Guidance

What we have learned putting AI to work inside real companies — choosing use cases, getting teams to adopt them, keeping the output reliable, and the Belgian rules that decide when you can go live.

Strategy and use cases

Choosing what to automate, tying it to KPIs, and proving it paid off.

• How to align AI use cases with company KPIsStart from a KPI you are already accountable for, not from a tool you saw demoed.
• How to spot AI use cases and tie them to management KPIsWatch for repeated work with a judgement step. Then walk it up to a KPI a director already reports, or it will not get funded.
• When to choose Claude and when to choose MistralBoth are serious choices. Decide on three things — how hard your hardest task is, where your data may go, and what the work costs at your volume — with the current figures side by side.

Data and determinism

Getting reliable, repeatable answers out of AI over your own data.

• How to test AI output before you put it in front of a customerBuild a small set of questions with known-correct answers, run it on every change, and measure the failures you actually care about.
• How to write a skill that does not invent thingsHallucination is mostly a design problem, not a model problem. Remove the situations where inventing is the only option left.
• Making AI skills deterministic in data applicationsMove every step that must be repeatable out of the model and into code the model calls.
• Use examples, not adjectives, to get the output you wantTwo or three real examples beat a paragraph of instructions. Show the edge cases and the refusals, not just the ideal answer.
• Write it down once: the context files that stop generic outputThe most common complaint about AI writing is that it could be about any company. The fix is four short documents everyone's prompts read from.

Governance and compliance

Policy, oversight, the EU AI Act, and what to write down.

• Belgium: you must tell the works council three months beforeFrom 50 staff, CAO 39 requires written information and consultation three months before you introduce technology with collective impact.
• The AI wrote it in English. In Flanders, that document is void.Employer-employee documents must be in the language of the region. A generated document in the wrong language is legally deemed never to have existed.

Rollout and ways of working

Getting teams to actually use it, and keeping it running.

• A better model will not fix a workflow you never wrote downA new model gives a better answer to the same question. It does not change the question, or what happens to the answer afterwards.
• E-invoicing is already mandatory. Use it as your first automation.Since January 2026 Belgian B2B invoices must be structured Peppol documents. That gives you machine-readable data you did not have before.
• Getting your team actually using AI, as a team leaderEnthusiasm decays in about three weeks. What survives is one workflow, one owner, and a habit of showing each other the output.
• Staying critical of output that sounds rightFluent writing suppresses scrutiny. Build habits that make checking automatic, because vigilance on its own does not survive week three.
• Winning time on lead generation without a GDPR problemAutomate the preparation, not the contact. The time is in research and drafting; the risk is in collecting and storing.

Frequently asked questions

The questions team leaders and owners ask us first.

• We want to start with AI — what do we do first?Pick one decision that a named person already owns, and improve it end to end.