable and standardised ERP core.

The principle is:

Customise where there is genuine business value. Standardise where standard functionality is sufficient.

AI does not remove the technical debt created by excessive customisation.

Legacy Architecture

Older ERP environments can create limitations that become more visible as organisations introduce AI.

Potential constraints include:

• limited integration capabilities;
• difficult access to enterprise data;
• outdated interfaces;
• rigid application structures;
• limited scalability;
• fragmented databases;
• unsupported technologies; and
• security or maintenance constraints.

AI may sit on top of an existing system, but it still needs access to the information and processes underneath it.

If the architecture cannot reliably expose the required data or business functions, the AI layer may have limited practical value.

This does not mean every legacy ERP must immediately be replaced.

The appropriate response depends on the organisation’s business requirements, technical constraints, cost, risk, and transformation objectives.

Possible approaches may include:

IMPROVE → INTEGRATE → MODERNISE → REPLACE

The important point is to evaluate the foundation rather than assuming that adding an AI tool will solve architectural limitations.

McKinsey similarly describes the ERP data and application foundation as critical to the reliability and scalability of AI-enabled enterprise operations.

Weak Security and Governance

As AI moves from providing information to recommending and executing actions, security and governance become even more important.

An AI assistant that summarises information requires access to data.

An AI agent that performs a business process requires access to systems and actions.

Those capabilities must therefore operate within clearly defined boundaries.

Organisations need to consider:

• Identity — Who is the AI acting for?
• Permissions — What is it allowed to access?
• Actions — What is it allowed to execute?
• Approvals — Which actions require human approval?
• Auditability — What did the AI access and do?
• Monitoring — How is its behaviour evaluated?
• Data protection — What information can be exposed or processed?
• Accountability — Who is responsible for the outcome?

Gartner’s 2026 research on ERP agent governance highlights the need for stronger controls as AI agents become increasingly embedded in ERP environments.

Governance therefore cannot be treated as a document created after deployment.

For AI-enabled ERP, governance needs to become part of the operating environment itself.

Weak Change Management

Even technically successful AI and ERP implementations can fail to create value if people do not adopt the new way of working.

AI can change how employees interact with ERP.

Instead of navigating multiple screens and reports, users may increasingly interact through conversational interfaces, recommendations, alerts, or AI-assisted workflows.

This changes more than the technology.

It can change:

• user responsibilities;
• approval processes;
• decision-making;
• training requirements;
• job workflows;
• performance expectations; and
• accountability.

Employees therefore need to understand not only how to use the technology, but also why the process is changing and where human judgement remains important.

Training, communication, process redesign, leadership support, and user adoption remain critical components of ERP transformation.

Gartner’s research on AI in ERP highlights organisational change alongside integration and data quality as important considerations when activating embedded AI.

AI can accelerate a process.

It cannot make people automatically adopt it.

The Netsense 6 Dimensions of an AI-Ready ERP

The limitations above point to a broader principle: AI readiness is not determined by a single ERP feature.

At Netsense, we synthesise the foundation into six dimensions:

1. DATA
   Can the organisation provide accurate, relevant, accessible, and governed data?
2. ARCHITECTURE
   Can the ERP and surrounding systems exchange information reliably and support the required AI capabilities?
3. PROCESSES
   Are business processes sufficiently defined, consistent, and structured for AI to understand and support them?
4. SECURITY
   Can access, permissions, data protection, and business controls be enforced?
5. PEOPLE
   Do users and leaders have the skills, understanding, and confidence to work with AI-enabled ERP processes?
6. GOVERNANCE
   Are accountability, monitoring, approvals, auditability, and risk controls clearly defined?

Together:

DATA + ARCHITECTURE + PROCESSES + SECURITY + PEOPLE + GOVERNANCE

These six dimensions are a Netsense framework and synthesis for assessing AI-ready ERP foundations. They are not presented as an external industry standard or a formal Gartner, McKinsey, or other research framework.

The purpose is practical: to help businesses evaluate whether their ERP environment is capable of supporting AI use cases responsibly and at scale.

Why the Six Dimensions Matter

Dimension	Key Question	What Happens If It Is Weak?
Data	Can AI access reliable business information?	Outputs may be inaccurate or incomplete
Architecture	Can systems exchange the required information?	AI may lack context or integration
Processes	Are workflows and rules clearly defined?	AI may reproduce process inconsistency
Security	Are access and permissions controlled?	AI actions can create security and compliance risks
People	Are users prepared to work with AI?	Adoption and value realisation may remain low
Governance	Can AI activity be controlled and audited?	Risks become harder to manage at scale

The six dimensions also reinforce an important idea throughout this article:

AI readiness is an organisational capability, not simply an ERP feature.

