Can AI learn how your manager decides?
Partly. AI can find patterns in past decisions — which orders were accepted, which quotes were discounted — but patterns are not reasons. It learns what usually happened, not why, and it misses the exceptions that were decided by phone or in a meeting. Behaviour data is useful to show where rules exist. The rules, their exceptions and their limits have to be confirmed with the person who decides, with a source. Whether AI should then act on them is a separate decision that stays with people.
A short story
A distributor let a model learn from years of credit decisions. It picked up that orders from one group of customers were almost always approved without a credit check, and started recommending exactly that.
What the data did not show: the finance manager approved those orders because a parent company guaranteed them — and that guarantee ended last year. The pattern was right about the past and wrong about today. The finance manager could have explained it in one sentence; nobody had asked. (An illustrative example, not a client case.)
What can AI learn from behaviour — and what not?
| Question | From behaviour data | Needs the person who decides |
|---|---|---|
| What was decided? | Yes, if the decision was recorded | — |
| Why was it decided? | No — only correlations | Yes: the rule and the reason |
| Rare exceptions | Weak — too few examples | Yes: when the rule does not apply |
| Decisions outside the system | No | Yes: calls, meetings, verbal agreements |
| Has the context changed? | No — it assumes the past continues | Yes: what is no longer true |
| A new situation | A guess from similar cases | The reasoning to apply |
What goes wrong when AI learns decisions only from behaviour?
- It repeats habits: including ones the manager already stopped
- It treats exceptions as noise: while the exceptions are where the judgement was
- It cannot explain itself: so nobody can check an answer before acting on it
- It does not notice change: and keeps deciding as before, with the same confidence
Should AI act on those decisions?
Decide that per decision, not per tool. A simple boundary helps: AI may advise, AI may prepare a decision for review, or AI never acts. Decisions with an impact on customers, money or safety usually belong in the first two.
Whatever the boundary, people need to see where an answer comes from. AI you can check explains why sourced answers matter more than confident ones.
Is learning from behaviour useless, then?
Not at all. Patterns in past decisions are a good way to find where the rules are, which cases are unusual, and where two people decide the same situation differently. Use them to prepare the conversation with the person who decides — not to replace it.
The combination is stronger than either part: the data shows what happened, the person explains why, and the captured rule carries both, with its source.
How MentX helps
MentX captures how your company really works: the knowledge of your key people and the processes they run. We offer both a hands-on service approach — mapping your processes and capturing the knowledge behind them, from meetings, workshops and a guided intake — and our own personality-aware knowledge base, with full source attribution, in your own Knowledge Space. We build on the tools you already use and add best-practice solutions where they fit.
We capture the reasoning with the person who decides — in conversations, and with real cases — and attach the source to every rule. Behaviour data is welcome as a starting point: it shows where to ask. DigiME is the first use case built on the MentX knowledge base: a digital knowledge copy of your key person you can converse with. It answers in his knowledge, his reasoning style and his voice of thought. It advises with sources; people decide. See decision support.
What to do this quarter
- Pick one recurring decision that depends on one manager.
- Collect a handful of recent cases, including an unusual one.
- Ask the manager to explain each: the rule, the exception, and what would change the answer.
- Write the rule down, with the date and the source.
- Decide for that decision whether AI may advise, prepare for review, or never act.
