PROOFCONSULTPEOPLE · EVIDENCE · IMPACTDecision Review
ENNL
Menu

Approach

Understand first. Then intervene.

A well-intentioned intervention can address the wrong problem. We examine not only visible events but the conditions, mechanisms and competing explanations behind them.

Four sources of information

We draw on research, organisational information, professional experience and what people involved can tell us. Sometimes these sources disagree. We then look for observations that could distinguish between possible explanations.

Law, professional authority, care and fairness set limits on the options from the outset.

1. Mechanism first

Labels such as “resistance to change” or “lack of ownership” describe an outcome and explain nothing. We ask what mechanism would produce the pattern: who gains or loses by acting, what people can see, what they are rewarded for, which coalition forms around which signal. An explanation helps identify a change worth trying. A label alone does not.

2. Select a few cues

Collecting more data does not help if it cannot distinguish between competing explanations. A cue is useful only if it would look different under different explanations. We deliberately choose a small number, often those that can be observed within days, and say what we did not measure and why.

3. Compare rivals, do not defend a favourite

Often, more than one explanation fits what we know. We write them down in their strongest form, then ask what each predicts. The aim is not to be right early; it is to find out cheaply which one is wrong.

Simple rules under uncertainty

When data are thin and the situation is messy, a long model often predicts no better than a short list of good cues. The research on fast and frugal heuristics, associated with Gerd Gigerenzer, makes this point: under real uncertainty, simple rules that ignore most information can match or beat elaborate analysis, and they are easier to check and to argue with.

Our version is a plain procedure. Rank the cues by how well each would separate the rival explanations. Look at the best one first. If it separates them, stop. If it does not, take the next. The price is honesty about what we skipped, so every Brief says what we did not look at and why. A fast rule can miss an interaction between two causes; we say when that risk is large enough to justify a second look.

What counts as credible

  • Strongest: records made before anyone asked the question, such as timestamps, logs and who attended what.
  • Good: direct observation of behaviour, and accounts from people with nothing to gain.
  • Useful but weighed: accounts from people with a stake, read together with what they stand to gain or lose.
  • Nearly worthless as evidence: seniority, confidence, and a story that has been repeated until it sounds like fact.

4. Weigh downside, reversibility and time

Two decisions with the same expected benefit are not equal if one can be undone next month and the other cannot. The more irreversible and costly the move, the more evidence we require before it; the more reversible, the earlier we try it and watch. Timing matters too: some uncertainty will resolve by itself if you wait, and some will only be resolved by acting.

Example: uncertainty and what would change the decision

  1. Situation (fictional)A manager considers replacing a team lead after a quarter of missed targets.
  2. Rival mechanismsThe lead lacks skill; or the lead was handed a role without decision rights; or the targets were set on assumptions that changed.
  3. UncertaintyWe cannot tell yet. Replacement is hard to reverse and affects a person’s career.
  4. What would change the decisionIf a short review of decision rights shows the lead cannot approve what the targets require, the move becomes a role fix, not a person swap. If the team gets the rights and still misses, the skill hypothesis gains weight.

A technique: break the plan before it breaks you

Gather the people who will carry out a plan and ask one question: how would we guarantee this fails? Give them ten minutes and ask for as many routes as they can find. Nobody is attacking anybody; everyone is allowed to be clever.

Then do two things. Turn each route to failure into a safeguard. And ask which routes are already under way. In a fictional rollout of a new scheduling rule, the list included “announce it on a Friday”, “let the people who lose a shift hear it second-hand” and “measure only the rule’s compliance, not the missed shifts”. Two of the three were already planned. People who say nothing in a meeting will often write a very good list. That is useful information about the meeting.

What to stop looking for

  • The single culprit. Most patterns have several contributing causes and a person to blame is the least useful of them.
  • A cue that cannot change the decision, however interesting it is.
  • The average opinion of a group, when the disagreement is where the information is.
  • Certainty before acting, when the cost of waiting exceeds the cost of a reversible mistake.

Where waiting costs

Waiting is a decision too, and it has a price. It is cheap when uncertainty will resolve by itself and nobody is harmed in the meantime. It is expensive when a small reversible step would reveal more than another month of discussion, when people bear a daily cost while we deliberate, or when a window closes. We state the cost of delay in the Brief next to the cost of error, so nobody can treat caution as free.

5. Review and revise

When making a decision, we also agree when to look back. What do we expect to see by then? What would suggest our explanation is wrong? Writing this down beforehand helps us learn from what actually happens.

  1. PredictState expectation and date.
  2. ObserveCollect the few cues, whatever they say.
  3. CompareCompare plainly, including misses.
  4. UpdateChange a belief or a threshold, in writing, and only then.

When something does not work, we may need to revise our approach. Read more about this in the Decision Review and on the CEIS page.