BusinessGPSOwn the capacity to change.

Layer 4: Judgement Intelligence

Articulate the judgement that protects your profit before you automate around it

v1.0 · 14 November 2026 · First issued 14 November 2026 · Custodian: BusinessGPS · Version index and changelog

Part of your profit depends on decisions only a few experienced people know how to make. Write that judgement down before AI takes over the work around it, or you automate the task and lose what protected the result.

Which decisions would become slower, more expensive or less reliable if your most experienced people were unavailable tomorrow?

The system would still run. The procedures would still exist. But someone would still need to spot the unusual transaction, challenge the delivery promise or stop a sensible-looking decision from causing a larger loss.

If that judgement sits with a handful of people, part of your profit depends on a capability the business has not secured. Your next AI investment is about to automate the work around it.

So the order matters. Articulate the judgement first, then automate. To articulate a judgement is to write down what the expert notices, weighs and decides, in words someone else can test.

Your experts are both your advantage and your limit on growth

A business often wins because its people know what competitors miss. They recognise what a customer really needs, see a supplier problem coming or know when a standard response will fail.

That expertise is worth protecting. But if every difficult case goes back to the same 2 people, their diary becomes a limit on growth. If the business expands, it can add decisions faster than it develops people able to make them.

So you need a deliberate choice about each kind of decision:

  • what to teach
  • what to support with better evidence
  • what must stay with a qualified expert

The aim is reliable capability across the business, with clear limits on who may decide.

Waiting for a resignation, a retirement or a reorganisation turns a manageable dependency into an urgent problem. The time to test cover is while the expert is still there to challenge it.

Your finance director sees the same risk as cost: one wrong credit release can wipe out a saving measured in minutes. Your programme director sees it at handover, when only the programme team can handle the hard cases.

Managerial Judgment is the judgement behind the action

We place this work in the Five-Layer Activity Model, our map of where work sits, from standard software to AI agents. Layer 3, Contextual Workarounds, covers the workarounds and manual activity. Layer 4, Managerial Judgment, covers noticing what matters, reading incomplete evidence, weighing the options and knowing when to stop or ask.

Take a hypothetical planner who changes a delivery promise. The system records the change. A spreadsheet may show the sums behind it. Neither explains why the planner distrusted the supplier’s date, or judged one customer commitment more important than another.

The planner may have spotted a real warning. They may also be applying an assumption that is out of date. Both need checking.

Experience, confidence and seniority are not enough to show that judgement is sound. Kahneman and Klein’s research links credible expert intuition to settings with learnable patterns and real feedback.[1]

So keep 3 questions apart:

  1. How was the decision made?
  2. Was the approach sound?
  3. Who is authorised to decide?

An expert’s explanation answers only part of the first question. It settles neither of the others.

Adding AI does not automatically improve the decision

A system that recalls an executive’s preferences, or writes in their style, has not shown that it can exercise their judgement. Agreement can feel useful while a weak assumption goes unchallenged.

A 2024 meta-analysis of 106 experiments compared people, AI and the two working together.[2] On average, the combinations beat people working alone. But they did worse than the better of the two working alone, whether that was the person or the AI.

That finding does not predict your programme’s result. It shows why the combination needs testing before it is trusted.

So compare the proposed arrangement with how the decision is made today, on cases that were not used to build it. Look at serious errors, uncertainty, escalation and business outcomes, as well as speed. Test whether the person can challenge the recommendation, not merely approve it.

Invest in decision support that proves its value with relevant evidence, real challenge and clear approval limits. A familiar conversational style proves nothing.

Our paper, Layer 4: Judgement Intelligence, sets out the evidence in full. It also sets out how we design decision support to meet that test.

People share their judgement only when they trust what happens next

The knowledge a programme most needs is often the hardest to say out loud. It might be an unofficial exception, a doubt about a rule, a decision that went wrong or a disagreement with a senior colleague.

People need a credible reason to share it. If they think an interview is a disguised performance review, you may get a polished account that leaves out the real decision.

Transformation Capability Management, the proposed standard we hold as custodian, names this in its sixth domain: social licence and absorption trust. Leaders need to explain four things:

  • what the work is for
  • who will see what people say
  • how people can correct it
  • what automation may mean for their roles

They must also be honest about consequences they cannot promise away.

Trust makes candid accounts more likely. It does not make every account correct. So rebuild actual cases and check what evidence was available at the time. Ask a second expert. Then test a similar case where the same answer would be wrong. The critical decision method and cognitive task analysis are established ways to do this.[3]

The commercial risk is direct. If people hold back the judgement the new model needs, you invest in a convincing but incomplete copy of your own capability.

Four questions test one decision before the next investment

Choose a recurring decision that moves margin, cash, delivery or control. Bring together its owner, an experienced practitioner and the person expected to provide cover.

  1. What does the expert notice that the system or procedure leaves out?
  2. What is the consequence of getting it wrong, and what evidence supports that view?
  3. Can another person reach a defensible decision on an unfamiliar case?
  4. What should be taught, supported, redesigned or kept under human authority?

Use the answers to set the next investment. A good result might be better training, better data or a clearer escalation rule. It might justify bounded AI support. Let the evidence decide which.

Before you automate a decision, see how far the decisions behind your change are written down.

The Baseline grades the first 9 of our 27 steps from your programme’s documents, including whether they record how competing goals were traded off.

Applying costs nothing. The fee is fixed and printed on the Baseline page. Cancel within 30 days of paying and you get the fee back in full, and you keep the report. The page also gives you the sentence to take to your board.

References

[1] Kahneman, D. and Klein, G. (2009). Conditions for Intuitive Expertise: A Failure to Disagree. American Psychologist, 64(6), 515 to 526. https://doi.org/10.1037/a0016755

[2] Vaccaro, M., Almaatouq, A. and Malone, T. W. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293 to 2303. https://doi.org/10.1038/s41562-024-02024-1

[3] Klein, G., Calderwood, R. and Macgregor, D. (1989). Critical Decision Method for Eliciting Knowledge. IEEE Transactions on Systems, Man, and Cybernetics, 19(3), 462 to 472. https://doi.org/10.1109/21.31053

The research behind this post

Layer 4: Judgement Intelligence — Articulating Judgement Before Automation

Version index and changelog

  • v1.0 (14 November 2026): first issued.