top of page

AI in Business Central: Why the right approach matters

Nov 12
9 min read

AI is creating new opportunities for Business Central users, but adding AI without understanding the wider business can also magnify existing problems.



The difference between a valuable improvement and an expensive distraction is rarely the technology alone; It is the quality of the decisions around it - which problem to solve, what data. which processes the solution depends on, who is accountable, and how the result will be tested and maintained…


This was a timely theme for the Business Central community at Microsoft Directions EMEA 2026, pointing to an important question for every Business Central customer:

how do you put AI to work responsibly in the real conditions of your business?


For organisations already using Business Central, the opportunity is not simply to “add AI”. It is to improve the way work gets done, while protecting the accuracy, control and trust on which the business depends; that takes a view of the whole business, careful implementation, and the right expertise around the table.


 The risk is not AI on its own. It is AI without context


Business Central is not an isolated tool. It sits within a network of people, processes, data, integrations, applications, approvals and customer commitments. A change to one part of that network can affect another.


A purchasing process, for example, may involve suggested quantities, supplier lead times, approval limits, warehouse capacity, cash-flow constraints and commitments to customers. An AI capability that looks useful when considered only as a way to generate a purchase suggestion, may have very different consequences when these wider dependencies are considered.


The same is true of sales, finance, inventory, service and reporting. A task that looks routine in a demonstration may involve exceptions, judgement and local knowledge in day-to-day operations. An AI system can help with some parts of the task, but it cannot be assumed to understand every business rule simply because it can read a record or produce a fluent answer.


That is why a “quick AI win” can turn into a difficult support problem. The first question should not be, “Where can we use AI?” It should be, What business outcome are we trying to improve, and what must remain true while we improve it?


How disconnected decisions create avoidable risk


AI is often discussed as if it were one capability. In practice, Business Central customers may encounter different features and approaches: built-in Copilot functionality, agents, integrations, custom extensions or AI services connected through other Microsoft tools. These are not all the same, and they do not necessarily have the same permissions, behaviour, cost, oversight or operational implications.

Treating them as interchangeable, or starting with a tool before defining the need, can lead to several kinds of business risk.


Risk - Solving the wrong problem


A business may invest in automating a task that is visible and easy to demonstrate, rather than one that is genuinely costly, slow or error-prone. The resulting solution may work as designed and still fail to deliver meaningful value.


For example, speeding up the preparation of a report may not solve the underlying problem if the report depends on inconsistent data, unclear definitions or repeated manual corrections. Automating a workaround can make the workaround faster, but it may also make its weaknesses harder to see.


A sound starting point is to identify the process, the people involved, the pain point and the outcome that would count as success. That makes it possible to compare options honestly, including the possibility that a process change, better training, improved data quality or a standard Business Central capability is more appropriate than a bespoke AI solution.


Risk - Automating a process that is not understood


Many business processes appear straightforward until someone maps the exceptions. What happens when a supplier changes a delivery date? When a customer has a credit hold? When a purchase invoice does not match the expected order? When an approval is delegated? When an item is substituted?


If those rules live only in the heads of a few experienced employees, an AI project may reveal a deeper process and knowledge-management issue. If the rules are unclear or inconsistent, the AI may produce inconsistent results too. If exceptions are not mapped, the system may not know when to stop and ask a person.


Before a task is automated, teams should understand both the standard route and the important exceptions. They should also decide which cases are safe to handle automatically, which require review, and which should remain with a person.


Risk - Trusting output because it sounds convincing


AI-generated language can be clear and confident without being complete or correct. In an ERP context, the important question is not whether an answer reads well. It is whether it is accurate enough for the decision being made, based on the right information, and presented with an appropriate way to verify it.

 

This matters particularly in finance, purchasing, stock management and customer-facing work. A mistaken summary may be inconvenient; a mistaken action or an unreviewed recommendation may affect postings, orders, supplier relationships, customer commitments or management decisions.


A responsible design gives users a way to review important outputs, understand where relevant information came from, and raise an exception. Human review should be designed into the process where the consequences warrant it, not added as a vague instruction to “check the AI”.


Risk - Making assumptions about permissions and data


One should not assume that every AI feature accesses or handles Business Central data in the same way. Microsoft’s current Business Central documentation says that its Copilot experience for working with Business Central data is governed by the user’s existing permissions, and describes that experience as read-only.


Microsoft also documents administrative controls for enabling or disabling individual Copilot and agent capabilities, managing access and applying data-governance settings. The details depend on the capability being used, so it is important to check the relevant documentation and configuration, rather than generalise from one feature to another.


Agents require particular care because their purpose can include carrying out steps, not merely answering questions. Microsoft’s documentation for its AI development toolkit describes agents as operating in Business Central with configured permissions and a user profile, and highlights transparency, intervention and change-control mechanisms.


The same documentation explains that an agent’s effective permissions can depend on the intersection of the agent’s permissions and the relevant user’s permissions. These controls are important, but they still need to be configured and tested appropriately for the customer’s specific scenario.


The practical lesson is simple: understand the exact capability, data flows, permissions and controls involved before enabling or extending it; Avoid relying on assumptions based on product labels or a demo.


Risk - Overlooking the systems around Business Central


Business Central often connects to Microsoft 365, Power Platform, banking services, e-commerce platforms, warehouse systems, payroll tools, document services and industry applications. AI introduced into one part of this environment may depend on data or steps in another.