A business may have access to advanced AI technology and still be unprepared to use it effectively.

Conversely, an organisation with strong data, connected architecture, disciplined processes, appropriate controls, capable people, and clear governance has a stronger foundation from which to evaluate and scale AI use cases.

The AI-Ready ERP Foundation

The relationship can be simplified as:

DATA
↓
ARCHITECTURE
↓
PROCESSES
↓
SECURITY
↓
PEOPLE
↓
GOVERNANCE
↓
AI-ENABLED BUSINESS VALUE

AI sits on top of this foundation.

It does not replace it.

That is why the most important question for an organisation evaluating AI in ERP may not be:

“Which AI feature should we activate?”

It may be:

“Is our ERP environment ready to use AI safely, reliably, and in ways that create measurable business value?”

That question provides the foundation for assessing whether an organisation should move forward with AI today, strengthen its ERP environment first, or pursue a broader ERP modernisation strategy.

---

Does AI Require a Cloud ERP?

No. AI does not automatically require a cloud ERP.

Businesses can use AI with different ERP environments, depending on the capabilities of the existing system, the AI use cases they want to implement, their integration architecture, data accessibility, security requirements, and the ERP vendor’s technology roadmap.

However, modern cloud ERP architectures can make certain aspects of AI adoption more practical.

Cloud platforms can provide scalable infrastructure, modern integration capabilities, continuously updated services, and easier access to new AI capabilities as they become available.

The distinction is important:

Cloud can make AI adoption more practical. Cloud alone does not make an ERP AI-ready.

Gartner’s 2026 research identifies cloud ERP as an important environment for the development of embedded AI, intelligent process automation, adaptive analytics, and AI-driven planning and forecasting. At the same time, Gartner notes that organisations still face challenges including data quality, integration complexity, skills gaps, and other adoption barriers.

Why Cloud Can Make AI Adoption More Practical

Modern cloud ERP platforms can provide several characteristics that support AI adoption.

Scalability

AI workloads can require additional computing resources as the volume of data, users, transactions, and AI-enabled processes increases.

Cloud infrastructure can provide the ability to scale these resources without requiring an organisation to design and maintain all underlying infrastructure itself.

This can be particularly relevant when AI moves from a limited pilot to multiple business processes.

For example, an organisation may initially use AI for financial reporting and anomaly detection.

Later, it may expand into:

• demand forecasting;
• procurement analysis;
• customer service;
• inventory optimisation;
• project forecasting;
• AI-assisted workflows; and
• agentic process automation.

The infrastructure requirements can therefore evolve as adoption expands.

Integration

AI rarely operates in isolation.

An AI capability may need information from the ERP, CRM, warehouse systems, e-commerce platforms, document repositories, manufacturing systems, or other business applications.

Modern cloud ERP ecosystems increasingly use APIs, integration services, connectors, and modular architectures to exchange information across these environments.

This can make it easier to connect AI capabilities with the broader enterprise technology landscape.

Gartner’s 2026 analysis describes connected data and flexible, cloud-based ERP architectures as important elements of an environment supporting embedded intelligence, AI agents, and adaptive experiences.

Data Accessibility

AI needs access to relevant business information.

Cloud architectures can make it easier to expose and connect data across applications and services, particularly when systems are designed around modern integration and data-access patterns.

But accessibility should not be confused with quality.

Making more data available does not automatically make that data accurate, consistent, or suitable for an AI use case.

The objective is therefore not simply:

More Data

but:

Accessible + Relevant + Reliable + Governed Data

McKinsey’s 2026 research on AI data readiness highlights data as a major constraint when organisations attempt to scale AI and emphasises the importance of a governed, reusable data foundation.

Continuous Platform Updates

Cloud ERP platforms are generally designed around vendor-managed updates and evolving platform capabilities.

This can allow organisations to access new functionality, security improvements, integrations, and AI capabilities without following the same upgrade model as heavily customised legacy environments.

This does not mean every new AI capability will automatically be available or suitable for every organisation.

Businesses still need to evaluate:

• functionality;
• licensing;
• security;
• data requirements;
• integration;
• governance;
• business value; and
• user adoption.

But a modern cloud platform can reduce some of the technical barriers associated with maintaining an ageing technology stack.

Modern APIs and Services

AI-enabled applications increasingly depend on the ability to interact with business systems through defined interfaces and services.

