Why ERP Is the Hardest Place to Put AI, and Why That Matters
There is no shortage of AI tools available to businesses right now. The question worth asking is why almost none of them operate inside the ERP — and what it means that some now do.
There is no shortage of AI tools available to businesses right now. Productivity assistants, document summarisers, meeting transcribers, code generators. The list grows every week. Most of them are useful. Most of them are also entirely separate from the systems that run the business.
That distinction matters more than it appears.
The system of record is different
A CRM holds contacts. A project management tool holds tasks. A communication platform holds messages. These are valuable, but they are not the system of record. The system of record is the ERP — the environment where purchase orders are raised, invoices are generated, subscriptions are managed, prices are applied, and financial periods are closed. It is the one system where a mistake has consequences that cannot simply be undone with an undo button.
This is why ERP is the hardest place to put AI — and why most AI tools stop at the boundary.
When AI assists with a document or a meeting summary, the cost of an error is a correction. When AI acts inside an ERP, the cost of an error is a wrongly posted transaction, a mispriced contract, an access permission granted to the wrong person, or a financial period that closes on data nobody audited. The stakes are categorically different.
The gap between recording and acting
Business Central is exceptionally good at recording what happens. A purchase order arrives over budget — BC records it. A subscription renews at the wrong price — BC records it. An approval sits unactioned for eleven days — BC records that too.
What BC was not designed to do is act. To reason about what has happened and decide what should happen next. To apply a rule, execute a workflow, or escalate a decision — not because someone clicked a button, but because the system understood the situation and knew what the business required.
This gap between recording and acting is where significant operational cost accumulates. Finance teams spend hours each week reviewing exceptions that the system flagged but could not resolve. Consultants build workflows in Power Automate that cover eighty percent of cases and break on the other twenty. Managers approve changes they do not fully understand because the system gave them no context.
The gap is not a people problem. The people are doing what the system requires of them. The gap is structural.
Why AI in ERP requires a different standard
There is a phrase worth internalising: the system of record demands a different standard of AI than any other application.
An AI assistant that gets something slightly wrong in a draft email is a nuisance. An AI system that gets something slightly wrong inside the ERP is a liability. The difference is not one of degree — it is one of category.
This is why the conversation about AI in ERP cannot be the same conversation as AI everywhere else. It cannot start with capability. It must start with governance.
Before any AI system acts inside Business Central, there are questions that must be answered. Not as a compliance exercise, but as a practical requirement of operating in a system of record.
Who approved this action? What rule triggered it? How confident was the system in its decision? What was the confidence threshold at which it proceeded automatically versus escalated for human review? And if the action turns out to be wrong — can it be reversed, and is there a complete record of what happened and why?
These are not difficult questions to ask. They are, however, difficult questions to answer if the AI system was not designed with them in mind from the start.
The governance question is the AI question
There is a version of AI in ERP that is genuinely useful. It acts within rules the business defines. It surfaces its reasoning before high-value decisions are executed. It maintains a complete audit trail — not as a retrospective record, but as a live output of every action taken. It routes exceptions to the right person with the right context. It does not act autonomously where the business has decided it should not.
This version of AI does not look like the AI tools most businesses have encountered so far. It does not ask to be trusted. It provides the evidence from which trust is earned — decision by decision, action by action, audit by audit.
Building this version of AI is harder than building a productivity assistant. It requires deep integration with the ERP’s data model, not a surface-level API connection. It requires the governance layer to be part of the design, not an afterthought. And it requires a company that understands both what the ERP can do and what the AI must not do without permission.
Why this matters now
Microsoft’s announcement of Project MIA at DynamicsMinds in May 2026 is a useful signal, even if its first wave is aimed at Dynamics 365 Finance and Supply Chain Management rather than Business Central. The direction of travel is clear: AI-assisted implementation is arriving at the centre of the Dynamics ecosystem, and the pace is accelerating. BC will not be exempt from that trajectory.
That acceleration makes the governance question more urgent, not less. The faster AI moves into ERP environments, the more important it becomes that the AI arriving there was designed to be auditable, governed, and safe to trust with the system of record.
The companies that will benefit most from AI in Business Central are not necessarily the ones who move fastest. They are the ones who move with the right foundation — data that is clean and connected, processes that are documented, and AI that operates within guardrails the business has defined and the board can see.
That is the standard that will matter. Not AI as an add-on to BC. AI as a native participant in what BC does — governed, auditable, and designed for the operational realities of mid-market businesses. The companies that get there first, and get there correctly, will be the ones that the governance conversation eventually lands on.
The conversation about AI in ERP is just beginning. The question is not whether to have it. The question is whether to have it on the right terms.
Let’s chat further.
"*" indicates required fields
