Is your business ready for AI?A practical ERP, data and process readiness checklist

Artificial intelligence is moving quickly from something businesses talk about to something they can actually use. AI capabilities are appearing inside ERP platforms, analytics tools, automation technologies and everyday business applications. For manufacturing and distribution businesses, the opportunities are becoming increasingly practical: finding information more quickly, analysing operational data, identifying patterns, improving forecasts, automating repetitive tasks and helping employees make better decisions.

But I think there is an important question businesses need to answer before asking what AI can do for them:

Is the business underneath it ready?

If customer records are duplicated, inventory information cannot be trusted, important processes happen outside ERP and different departments disagree about their numbers, adding AI does not remove those underlying issues. In some cases, it can simply expose them more quickly.

After working on ERP implementations, upgrades and business improvement projects, I’ve seen how much difference the foundations make. The organisations best placed to take advantage of new technology are not necessarily those adopting it first. They are often those that understand their processes, maintain dependable data and know what business problem they are trying to solve.

So before starting an AI project, I would assess six areas:

Data → Process → Systems → Governance → People → Use Case

Together, these form the Blue Sky ERP AI Readiness Framework.

Key takeaway: What does AI readiness mean for an ERP business?

AI readiness is the ability of a business to use artificial intelligence effectively, safely and for a clearly defined purpose. For manufacturing and distribution businesses, that usually depends on more than access to an AI tool.

It requires:

• Reliable and accessible business data
• Consistent operational processes
• Connected systems
• Appropriate security and governance
• Employees who understand how AI will support their work
• A worthwhile business problem to solve

ERP can form an important part of that foundation because it contains information across areas such as customers, suppliers, inventory, purchasing, production, sales and finance. But having ERP data is not the same as being ready to use it effectively with AI.

AI is becoming part of ERP

The conversation around AI and ERP has changed significantly. AI is no longer necessarily a separate technology experiment sitting outside the systems people use every day. Increasingly, AI capabilities are being built into business applications and operational workflows.

For Epicor customers, Epicor Prism is one example of that change. Epicor is developing AI capabilities that allow users to interact with ERP information using natural language, surface insights and increasingly support actions within ERP workflows.

That direction is important. For years, ERP has primarily been a system of record: it records orders, purchases, inventory movements, production activity, invoices and financial transactions. The next stage is about helping people make better use of that information, but there is an obvious dependency. The value of AI depends heavily on the quality, context and accessibility of the information and processes it is working with.

Before asking: “How can we introduce AI?” I think businesses should first ask: “What will AI actually have to work with?”

The Blue Sky ERP AI Readiness Framework

I would assess AI readiness across six connected areas.

1. Data — Can you trust the information?

Start with data. Your ERP may contain thousands or millions of records, but quantity is not the same as quality. Consider customer records.

If the same customer exists under several accounts, addresses are inconsistent, classifications are missing and sales information is maintained separately outside ERP, an AI application attempting to analyse customer behaviour starts from an incomplete picture.

The same applies to:

• Part and product records
• Supplier information
• Inventory
• Bills of material
• Routings
• Costs
• Lead times
• Pricing
• Sales history
• Purchasing history
• Production transactions
• Delivery performance

Before considering an AI use case, ask:

• Is the relevant data complete?
• Is it entered consistently?
• Are duplicate records controlled?
• Is master data actively maintained?
• Can we identify the authoritative source?
• Do users trust the information?
• Can we explain what important fields actually mean?
• Is historical information sufficiently consistent to analyse?

Inventory is a good example. As we explored in Why Your Inventory Is Never Accurate (And Why It’s Usually Not a Warehouse Problem), inventory discrepancies often originate much earlier in the process through delayed transactions, inconsistent working practices, disconnected systems and manual workarounds.

AI doesn’t make those underlying transactions more accurate simply by analysing them. One useful test is surprisingly simple: Ask three departments for the same business number. If they return three substantially different answers and nobody can explain why, AI readiness probably isn’t your immediate problem. Data confidence is.

Our guide to Why Manufacturing Reports Disagree: How to Create One Trusted Version of the Truth looks at this problem in more detail, including the effect of different source systems, definitions, refresh times and manually maintained spreadsheets.

2. Process — Does the business operate consistently?

Good data and good processes are closely connected. If employees follow different processes for the same transaction, the resulting data will reflect those differences.

For example, imagine two teams processing customer orders. One enters information immediately into ERP. Another records part of the information in a spreadsheet and updates ERP later. A third manages exceptions through email.

The business may ultimately complete every order successfully, but the digital record of how those orders progressed is inconsistent. That matters when technology begins looking for patterns, exceptions or recommendations.

