AI connected to your ERP

An AI that cannot read your ERP knows nothing about your business. We build that connection, with rights, roles and performance respected.

In short

Ask any AI tool what the lead time is on item 4471 for customer Vandenbulcke, and you get a friendly answer that means nothing. The model does not know what your item numbers stand for, cannot see your price agreements and has no idea about your stock.

That is not a shortcoming of the technology, it is a question of access. The moment an AI agent can query your ERP, the nature of what it can do changes. It stops answering on the basis of what sounds plausible and starts answering on the basis of what is in your system.

Business Central: standard and custom

Microsoft now ships a number of agents inside Business Central itself, among others for incoming invoices and sales orders, plus a design environment for building simple ones. There is also a layer that lets external AI applications talk to Business Central securely.

We are not arguing against that. What comes as standard, use as standard. We come in where Microsoft stops: your own pricing and discount logic, connections to systems outside the Microsoft landscape, processes that span several packages, and everything your business does differently from the rest of the market.

What you need to start

An ERP with an API, or a version for which one is available.

Someone on your side who knows how the data in the system is structured.

One well-scoped first application. Not: open everything up. But: this one process.

A decision on where your data may be processed. More on that.

WHAT CHANGES ONCE AI CAN READ YOUR ERP

A customer asks through the webshop where their order is and gets the real answer, not a referral to customer service.
A salesperson asks which customers are ordering less than last year and gets the list, without anyone building a report.
An incoming PDF purchase order is read, matched to the right customer and item numbers, and set up ready as an order.
An incoming invoice is checked against the purchase order and the delivery, and the discrepancies float to the top.
A colleague asks for an overview in plain language and gets a table or a chart instead of an export to sift through.

THE PACKAGES WE WORK WITH

Our clients run on widely varying systems. We have been connecting webshops, portals and applications to them for years:

For an AI agent, that story does not fundamentally change. It is simply a new user of the same integration. It just asks more questions than a human, and faster.

HOW WE SET UP AN INTEGRATION

Inventory. Which data does the agent need, and which data must it never see. The second is as important as the first.
Access layer. We build a layer that decides what the agent may read and write. The model never talks to your database directly.
Rights and roles. The agent inherits the rights of the user who addresses it.
Read before write. Almost every project starts read-only. Write access comes once the results check out.
Audit log. Every query and every action is recorded. In a dispute you can trace exactly what happened and on what basis.

GOOD TO KNOW

An AI agent allowed to write straight into your ERP with no access layer and no log is not an efficiency gain, it is an open risk.

We build that layer as standard. Not because it looks good in a quote, but because otherwise you cannot show what went wrong at the first incident.

YEARS OF .NET EXPERIENCE

1,000+

PROJECTS DELIVERED

100%

SUCCESSFUL ERP INTEGRATIONS

IN-HOUSE SPECIALISTS

Keeping speed and cost under control

An agent that queries your ERP ten times per question makes your system slow and your bill high. So we build in caching for data that does not change by the second, and we limit up front which queries the agent may run. That is ordinary sound engineering, and it is the difference between a pilot and something still running a year later.

Why companies choose IDcreation

The Belgian Defence has been our largest client for over ten years. What meets those security requirements will hold up with you.

No standard package you have to bend to, but software that follows the way you work.

Our own hosting in Belgium, with monitoring, backups and a single point of contact.

Since 1996 we have built in Microsoft technology. One strong foundation makes us fast and thorough.

Frequently Asked Questions

Does this work if our ERP is an older package?

Often yes. If an API, a web service or even a decent export option exists, we can work with it. In thirty years we have integrated plenty of systems that were officially out of support.

Can the AI agent create orders in our ERP itself?

Technically it can. Whether it should depends on the process. We almost always start read-only and grant write access only once the agent's proposals demonstrably check out.

Does the AI then see all our company data?

No. We determine up front which tables and fields are reachable, and the agent inherits the rights of the user addressing it.

What is an MCP server?

It is a standardised way to give an AI application secure access to a system. Instead of building a separate integration for every tool, the AI talks to your systems through one agreed interface.

Will this slow down our ERP?

Not if it is built properly. We limit the number of queries, cache what does not change constantly and measure the load from day one.

Do we have to clean up our data first?

Rarely in full. We look at which clean-up is genuinely needed for the chosen application.

Want to know what AI can do with your ERP data?

Tell us which package you use and which process takes the most manual work. We will look at feasibility and sketch an approach.