AI in business intelligence for SMEs: what works, and what doesn't.

Almost everything you read about this comes from a supplier. Here’s the honest version: where AI in BI really delivers, where it costs money without results, and which step is best to take first.

Data & BI

Last checked: August 16, 2026

What does work

Four applications that prove themselves in SMEs. They have something in common: you can check the answer yourself.

Report what's different

Not you opening the dashboard, but the dashboard sending you a message when a number falls outside the range. The biggest benefit, the lowest risk, and you don't need a large model for it.

Categorize free text

Emails, tickets, complaints and quote requests are the largest untapped data source in SMBs. A model can reliably categorize them by subject, urgency and tone. You can see the text and its categorization side by side, so errors stand out.

Ask questions in plain language

Works well—but only on top of a model where the definitions are already established. It then translates your question into a query against that model instead of inventing the logic.

Explain what's on a dashboard

A two-sentence summary below a chart, for people who don’t read the chart. Small, but it increases the use of reports more than a prettier dashboard does.

What doesn’t work

These are the four I see fail most often, usually after someone has already paid.

AI on a messy data model

This is the most important one. AI fed with messy data produces convincing mess: the answer sounds good, is full of figures, and is wrong. Nobody notices, because there’s no source to check it against. No tool can solve this.

“Just ask anything” without definitions

If revenue per department means something different to each person, the model gives a different answer to the same question every time. Not because it's bad, but because the question has no single clear answer.

Predicting with too little history

At SME volumes, the margin of error is often greater than the effect you’re trying to influence. A forecast that’s off by ten percent is useless if your margin is eight percent.

AI instead of someone who understands the numbers

A model answers the question you ask. It doesn’t tell you that you’re asking the wrong question. That still takes people.

The disadvantages rarely mentioned

Not to scare you off. To arrange them in advance rather than afterward. How to set up that model correctly first is shown at Power BI for SMEs.

Confidently wrong numbers

A language model rarely says, “I don’t know.” It gives you a number. Without a traceable source, you can’t tell the difference between a reliable answer and a made-up one.

No record of how a figure was arrived at

With a dashboard, you can drill down to the source. You can’t do that with an AI answer in a chat window. That’s a problem for figures sent to the bank or an accountant.

Where your numbers go

Asking a question means sending along your data. With a public chat interface, it leaves your environment. With Copilot inside your own tenant, it doesn’t. You need to look into that difference and document it.

Cost per inquiry

Every question costs money. You won’t notice it with a handful of users. You will with a dashboard that generates a summary a hundred times a day.

What’s possible for you?

Five questions. Then you'll see which application suits your situation, what technology can make it work, and what the first quick win is. You don't have to fill anything in.

Frequently asked questions

Do I need a data warehouse for AI in BI?

Not for everything. Flagging exceptions and categorizing free text can be done without one. You don't want to ask questions in plain language without a model that has the definitions set out.

Can Power BI Copilot do this already?

To some extent. Copilot works with your semantic model, so everything depends on how good that model is. If the model is messy, Copilot gives messy answers — neatly formatted.

Is it safe to send our figures to an AI model?

That depends on which figures and which model. Within your own Microsoft environment, the data stays with you. Not so with a public chat service. Document which data must not leave which environment.

About how much does this cost?

The technology is usually the cheapest part. The work lies in cleaning up definitions and sources. The business scan will show what that means for you.

The first step is almost never AI

Choose the number that causes the most arguments. Document on one A4 sheet exactly what it means and where it comes from. That takes an afternoon, costs nothing, and makes every next step easier.

Schedule a call

Ask this site

Ask your question about AI, the AI Act, integrations or digitalization. The answer comes from the pages of this site, with references to where it can be found.

The way it works now

The appointment is in one colleague's mailbox
Data is manually retyped from one system to another
The person who knows is on vacation
No one can see afterwards why this decision was made

As it is

The agreement is stored with the client, findable for everyone
Once entered, information flows through to the rest
Work continues, even if someone is absent
Every decision has a source, an owner, and a date

drag with your finger

What if Peter leaves? Two out of five rules on this page disappear. That's how it feels when knowledge isn't part of the company.

What is your next step?

Free self-assessmentTwelve click questions, three minutes, immediate score and your biggest risk. Fresh LookBusiness case, benchmark and priority. Settled once we build after this. What it costsFixed price per package, no price per user, no hourly billing. From the fieldWhat changed at companies that started with this.

Also good to know: frequently asked questions · security and data · terms and conditions · partners · research and figures · templates · what's changing · where you start

Ready for the next step?

Move from explanation to a concrete next step for your organization.

View the next step