Before applying AI to a process, understand how that process actually works today. Ask:

• Is there an agreed process?
• Do employees actually follow it?
• Where does information leave the ERP?
• Which steps rely on spreadsheets?
• Where is information entered more than once?
• Which decisions depend on email or informal communication?
• Which exceptions aren’t recorded consistently?
• Where does the documented process differ from reality?

This follows the same principle we apply to automation: Don’t use technology to accelerate a process you haven’t understood.

In our recent article The Hidden Cost of Operational Workarounds in Manufacturing, Dr Dele Oyebande explored how temporary spreadsheets, manual processes and offline systems can gradually become part of normal operations.

Those workarounds matter for AI readiness because they may contain important business context that isn’t captured in the ERP at all. Sometimes an AI project will identify an opportunity for improvement. Sometimes the better answer is simply to fix the process first.

3. Systems — Can information flow where it needs to?

Few businesses operate entirely within one application. A manufacturing or distribution organisation might use:

• ERP
• CRM
• Warehouse systems
• Ecommerce
• Quality systems
• Production systems
• CAD or engineering applications
• Power BI
• Microsoft 365
• Supplier or customer portals
• Specialist industry applications

That isn’t necessarily a problem. The question is whether those systems form a coherent information environment. If employees regularly export information from one application, manipulate it and manually enter it into another, you may have an integration problem rather than an AI opportunity.

Review:

• Which systems contain important operational information?
• Which application is authoritative for each type of data?
• Which integrations already exist?
• Where is information transferred manually?
• Are integrations reliable and monitored?
• Can the necessary information be accessed securely?
• Are there important data silos outside ERP?
• Is business logic applied consistently when systems exchange information?

AI becomes considerably more useful when it can operate within a connected environment rather than trying to compensate for fragmented systems.

This is also why Integration & Automation should be considered as part of the wider AI-readiness discussion. Connecting systems and removing unnecessary manual hand-offs can sometimes create value before AI is introduced at all.

4. Governance — Who owns the data and the decision?

This is one of the areas I think could easily be overlooked as businesses experiment with AI.

• Who is responsible for the information being used?
• Who should have access to it?
• Who determines whether an AI-generated recommendation is appropriate?
• What happens when it is wrong?

The answers will depend on the use case, but they should not be left until after implementation. Consider an AI application recommending purchasing action. There is a significant difference between: “These five parts appear likely to require attention.” and: “Create purchase orders for these five parts.”

The second moves from insight towards action. That means businesses need to think about:

• Data ownership
• User permissions
• Confidential or commercially sensitive information
• Approval authority
• Auditability
• Human review
• Responsibility for outcomes
• How exceptions are handled

ERP environments already contain security, roles and approval structures. AI initiatives should work with appropriate business controls rather than bypassing them. The objective shouldn’t be to remove people from every decision. It should be to decide where human judgement adds value and where technology can safely remove unnecessary effort.

5. People — Will employees understand and trust it?

A technically successful project can still fail if nobody uses it. ERP implementations have taught the industry this lesson many times.

You can configure a system correctly, migrate the data and make it technically available, but if users don’t understand why a process is changing or how the new approach helps them, adoption will suffer.

AI will be no different. Employees may reasonably ask:

• Why are we introducing this?
• What information is it using?
• Can I trust the answer?
• What am I expected to do with the recommendation?
• What happens if it’s wrong?
• Is it replacing part of my role?
• Who do I speak to when something doesn’t look right?

Those aren’t barriers to adoption, they’re sensible questions. Involving the people who perform the process every day can also improve the project itself. They often understand exceptions and operational realities that aren’t obvious from a process map.

Before introducing AI into a workflow:

• Explain the problem being solved
• Involve relevant users early
• Define where human judgement remains necessary
• Provide appropriate training
• Create a way for users to challenge or report poor outputs
• Measure whether the tool is actually helping them

AI adoption should be treated as a business change project, not simply a software deployment.

6. Use Case — What problem are you actually trying to solve?

This is where I would start before spending significant time looking at products.

Not: “Where can we use AI?” But: “Which business problem are we trying to solve?” There are plenty of potentially useful areas to investigate. For a manufacturing or distribution business, examples might include:

• Finding ERP information more quickly
• Analysing purchasing or supplier information
• Identifying unusual inventory patterns
• Improving demand forecasting
• Supporting customer service
• Investigating operational exceptions
• Reducing repetitive administrative work
• Analysing documents
• Helping employees access internal knowledge
• Supporting reporting and analysis
• Identifying orders or transactions requiring attention

But every use case should have an outcome.

Instead of: “Use AI in purchasing.” Define something more useful: “Reduce the time purchasing spends identifying and investigating material shortages.”