Modern APIs can allow AI capabilities to retrieve information, interact with business processes, or initiate approved actions without requiring users to manually move information between systems.

This becomes increasingly important as organisations move from AI-generated insights toward AI-assisted and agentic workflows.

The architecture therefore matters:

ERP DATA → API / INTEGRATION → AI CAPABILITY → DECISION / RECOMMENDATION → CONTROLLED ACTION

The ERP remains part of the operational foundation rather than simply becoming a data source for an external AI tool.

Why Cloud Alone Does Not Make ERP AI-Ready

Moving an ERP to the cloud does not automatically solve the underlying problems that can limit AI adoption.

A cloud ERP can still have:

• poor-quality master data;
• inconsistent business processes;
• disconnected applications;
• excessive customisation;
• weak governance;
• inadequate security controls;
• unclear ownership of data;
• insufficient integration;
• limited user adoption; or
• business processes that are not suitable for automation.

The same fundamental principle therefore applies:

Cloud is an architectural characteristic. AI readiness is an organisational and technological capability.

Consider two businesses.

Business A has a cloud ERP but inconsistent customer records, fragmented systems, poorly defined approval processes, and limited data governance.

Business B has a well-integrated ERP environment with reliable master data, standardised processes, defined permissions, strong governance, and accessible business information.

The fact that both use cloud ERP does not mean they have the same AI readiness.

This is why Gartner identifies data quality, integration, and organisational change as considerations when organisations activate embedded AI in ERP.

When an Existing ERP May Still Be Viable

Businesses should not assume that an existing ERP must be replaced simply because they want to adopt AI.

The first step is to evaluate whether the current environment can support the intended AI use cases.

A practical assessment should examine five areas.

Assessment Area	Key Question
Architecture	Can the current ERP expose the required data and business functions?
Integration	Can it connect reliably with AI tools and surrounding systems?
Data Quality	Is the data accurate, consistent, relevant, and sufficiently governed?
Vendor Roadmap	Is the ERP vendor investing in AI, APIs, security, and modern platform capabilities?
AI Requirements	Can the current environment support the specific AI use cases the business wants to implement?

This creates a more useful question than simply asking:

“Is our ERP old?”

The better question is:

“Can our current ERP architecture provide the data, integration, security, and scalability required by the AI use cases we want to implement?”

An older ERP may remain viable for some AI use cases if its data and integration architecture are sufficiently accessible and reliable.

Conversely, a relatively modern ERP may still require significant remediation if its data, processes, or integrations are poorly structured.

When Modernisation May Be Necessary

There are situations where the existing ERP environment may create structural limitations.

Examples include:

• the ERP cannot provide the required integration capabilities;
• critical data is fragmented across unsupported systems;
• the architecture cannot scale with business requirements;
• extensive customisation makes upgrades and integrations difficult;
• security or compliance requirements cannot be adequately supported;
• the vendor roadmap no longer aligns with the organisation’s needs; or
• the ERP cannot support the business processes required for future operations.

In these circumstances, AI may expose an existing ERP problem rather than solve it.

McKinsey’s 2026 research makes a similar point: while AI agents can be layered over existing enterprise applications, organisations can encounter a ceiling when legacy systems limit scale, control, and consistency. McKinsey therefore argues that thoughtful ERP modernisation and improvements to the data foundation remain important for scaling AI.

The appropriate response is not automatically “replace the ERP.”

Instead, businesses can evaluate four strategic paths:

IMPROVE → INTEGRATE → MODERNISE → REPLACE

Improve

Keep the existing ERP while improving:

• data quality;
• process consistency;
• governance;
• security;
• configuration; and
• user adoption.

This may be appropriate when the core ERP remains capable of supporting the business and the main limitations are foundational.

Integrate

Retain the ERP while connecting it with modern applications, data platforms, AI services, or other systems.

This can be useful when the ERP remains a reliable system of record but does not provide every capability required by the organisation’s broader technology environment.

Modernise

Upgrade, restructure, reconfigure, or migrate parts of the ERP environment to address architectural or operational limitations.

The objective is to create a stronger foundation without necessarily treating the entire ERP as obsolete.

Replace

Consider a new ERP when the existing platform has fundamental limitations that prevent the organisation from meeting its business, operational, integration, security, scalability, or technology requirements.

ERP replacement should therefore be treated as a business transformation decision—not simply an AI adoption decision.

