By Rajib Lochan Huzuri
Recently I sat down with the finance director of a mid-sized supply chain company. They’d been live on Business Central for over a year, one of those implementations that had actually gone well by ERP standards. They’d closed a full financial year in the system, the warehouse mobile app was working, integrations with their partners were good, and purchase orders moved through smoothly. Users had settled into the new processes.
Then, in the middle of the conversation, the finance director asked me something that had nothing to do with any of that.
“Can Business Central just tell me which of our key customers are going to cause a cash flow problem next month?”
My first reaction was automatic. I started walking him through how to get there — open Customer Ledger Entries, filter the open invoices, check the due dates, review payment terms, run the aged receivables report, maybe export it to Excel and compare it against historical payment behaviour.
I got about halfway through explaining it and realised I was giving him a lesson on Business Central when he was asking me a business question. He hadn’t asked which page to open. He was asking what was happening in his business, and what deserved his attention. I’d been answering that kind of question the same way for years.
That conversation stuck with me. It made me wonder whether the way we teach people to use ERP is about to change.
For Years, I Taught People How to Use ERP
Over the years I’ve trained hundreds of users, and the pattern is usually the same. We show them the pages, explain the filters, explain dimensions and posting groups, and eventually they become comfortable enough to get through their daily work.
Where’s the customer ledger? How do you filter this page? Which report do you run? How do dimensions work? What happens after you post a document, and where does the transaction go? For finance teams we explain posting groups and ledger entries. For warehouse teams, item availability and reservations. For sales teams, customers, items, prices, discounts, orders. That knowledge still matters, I don’t think it’s going anywhere. But I have never met a sales manager who wakes up in the morning thinking, “I hope I get to use the customer ledger today.” They want to know which orders are in trouble. Finance wants to know where the cash is stuck. Operations wants to know why the same items keep running short. The ERP is the tool. That’s not really what they care about.
Business Central holds a huge amount of the information that answers those questions. Getting to it has always meant knowing where to look. That’s the part starting to change.
The Finance Team That Didn’t Need Another Report
That question from the finance director reminded me of another customer, a few years back. Their collections team had a familiar problem of overdue receivables. They had reports, customer ledger entries, ageing analysis, dimensions, Excel, even a standing Monday morning collections meeting. And every Monday, someone still spent hours pulling it all together.
When we sat with the team, it became clear the problem wasn’t a lack of information, they had plenty. The problem was deciding what mattered. Nobody needed another report listing every overdue customer. What they actually wanted to know was “which five customers should I worry about today, and why?” Once we got into the detail, I realised they weren’t short of reports at all. They were short of time. Someone was still having to look at all those reports and decide which customers actually mattered that morning.
Picture an agent with the right access to Business Central data and other relevant business context. Instead of spitting out a list of overdue invoices, imagine the agent coming back and saying that one of the company’s larger customers has a sizeable overdue balance, several open orders sitting on credit limit check, and a recent dispute with the customer may explain why payment has slowed. That’s not the same thing as running an aged receivables report. It’s an interpretation, several pieces of information read together in context.
I want to be precise here, though — this isn’t a claim that Business Central has an out-of-the-box button labelled Predict My Cash-Flow Problems. It doesn’t. What I’m describing is the kind of scenario you can build once AI agents are connected to Business Central’s data and business logic, including through Microsoft’s Business Central MCP server, with the right configuration and permissions behind it.
Microsoft documents the Business Central MCP server as a way for AI clients to interact conversationally with Business Central data and capabilities. By default, access is read only to whatever API pages are exposed, anything beyond that — creating, modifying, or deleting records, or triggering business actions — needs explicit configuration.
There’s a real gap between what the technology makes possible and what a customer should assume already exists in the product today. As a consultant, I think that gap is worth being careful about.
The Stock Shortage That Wasn’t Really a Stock Shortage
One of the sharper lessons of my career came from a manufacturing customer with a serious looking problem. Business Central kept telling them they didn’t have enough stock of several key raw materials, while the warehouse team could physically point to pallets of the stuff sitting in the aisles. The operations manager’s opening request was simple: “we need a report to fix this.”
My first instinct was a planning parameter issue, so we went through the usual suspects — safety stock, lead times, reordering policies, demand, purchase orders. Two days in, we still hadn’t found a single configuration problem that explained the whole picture. It was only once we started watching what actually happened on the warehouse floor that things clicked.
Goods were being physically received, but some purchase receipts weren’t getting posted until a day or two later. Several vendor lead times hadn’t been touched in years. Salespeople occasionally changed promised delivery dates without checking availability at the right location. None of those, on its own, looked like a serious problem. Together, they painted a picture in Business Central that had drifted a long way from what was actually happening in the warehouse.