Or: “Help customer service find accurate order information without manually checking several systems.” Or: “Reduce the time required to investigate common ERP support questions.”

Once the problem is clear, you can assess:

1. What information is required?
2. Is that information reliable?
3. Which process does it affect?
4. Which systems are involved?
5. What decisions could AI support?
6. Where is human approval required?
7. How will we measure whether it worked?

That creates a much stronger basis for investment.

AI Readiness Checklist: Is your business ready for AI?

Use the Blue Sky ERP AI Readiness Checklist to assess six areas of your business: data, processes, systems, governance, people and use case.

You don’t need to answer “yes” to every question before experimenting with AI. But the more “no” answers you have, the more useful it may be to improve the foundation first.

Data

☐ We know which systems contain the data required for our proposed AI use case.
☐ Important master data is actively maintained.
☐ Duplicate and incomplete records are understood and controlled.
☐ Users generally trust the underlying ERP information.
☐ Important business definitions are documented.
☐ We can trace key information back to an authoritative source.

Process

☐ The process we want to improve is clearly understood.
☐ Employees generally follow a consistent process.
☐ We know where spreadsheets and manual workarounds are used.
☐ Important exceptions are recorded rather than held only in people’s heads or inboxes.
☐ We have challenged unnecessary steps before attempting to automate them.

Systems

☐ We understand which applications are involved in the process.
☐ Each important type of information has a defined source.
☐ Critical systems are appropriately integrated.
☐ We understand where employees manually move information between applications.
☐ The required information can be accessed securely.

Governance

☐ Ownership of important data is clear.
☐ Appropriate access permissions are defined.
☐ We know which AI outputs require human review.
☐ Responsibility for decisions remains clear.
☐ We have considered security, confidentiality and audit requirements.
☐ There is a process for dealing with incorrect or unexpected outputs.

People

☐ Users understand why the AI initiative is being considered.
☐ Relevant operational employees are involved.
☐ Training requirements have been considered.
☐ Users understand when they should question an AI output.
☐ We have identified who will own the solution after implementation.

Use Case

☐ We can describe the business problem in one sentence.
☐ The use case has a measurable outcome.
☐ We know the current baseline.
☐ AI offers a meaningful advantage over simply improving the existing process.
☐ We can start with a controlled scope.
☐ We know how success will be measured.

How ready is your business?

I wouldn’t turn this into an artificial score where 24 points means “AI ready” and 23 means “not ready”. Businesses are more complicated than that. Instead, look for patterns.

If most of your uncertainty is in Data, start with data quality and reporting confidence. If it is in Process, understand how the work actually happens before automating it. If it is in Systems, investigate integration and information flow. If it is in Governance, establish ownership, permissions and appropriate controls. If it is in People, work on involvement, training and adoption. If it is in Use Case, don’t buy anything yet.

Define the problem first. That is far more useful than a generic AI maturity score.

Where does ERP fit into AI readiness?

ERP can provide a particularly valuable foundation because it sits across so many business processes. Depending on the organisation, ERP may contain or connect information relating to:

• Customers
• Suppliers
• Sales orders
• Purchasing
• Inventory
• Production
• Costs
• Finance
• Shipping
• Demand
• Capacity
• Quality

That makes ERP information potentially valuable for AI, but ERP should not automatically be assumed to contain everything AI needs.

Some information may sit in CRM, engineering, warehouse, ecommerce, quality or other systems. Some may exist in documents. Some may still be held in spreadsheets. The goal should therefore be connected, governed information, not simply “put everything into ERP”.

This is closely related to the challenge explored in Future Proof Datasets: adding an AI assistant on top of reporting doesn’t remove the need for a well-structured, understandable data foundation.

For Epicor customers, this is becoming increasingly relevant as AI capabilities such as Epicor Prism become part of the ERP environment. The technology is moving forward. The question for individual businesses is whether their own operational foundations are moving forward with it.

Five warning signs you’re not ready to scale AI yet

Experimentation is useful. But I would be cautious about scaling an AI initiative if:

1. Nobody trusts the underlying data

If every AI answer immediately needs somebody to check three spreadsheets, the foundation needs attention.

2. The process isn’t understood

If five employees describe the same process in five different ways, map the process first.

3. Nobody owns the outcome

If IT thinks Operations owns the project and Operations thinks IT owns it, establish accountability before going further.

4. There is no measurable business problem

“Using AI” is not a business outcome.

5. The proposed solution adds another disconnected system

If an AI project creates another information silo or manual hand-off, it may increase complexity rather than reduce it.

Start small, but start with something that matters

Being cautious about AI doesn’t mean doing nothing. In fact, waiting until every data-quality issue and process problem has been eliminated isn’t realistic either. No business has perfect data. No process is perfect.