Cloud ERP and the Future of AI-Enabled ERP

The relationship between cloud ERP and AI is likely to become increasingly important as ERP platforms evolve toward embedded intelligence, AI-assisted workflows, and agentic capabilities.

Gartner’s 2026 research describes cloud ERP as a key environment for developments including intelligent process automation, adaptive analytics, AI-driven planning and forecasting, and AI agents.

McKinsey, however, provides an important counterbalance: even as AI agents increasingly interact with enterprise systems, the underlying ERP data, application logic, and system-of-record capabilities remain important for reliability, auditability, compliance, and consistency.

This suggests that the future is not simply:

CLOUD ERP → AI

It is closer to:

ERP FOUNDATION → CONNECTED DATA → AI CAPABILITIES → INTELLIGENT WORKFLOWS → CONTROLLED ACTION → MEASURABLE VALUE

Cloud can provide an important architectural foundation.

But the business value comes from what the organisation can reliably do with its data, processes, systems, people, and AI capabilities.

The Executive Decision

For organisations evaluating AI and ERP together, the decision should therefore not begin with:

“Do we need to move to the cloud because we want AI?”

It should begin with:

“What AI-enabled business outcomes are we trying to achieve, and can our current ERP environment support them?”

From there, businesses can assess whether the appropriate path is to:

IMPROVE → INTEGRATE → MODERNISE → REPLACE

The right path depends on the organisation’s current architecture, data quality, processes, security requirements, integration landscape, vendor roadmap, business objectives, and AI use cases.

The goal is not simply to have a cloud ERP.

The goal is to build an ERP environment that can provide the data, context, intelligence, controls, and scalability required to create measurable business value from AI.

---

Is Your ERP Ready for AI?

Having an ERP system with AI capabilities does not necessarily mean that an organisation is ready to use AI effectively.

AI readiness depends on more than whether an ERP vendor offers an AI assistant, predictive analytics, generative AI, or AI agents.

It depends on whether the organisation has the data, architecture, processes, security, people, and governance required for the specific AI use cases it wants to pursue.

This distinction matters because AI readiness is not a permanent label.

A business may be ready to use AI for one relatively simple use case but not ready for another use case that requires deeper integration, higher data quality, greater automation, or more significant business authority.

For example, an organisation may be ready to use AI to:

• summarise management reports;
• identify unusual transactions;
• classify documents; or
• provide conversational access to ERP information.

But it may not yet be ready to allow an AI agent to:

• create or modify financial transactions;
• change supplier terms;
• release payments;
• automatically adjust production schedules; or
• execute high-value business decisions without human approval.

The level of readiness therefore depends on the use case, risk, data requirements, integration requirements, and degree of automation involved.

Gartner’s recent research on ERP data readiness similarly emphasises assessing data against the intended AI use case, while its broader ERP research identifies integration, data quality, change management, and standardisation as important considerations for embedded AI.

AI Readiness Is Use-Case Specific

There is no single switch that changes an organisation from “not AI-ready” to “AI-ready.”

Instead, readiness should be evaluated against a specific business objective.

Consider three different use cases:

AI Use Case	Typical Requirements	Relative Control Requirement
Management reporting assistant	Reliable ERP data, reporting context, user access controls	Lower
Cash-flow forecasting	Historical financial data, current transactions, appropriate forecasting context	Moderate
AI agent executing financial workflows	Reliable data, integrations, business rules, permissions, approvals, monitoring and auditability	Higher

The more an AI capability moves from information → recommendation → action, the more important the underlying foundation becomes.

This can be expressed as:

ASSIST → RECOMMEND → EXECUTE WITH APPROVAL → EXECUTE WITHIN DEFINED CONTROLS

The objective is not to maximise autonomy.

The objective is to establish the appropriate level of AI capability for the business problem and risk involved.

The Six Dimensions of AI Readiness

A practical AI-readiness assessment can be organised around six dimensions:

1. Data
2. Architecture
3. Processes
4. Security
5. People
6. Governance

These dimensions form the Netsense 6 Dimensions of an AI-Ready ERP framework introduced earlier in this article.

They should be viewed as a practical Netsense synthesis for assessing ERP readiness—not as an external industry standard or a formal Gartner or McKinsey framework.

1. Data

AI needs reliable and relevant information.

Executives should ask:

• Is our master data accurate?
• Are customer, supplier, product, inventory, and financial records consistent?
• Can we identify the authoritative source for important information?
• Is historical data sufficiently complete for the intended use case?
• Can relevant structured and unstructured data be accessed?
• Do we understand data lineage and ownership?
• Can data quality be monitored continuously?