No report would have fixed that. And AI wouldn’t have fixed it either — an agent might have spotted the discrepancy faster than we did, but if the underlying data was wrong, it would just as easily have produced a confident, well-written explanation built on the wrong foundation. That’s something worth flagging to every customer who gets excited about AI and ERP together: an intelligent agent sitting on top of a poorly implemented ERP doesn’t create an intelligent business on its own. If anything, it exposes bad processes faster, and can make bad decisions easier to act on before anyone checks them. That experience is why I’m still cautious when someone tells me that AI will solve their ERP problems. Before we talk about the agent, I want to know whether the ERP is telling us the truth.
“I Don’t Want Another Dashboard”
Another conversation has stuck with me for a different reason. We’d spent weeks building dashboards for a managing director — Power BI, KPIs, charts, drill-downs, Role Centers, the works. When we presented it, he looked at the screen for a while and said something like “This is impressive. But I don’t want fifty things to look at. I want the three things happening today that could cost me money if I don’t act.”
I remember looking at him and thinking, “He’s right.” We’d spent weeks giving him more information when what he really wanted was less noise. We’d solved the reporting problem. We hadn’t solved the decision problem.
A dashboard gives you information. An agent can help you find the exceptions buried inside it. It’s a subtle difference, but it matters — instead of “show me today’s sales,” you ask why sales are below expectation this week. Instead of “list all overdue customers,” you ask which ones need attention today and what’s causing the delay. Instead of “show me inventory below safety stock,” you ask which shortages are actually likely to affect this week’s deliveries. That’s a different way of working with an ERP system.
Where Does MCP Actually Fit?
I’ve deliberately kept the technical side light so far, because MCP isn’t really the story here — the changing relationship between people and ERP is. But it’s worth explaining briefly, because it’s the plumbing underneath a lot of what I’ve described.
I don’t think most finance or operations managers need to know what MCP stands for. They need to know what it changes.
For me, the simplest way to think about it is Business Central remains underneath everything. MCP gives an AI application a more standard way of reaching the Business Central capabilities we’ve chosen to expose.
Microsoft now provides a Business Central MCP server, which means compatible AI clients — Copilot Studio, and other MCP-compatible tools — can connect to a Business Central environment and, depending on configuration, work with its data and business logic conversationally.
Business Central stays the system of record. The agent is just another way in. And that’s where “talking to your ERP” gets interesting, but I would still be very careful about what the agent is allowed to do. Some of the AI conversation out there gets a bit too optimistic on this point. This is where I would slow the customer down. Connecting an agent to an ERP isn’t the same as giving it free rein. Microsoft’s own documentation is clear: by default, the MCP server gives read-only access to whatever API pages are exposed. If you want an agent to create, modify, or delete data, or take a business action, that has to be explicitly configured and permissioned — nothing happens by default just because it would be convenient.
That’s exactly how we want integration to behave. Imagine someone asking an agent to “create a purchase order for this supplier.” Sounds harmless, until you start asking the obvious follow-ups. Which supplier? Which company? Which items, at what price, from which location? Is the supplier even an approved vendor? Does this person have the authority? Is it a capital purchase that needs a second level approval?
The technology being able to do something isn’t the same as it should. ERP has always been about controls, and AI doesn’t remove them — if anything, it forces you to think harder about exactly where they sit.
And this is where I think it gets more interesting. Business Central isn’t only moving toward answering questions. Microsoft is already introducing agents that actually perform specific jobs. Microsoft’s Sales Order Agent is a good example. It can monitor a mailbox, read customer requests, identify the customer, find the items they’re asking for, check availability, and prepare a sales quote, and, depending on configuration, carry that through to a confirmed sales order. A person stays involved at defined points, including reviewing and approving the communication.
That’s more interesting to me than a chatbot answering “what’s my inventory.” The difference is that the agent is actually participating in a business process — it has a task, access to specific information, defined permissions, and checkpoints where a person can step in. That’s less “chat with your ERP” and more “work alongside an ERP agent,” and I think that is going to matter more over time.
There’s Still a Lot of ERP Underneath the AI
I don’t want to suggest any of this makes traditional ERP knowledge less relevant. It’s the opposite. Anyone who’s spent real time implementing Business Central knows that a simple sounding business question can sit on top of a genuinely complicated process.
This is where my implementation experience makes me a little less enthusiastic than some of the AI conversations I see online. A question like “Can we deliver this order next Friday?” sounds simple to the customer. Anyone who has actually implemented Business Central knows it isn’t. The honest answer depends on current inventory, location, reservations, purchase orders, production orders, transfer orders, lead times, planning parameters, capable-to-promise logic, and warehouse processes, all at once. Microsoft’s Sales Order Agent, for instance, can draw on availability information and capable-to-promise functionality when it’s configured to, which is a good illustration of just how much ERP logic sits underneath what looks like a simple question.
