✓ AI-driven. Human-focused. ✓ Secure knowledge without a new system ✓ AFAS · Exact · Microsoft 365 integration ✓ Sources visible with every important answer ✓ From signal to owner and action ✓ Get it done without muddling through.
✓ Fixed price, no hourly billing ✓ Up and running in two weeks ✓ For Dutch SMBs
Bedrijfsgeheugen
SolutionsPlatformPricesKnowledgeAbout usMoreLog inSign up

AI in data engineering for SMEs: what makes sense, what doesn't.

A model writes a transformation in minutes that would take someone a day. That’s true. It also quietly writes incorrect SQL that looks perfectly fine. Here’s where the boundary lies, and which application is best to start with.

Data & BI

Last checked: August 16, 2026

What is worthwhile

Ranked by how much it delivers compared with how much can go wrong.

Explain what's already in place

The best first application, and almost nobody starts here. Have a model read the script nobody understands anymore and write an explanation of it. The risk is zero: you change nothing; you make visible what’s already there.

Come up with controls

Can this field be left empty, what values belong in it, and how many rows are normally added each day? Models are good at this, and people rarely think of it. This catches the failures you only notice three weeks later.

Convert Excel logic to SQL

That one spreadsheet with forty nested formulas the business runs on. A model reads it and turns it into readable, testable logic.

Mapping an unfamiliar system

When integrating with a system you don’t know, having someone explain the column names and relationships before you start can save days.

What’s pointless

Run pipelines autonomously

A model that’s allowed to intervene on its own when something changes sounds appealing, but it isn’t. Resolving schema drift without someone keeping an eye on it means you only discover errors once they’ve affected your figures.

Trusting generated SQL without checking it

This is the most dangerous one. Generated SQL always looks right. You only notice a join that duplicates rows or a slightly incorrect filter when the figures have already been off for months.

Data quality by feel

“Ask the model whether the data is correct” doesn’t work. Quality is an agreement about what counts as good, not an opinion. You have to make that agreement yourself; then a model can test against it.

Replace orchestration

What runs when, what waits on what, and what happens when something goes wrong — that belongs in an orchestrator, not in a prompt.

The disadvantages

Silent errors

The biggest risk in this whole field. You notice code that fails immediately. You never notice code that calculates just a little differently than intended unless you test for it.

Review burden

Writing faster means more to check. If no one has time to review it, you're shifting the problem instead of solving it.

Access to production

A model that can see your real data is another place that data passes through. For most work, that isn’t necessary: the schema and column names are enough.

Dependence on something no one understands

Exactly the problem you wanted to get rid of, in a new form. So always have documentation generated along with the code.

What’s possible for you?

Five questions about how your data currently moves. Then you see which application fits, with which technique, and what the first quick step is.

Frequently asked questions

Will AI replace our data engineer?

No. It makes writing faster and checking more important. The work shifts from typing to reviewing, which requires more knowledge, not less.

Can a model document our existing connections?

Yes, and that’s the application that delivers the most for the least risk. You don’t change anything; you make what’s already there visible.

Does our production data need to go to a model?

For most work, no. Code, schemas and column names usually don't contain personal data. Use made-up example rows when the model needs to understand something.

We don’t have any pipelines, just exports. Is this still for us?

Then this is the perfect time. Manual exports are the easiest thing to replace, and you're not yet locked into choices someone made years ago.

The first step

Don't let a model build something new, but explain what's already running. Take the script you're most afraid of and have it explain line by line. You discover things you didn't know, and it costs you an hour.

Schedule a call

Read more: AI adoption in SMEs — integrating systems in SMEs — AI in business intelligence.

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