Gartner’s research emphasises that AI-ready data must be evaluated according to the intended use case, including whether the data is representative, contextualised, governed, and suitable for the task.

McKinsey similarly identifies data readiness as a major constraint on scaling enterprise AI and emphasises the need for governed, reusable data foundations.

Key question:

Can we trust the data that AI will use to make recommendations or decisions?

2. Architecture

AI needs access to business context.

Executives should ask:

Can our ERP integrate with AI services?
Are APIs or other integration mechanisms available?
Can AI access the systems containing authoritative information?
Can information move reliably between ERP and surrounding applications?
Can the architecture scale as AI use cases expand?
Are legacy technologies creating integration constraints?

A modern AI architecture is not simply an AI model connected to an ERP database.

It may involve:

ERP → APIs / Integration → Data & Context → AI Capability → Business Workflow

McKinsey’s research emphasises that the ERP data and application foundation remains important even as AI agents become more capable, because reliability, auditability, compliance, and consistent business logic still depend on the underlying enterprise architecture.

Key question:

Can our technology environment provide AI with the business context it needs?

3. Processes

AI works within business processes.

Executives should ask:

Are our core processes clearly defined?
Are workflows consistent across departments?
Are approval rules documented?
Are exceptions understood?
Are unnecessary manual steps creating avoidable complexity?
Which processes should be standardised before automation?
Which decisions require human judgement?

A business should be cautious about automating a process that is already poorly defined.

If five departments follow five different versions of the same process, AI may simply reproduce or accelerate that inconsistency.

The objective should therefore be:

Standardise → Simplify → Automate → Optimise

Gartner identifies the adoption of standard ERP functionality and avoidance of unnecessary customisation as important elements of creating more adaptable ERP environments.

Key question:

Are our processes structured enough for AI to understand and support them reliably?

4. Security

AI introduces another layer of access to enterprise information and processes.

Executives should ask:

What information can an AI capability access?
Which users can invoke it?
What actions can it perform?
Are permissions inherited from existing ERP roles?
Which actions require human approval?
How is sensitive information protected?
Can AI activity be audited?
Are segregation-of-duties requirements preserved?

This becomes increasingly important as organisations move from AI assistants toward AI agents.

An AI system that can read information has one type of risk.

An AI agent that can modify records or initiate transactions introduces another.

McKinsey’s 2026 research on AI trust highlights that increasing AI autonomy raises the consequences of unintended actions and makes governance and controls foundational to responsible deployment.

Key question:

Can AI operate within clearly defined security, permission, and control boundaries?

5. People

AI adoption is also an organisational capability.

Executives should ask:

Do employees understand how AI will change their workflows?
Do users know when to trust an AI recommendation and when to verify it?
Are employees trained to work with AI-enabled ERP processes?
Are new roles or responsibilities required?
Do managers understand how AI changes decision-making?
Is there a clear process for reporting errors or unexpected AI behaviour?

Technology adoption can fail even when the underlying technology works.

Gartner identifies behavioural change, training, and AI literacy as important elements of scaling AI capabilities.

The objective is not to make every employee an AI specialist.

It is to ensure that people understand:

What AI can do → What AI cannot do → When to trust it → When to verify it → When human judgement is required

Key question:

Are our people prepared to work with AI rather than simply receive AI technology?

6. Governance

AI governance becomes particularly important when AI moves from providing information to influencing or executing business processes.

Executives should ask:

Who owns each AI use case?
Who approves deployment?
What actions can AI perform?
What actions are prohibited?
When is human approval required?
How are AI outputs evaluated?
How are errors and exceptions handled?
Can AI activity be monitored and audited?
How are changes to models, instructions, workflows, and permissions controlled?
How is business value measured?

For agentic AI, governance must extend beyond the question of whether an AI output is accurate.

The organisation must also consider whether the AI acted appropriately.

McKinsey’s 2026 research identifies governance, risk management, data and technology, and agentic AI controls as important dimensions of AI trust maturity.

Key question:

Can we control, monitor, audit, and improve AI after it goes into production?

Questions Executives Should Ask Before Activating AI

A practical executive assessment can bring the six dimensions together.