That’s why I don’t think AI makes ERP consultants irrelevant, I think it does the opposite. A consultant still has to understand what the answer actually means. Somebody has to know whether the underlying data can be trusted. A human has to decide what an agent should and shouldn’t be allowed to do, understand the consequences of posting a transaction, and recognise when an answer just doesn’t make business sense. None of that goes away. If anything, it becomes the more important part of the job.
How My Advice to Customers Has Changed
Ten years ago, if a customer asked how to prepare for a Business Central implementation, I’d have started with — get users early to the system and train them, learn the navigation, understand the processes, know where the reports live, how a transaction moves through the system. I still give that advice.
This is probably the biggest change in my own thinking: what decisions do system users actually need to make every day? That’s a different starting point. Instead of designing around pages and reports, you start with intent — what does the sales manager genuinely need to know each morning? What’s the finance director spending two hours untangling every Monday? Where does the warehouse manager keep having to chase down the same exceptions? Which decisions are routine, and which need real judgment? Which of them could an agent help with, and which should always stay with a person? Those questions are becoming part of how you design an ERP implementation, not an afterthought once it’s live.
What I’d Tell a Company Starting a Business Central Project Today
When I sat down with a company beginning a Business Central implementation now, my advice looks a little different from a decade ago.
- Fix the process before you automate it. Don’t customize Business Central to preserve the habits of the old system — an agent doesn’t turn a bad process into a good one, it just runs the bad process faster.
- Take master data seriously, properly seriously. Item data, customer data, vendor data, lead times, dimensions — these have always mattered, but they matter more now. A person can often sense when something looks off. An agent won’t, unless you’ve deliberately built that context into the system.
- Design around decisions, not screens. Don’t start with “which page should we customize” — start with what the user actually needs to know, and what decision they’re trying to make.
- Take security seriously, early. An agent should never become a back door into ERP data. Permissions, identity, and segregation of duties still matter just as much as before. Microsoft’s MCP architecture authenticates through Microsoft Entra ID, and operations run under the user’s own identity and permissions, which supports proper auditability — that’s the kind of foundation I’d want underneath any enterprise implementation.
- Start with one real problem instead of trying to make the whole ERP conversational on day one. Pick something genuinely painful — overdue receivables, sales order exceptions, inventory shortages, vendor invoice processing — solve it properly, and learn from that before expanding.
- And keep people in the loop wherever the consequences are real. I’m comfortable with an agent flagging an exception, drafting a document, or preparing a response. I want to see how approval and accountability actually work before I’m comfortable with an agent posting a financially significant transaction without a person involved. The technology may eventually support more autonomy than that. Whether it should is a business decision, not a technical one.
What Does It Actually Mean to Be a Business Central User?
This is the question I keep coming back to. If a user can ask which customers are likely to create a collections problem this month and get a useful answer without knowing which page holds the customer ledger, or a sales manager can ask which orders are at risk of missing their delivery date without checking three different screens, something has shifted. Not necessarily in Business Central itself. It’s still the ledger, still the place transactions get recorded, still where the business rules and processes live. What can change is how much of that machinery a person has to touch directly to get an answer out of it.
I’ve been thinking about this quite a lot lately. If someone can ask Business Central a question in plain English and get to the answer without knowing which page contains it, are they still “using ERP” in the way we have traditionally understood it? I don’t know where this ends. And I’m not convinced anyone does yet.
After Years of Teaching People Where to Click
I’ve spent a large part of my career telling people how to use ERP software. These days I find myself asking a different question, not “which page should this person open,” but how much of Business Central someone actually needs to see to do their job well.
I don’t think the answer is “none.” ERP knowledge still matters. Good processes and good data still matter. Controls still matter. Experienced people still matter more, not less. But the way people interact with all of that is starting to shift. Maybe the future user doesn’t think “I need to open Business Central,” maybe they think “why did our margins drop,” or “which customers need my attention today,” or “what’s going to stop us shipping this week.” And maybe Business Central, working alongside an intelligent agent, is what gets them to that answer.
I don’t know if we’re heading toward a world where people rarely open Business Central. I doubt it will be that simple. But I do think we’re heading toward a world where users spend less time figuring out where the answer is and more time deciding what to do with it. And after spending most of my career teaching people where to click, I find that a pretty interesting change. So I’ll end with the question I’ve started asking customers myself:
“If you could ask your Business Central system anything, without knowing which page, report, or screen held the answer, what would you ask it first?”