 

If an integration is fragile, a data field has inconsistent meanings, or a process relies on a manual hand-off, an AI layer does not automatically repair the underlying issue. It may simply make the chain more complex. A change can also affect who owns support, how failures are diagnosed and which supplier is responsible for resolving them.


This is one reason a holistic assessment matters for your business; it should include the process, the data, the users, the integrations, the security model and the operational support and not just the feature being considered.


Risk - Underestimating change for people


A system change is also a change to how people work. Users may worry about the accuracy of AI output, whether tasks or roles will change, or what they are expected to check. Managers may assume that a new capability will be adopted automatically once it is available.


If users do not understand what the tool can and cannot do, they may either avoid it or trust it too readily. If the process changes but training and responsibilities do not, work may be duplicated or important checks missed.


Good adoption is practical. It explains why the change is being made, how the new process works, what users should review, when to escalate a problem, and who remains accountable for the outcome. The goal is not merely to switch on a feature; it is to help people use it safely and consistently.


Risk - Failing to plan for testing and ongoing ownership


A successful demonstration is not the same as a dependable production process. Real usage brings varied data, unusual cases, permission differences and changes to the surrounding system. A capability that works for one clean example may behave differently across a wider range of records and scenarios.


Testing should therefore cover representative cases, exceptions, failure paths and user roles, not only the good path. Teams should establish what counts as an acceptable result, what needs human approval, how problems are reported, and who is responsible for changes after launch.


The business-wide view: what to examine before starting


Enhance Dynamics knows that a considered AI project begins with the business, not the model. Before choosing a feature or commissioning a development, we ask questions such as:


  • What business outcome does our customer want? The specifics about the problem and the improvement sought.


  • Who performs the process today?  Including the people who handle exceptions and downstream effects.


  • What data does the process depend on? We check quality, ownership, definitions and access.


  • Which Business Central features, extensions and integrations are involved?


  • What may the AI read, recommend or do? Defining boundaries clearly.


  • Where is human review required? Identifying actions or decisions that need approval.


  • How will we test it? Including exceptions and realistic operating conditions.


  • How will we measure value and risk?  Agreeing what success looks like before launch.


  • Who owns it after implementation? Assigning responsibility for support, configuration and review.


This does not mean every project needs a lengthy programme or a large governance committee. It means the level of planning should match the business impact and the degree of autonomy involved, which an expert Business Central partner, like us, can provide.


The reward of getting it right


The risks are clear, but the point is not to avoid AI. With the right use, appropriate safeguards, and expertise, AI can help people work more effectively and make better use of the information already available to them, boosting business productivity.


A well-designed solution may help reduce repetitive work, make relevant information easier to find, support consistent processes or give employees more time for tasks that need judgement and customer understanding. It can also help a business build on its existing investment in Business Central, provided the new capability fits the way the organisation actually operates.


The rewards are more likely when the project is grounded in practical business needs:


  • A clearer process:  Your teams understand what should happen, who owns each step and when an exception needs attention


  • More useful automation:  Technology supports the work that matters instead of automating a visible but low-value task


  • Better user confidence:  Your people know what the AI is intended to do and where they remain responsible.


  • More reliable decisions: Data, permissions and review points are considered as part of the design


  • A stronger foundation for future change:  Your business learns how to assess, test and govern AI capabilities rather than treating each new feature as an isolated experiment


These outcomes do not come from AI alone. They come from matching the technology to a real need, preparing the process and people, and keeping a clear line of accountability.


We start with a controlled, measurable use case


We believe that a sensible way forward is to choose one use case with a clear business owner and a manageable scope. Defining the current process first, including the time it takes, the common exceptions and the problems users encounter. Then agreeing what the proposed change will do, and what it will not do.


Setting measures before implementation is key. Depending on the use case, these might include time spent on a task, the frequency of corrections, the number of cases escalated, user confidence or another operational measure. Choosing measures that reflect business value, not simply how often a tool is used.


Next, we agree the controls: who can access the feature, what permissions it needs, where a person must approve an action, how users will review output, and how an issue can be raised. We test the use case against realistic examples before widening access.


Once it is in use, we review performance and user feedback, adjusting the process when the evidence shows that something is not working as intended.


A controlled pilot is not just a smaller rollout. It is a chance to learn whether the business assumption is correct, whether the data is fit for purpose, and whether the proposed process is practical for the people who will use it.


Why the right Business Central partner matters


We are an experienced partner, and can help connect the technology choices to the realities of your business. That means asking questions beyond “Can this be built?” We look at whether it solves the right problem, Fits your existing Business Central environment, dependencies, how access will be controlled, how the solution is tested and supported, and what happens when the process or product changes…


Sometimes the right outcome will be a standard Business Central capability; sometimes it may involve better configuration, an improved process, training, an integration or a carefully scoped extension. AI should be selected because it fits the need, not because it is the newest option.


Enhance Dynamics also brings knowledge of the Microsoft business applications ecosystem, with a wealth of added-value partner solutions, and experience in helping customers plan and implement these solutions.

 

For current Business Central users, the important next step is to simply have a no-obligations conversation with us. AI can create real value for Business Central customers like you, but only when it is introduced with a clear view of the whole business. Without that view, organisations risk automating the wrong process, amplifying poor data, overlooking exceptions, confusing users or adding a new point of failure to an already complex environment.


With a considered approach, those risks can be addressed before they become operational problems.


So, if you are considering AI in your existing Business Central environment, or want to assess an idea before committing to it, get in touch with us to discuss your business needs, the risks and a sensible way forward.





Comments


Commenting on this post isn't available anymore. Contact the site owner for more info.
bottom of page