Data
Is our business data reliable?
Are critical master-data records accurate and consistent?
Can we identify the authoritative source for key information?
Is the data relevant to the AI use case we are considering?
Can data quality be monitored continuously?
Architecture
Can our ERP integrate with AI services?
Can relevant systems exchange data reliably?
Are APIs and integration capabilities available?
Can AI access the business context required for the use case?
Can the architecture scale if the pilot expands?
Processes
Are our core business processes standardised?
Are approval rules clearly defined?
Do we understand process exceptions?
Are there unnecessary manual steps or workarounds?
Is the process stable enough to automate?
Security
What information will the AI access?
What permissions will it have?
What actions can it perform?
Which actions require human approval?
Can AI activity be audited?
People
Are users prepared for AI-enabled workflows?
Do employees understand how responsibilities will change?
Is appropriate training available?
Do managers understand the implications for decision-making?
Is there a process for handling AI errors and exceptions?
Governance
Who owns the AI use case?
Who is accountable for its business outcome?
What controls and approval thresholds apply?
How will performance be monitored?
How will business value be measured?
What happens when the AI produces an unexpected result?
Vendor and Platform

Executives should also evaluate the ERP vendor itself.

Ask:

Does the vendor have a credible AI roadmap?
Are AI capabilities embedded within the ERP or dependent on disconnected tools?
How is enterprise data protected?
How are AI permissions and governance handled?
Can AI capabilities be extended as business requirements evolve?
How frequently is the platform updated?
Can the organisation scale from one AI use case to multiple workflows?

This matters because AI readiness is not only a question of the organisation’s current state.

It is also a question of whether the technology platform can support the organisation’s future direction.

From AI Experiment to AI-Ready Operating Environment

A business may have successfully completed an AI pilot without being ready to scale AI across the organisation.

For example, a finance team may deploy an AI assistant to summarise reports.

The pilot works.

But scaling AI into procurement, inventory, manufacturing, customer service, and financial workflows may reveal additional requirements:

More data → More integrations → More permissions → More governance → More users → More monitoring

This is why AI readiness should be treated as a capability that develops over time.

Gartner describes AI-ready data as an ongoing practice rather than a one-time state, while McKinsey similarly emphasises the need for governed and reusable data foundations as AI moves from experimentation toward scale.

The objective is therefore not:

“Become AI-ready once.”

It is:

“Build the capability to continuously assess, deploy, govern, and scale AI use cases.”

AI Readiness Categories

Rather than assigning an arbitrary numerical score, businesses can use descriptive categories to understand the condition of their ERP foundation.

Strong Foundation

The organisation has a relatively strong foundation for the AI use case being considered.

Typical characteristics may include:

reliable and accessible data;
connected architecture;
clearly defined processes;
appropriate security controls;
users prepared for adoption; and
established governance and measurement.

This does not mean every AI use case is immediately appropriate.

It means the organisation has fewer foundational barriers to evaluating and scaling suitable use cases.

Foundation Requires Strengthening

The organisation has several elements required for AI adoption, but specific gaps should be addressed before scaling.

Examples may include:

inconsistent master data;
incomplete integrations;
process variations;
limited AI skills;
unclear governance; or
insufficient monitoring capabilities.

A focused remediation programme may allow the organisation to proceed with selected lower-risk use cases while strengthening the foundation for more advanced applications.

Significant Foundation Gaps

The organisation has substantial limitations that may prevent AI from delivering reliable or scalable value.

Potential indicators include:

unreliable or fragmented data;
disconnected systems;
highly inconsistent processes;
significant legacy constraints;
unclear security controls;
limited organisational readiness; or
no clear governance model.

In this situation, the immediate priority may be to strengthen the ERP foundation before attempting advanced AI automation.

AI Readiness Is a Journey, Not a Checkbox

The three categories should not be interpreted as permanent labels.

An organisation can move from:

SIGNIFICANT FOUNDATION GAPS
↓
FOUNDATION REQUIRES STRENGTHENING
↓
STRONG FOUNDATION

And the assessment should be repeated as the organisation introduces more sophisticated AI use cases.

A business that is ready for AI-assisted reporting today may require additional capabilities before introducing AI agents that can execute financial or operational workflows.

This is why the appropriate sequence is often:

USE CASE → REQUIREMENTS → READINESS ASSESSMENT → GAP IDENTIFICATION → FOUNDATION IMPROVEMENT → PILOT → MEASUREMENT → SCALE

The technology should follow the business requirement—not the other way around.

The Executive Question

The most useful question is therefore not:

“Is our ERP AI-ready?”

It is:

“Is our ERP environment ready for the specific AI use cases we want to pursue—and what gaps must we address before those use cases can create measurable business value?”

That question changes the conversation from AI adoption as a technology project to AI readiness as a business capability.

And once the readiness gaps are understood, the next question becomes more practical:

What should the business invest in first?