A more practical approach is to choose a contained, worthwhile use case. Select something where:

• The problem is understood
• The data is accessible
• The risk is manageable
• The people involved can participate
• The result can be measured

Establish the baseline, test the approach, validate the outputs and learn from the users. Then decide whether to expand it. A small project that solves a real problem is more valuable than a large AI programme without a clear purpose.

Ten questions to ask before your first AI readiness meeting

If AI is appearing on your management agenda, I would start with these:

1. What specific business problem are we trying to solve?
2. What data would AI need to solve it?
3. Do we trust that data today?
4. Where is that information currently stored?
5. Is the underlying business process consistent?
6. Where are people currently compensating with spreadsheets or manual work?
7. Who owns the data and the process?
8. Which decisions must remain subject to human review or approval?
9. How would employees know whether an AI output was wrong?
10. What measurable improvement would make the project worthwhile?

If those questions generate a useful discussion, you’re already making progress. If nobody can answer them, that’s useful information too.

AI readiness may expose wider ERP opportunities

One of the useful things about assessing AI readiness is that it can reveal improvements worth making regardless of whether an AI project goes ahead. You might discover:

• Duplicate master data
• Unused ERP functionality
• Poorly defined processes
• Reporting inconsistencies
• Manual integrations
• Unnecessary spreadsheets
• Missing ownership
• Training gaps
• Repetitive administrative work
• Business-critical knowledge held by one employee

Fixing those issues can improve the organisation today. Better inventory information helps purchasing today. More consistent processes help operations today. Better integration removes duplicate entry today. Clearer reporting helps management today.

AI can then build on a stronger operational foundation rather than becoming another layer placed on top of existing complexity.

Final thoughts

AI is going to become a more normal part of ERP and business software. For me, that makes readiness more important, not less. Manufacturing and distribution businesses don’t need to become AI companies. They need to understand where AI can improve the way their own business operates.

That starts with the basics:

Can we trust our data? Do we understand our processes? Are our systems connected appropriately? Do we have the right controls? Are our people involved? Are we solving a real problem?

If the answer to some of those questions is no, that doesn’t mean your business should avoid AI. It tells you where to start.

At Blue Sky ERP, we work with manufacturing and distribution businesses to understand how their ERP systems, data and processes operate in practice, and where technology can deliver meaningful improvement. If AI is starting to appear on your technology roadmap, but you’re unsure whether your ERP environment and wider business are ready for it, talk to the Blue Sky ERP team about where you are today and what needs to happen next.

Planning for AI but unsure where to start?

Talk to Blue Sky ERP about your ERP and AI readiness.

Frequently Asked Questions

What does AI readiness mean for a business?

AI readiness means having the data, processes, systems, governance, people and clearly defined use cases required to use artificial intelligence effectively. It is not simply a question of having access to AI software.

Does a business need perfect data before using AI?

No. Few organisations have perfect data. The important question is whether the data required for a particular use case is sufficiently accurate, consistent, accessible and governed for the decision or process it will support.

Why is ERP important for AI?

ERP contains and connects important operational information across areas such as sales, purchasing, inventory, production and finance. This can provide valuable context for AI applications, although other systems and data sources may also be required.

Can AI fix poor ERP data?

AI may help identify patterns or anomalies, but businesses should not assume it will automatically correct underlying data-quality problems. Poor master data, inconsistent processes and unclear ownership still need to be understood and managed.

What should a business do before implementing AI?

Start with a specific business problem. Identify the process and data involved, assess data quality, understand the systems required, establish governance and ownership, involve the relevant users and define how success will be measured.

How can manufacturers and distributors use AI with ERP?

Potential uses include finding ERP information through natural-language queries, analysing operational data, supporting forecasting, investigating exceptions, improving customer service, processing documents and reducing repetitive administrative work. The appropriate use case depends on the organisation’s processes, data and technology environment.

What is Epicor Prism?

Epicor Prism is Epicor’s portfolio of industry-focused AI capabilities designed to help users interact with ERP information and workflows using technologies including natural-language interaction and AI agents. Availability and functionality can vary by Epicor product, region and licensing, so businesses should confirm the capabilities relevant to their environment.

Is Epicor Kinetic ready for AI?

Epicor is introducing AI capabilities into its ERP portfolio, including Epicor Prism within Kinetic. However, a technically AI-capable ERP platform does not automatically mean an individual organisation is ready to use every AI capability effectively. Data quality, processes, integrations, governance and user adoption remain important considerations.

What should we do if we don’t know whether we’re ready for AI?

Start by identifying one worthwhile business problem and assessing the data, process, systems, governance and people involved. If that exposes wider ERP or operational issues, address those foundations before committing to a larger AI programme.