Foundation — one question: why do margins diverge from what the contract promises?

The owner asked one question. Every month the realised margin diverged from what the contract had promised. Sometimes a few percent, sometimes tens of percent. Dashboards showed the outcome — never the cause.

The question sounded simple. In practice it touched everything: contract terms, invoicing rules, attendance and extras, purchase prices, tariff indexation, discount clauses, and the many places where manual adjustments happened. The answer was scattered across at least five source systems and dozens of people.

What we tried and what did not work

First dashboards. Every KPI was correct in isolation, but no dashboard could answer the "why". Dashboards show what happens. They never explain why.

Then data analysts with SQL. They could reconstruct per contract, but needed weeks per period. As soon as a new period started, the work restarted by hand.

Then a BI tool with "AI insights". It generated plausible explanations — without verification. Every statement sounded good and no one could prove it was true.

The real lesson

The question could not be answered without simultaneously checking the full chain: contract → invoicing → attendance → purchasing → indexation → discount. Humanly impossible within an acceptable turnaround. Mechanically possible — provided the machine asks the question the way a controller would, not the way a model would guess.

That became the foundation: a system that works the way a controller thinks, not the way a model writes. Every later choice — the agents, the thinking model, the sovereign decision layer — comes back to this one question.

THE LESSON A good question is worth more than ten dashboards. The whole platform was later built on exactly this foundation: one concrete question that touches the full value chain.