The Decision Brief

Decision Brief · structure

1. 1. The decisionWhat is being decided, by whom, by when, and what would count as good.
2. 2. Observed realityWhat can be observed, with sources, kept apart from interpretation.
3. 3. Rival hypothesesTwo or three serious explanations, each open to being wrong, none a straw man.
4. 4. Discriminating cuesThe few signals that would tell the hypotheses apart, and how to get them cheaply.
5. 5. Decision thresholdHow strong the evidence must be before acting, given the cost of being wrong in each direction.
6. 6. ActionThe recommended step, who does what, and the first visible behaviour.
7. 7. UncertaintyWhat remains unknown, and how strong the evidence behind each statement is.
8. 8. Reversal conditionsWhat would make us change course, agreed before emotions are invested.
9. 9. PredictionsExplicit, dated expectations that reality can contradict.
10. 10. Outcome reviewWhen and how predictions are compared with outcomes, and by whom.
11. 11. EthicsWho is affected, who has not been heard, decision rights, consent and conflicts of interest.
12. 12. Complexity budgetHow much machinery the decision deserves: what we deliberately did not model or measure.

Fictional Brief · a 40-person accounting firm

1. 1. DecisionThe managing partner must decide within five weeks whether every client filing needs a mandatory second review by a partner. Errors found after filing rose from about 2 to about 6 per 100 filings over a year (invented figures). Good outcome: the error rate returns near the old level without halving the firm’s throughput.
2. 2. Observed realityMost errors are small, a few are costly. Errors cluster in March to May. Partners say they review “everything important”. The query log, where juniors ask questions before filing, shows far fewer entries than two years ago. Nobody has examined why.
3. 3. Rival hypothesesA: seniors are overloaded at peak, so reviews are shallow. B: juniors have learned that questions are met with irritation, so doubts are not raised and uncertain work is filed as if it were certain. Both are plausible and both fit what is known.
4. 4. Discriminating cuesTake the twenty most recent errors. If A: they concentrate in peak weeks and in files where the reviewer spent least time, and juniors’ query rates are unchanged. If B: they concentrate in files where a junior was unsure at the time, and query rates fell exactly when a partner last reacted sharply. Add a five-question anonymous probe: “Last time you were unsure, did you ask? What happened?” About one week of work.
5. 5. Decision thresholdA blanket second review costs partner hours and is hard to withdraw once clients expect it. It needs stronger evidence than a six-week change in how questions are received. We act on the cheap, reversible step with moderate evidence; the costly one waits for strong evidence.
6. 6. ActionWhatever the finding: a second review for the three filing types with the costliest errors. If B is supported: the managing partner commits to answer any pre-filing query within a day and without comment on the question. If A is supported: temporary capacity in peak weeks. First visible behaviour: queries logged again.
7. 7. UncertaintyTwenty files is a small sample. Errors are found after filing, so the data hides errors not yet found. Evidence on A and B is moderate at best. The probe only reaches juniors willing to answer.
8. 8. Reversal conditionsIf B is supported but queries and errors do not move within eight weeks, B is wrong or insufficient, and the blanket second review gets another hearing. If errors are spread evenly across the year and across files, neither A nor B is the main cause, and client mix is next.
9. 9. PredictionsIf B is right: the query log doubles within six weeks, and errors in junior-prepared files fall by a third within the next filing season. If neither happens, B is wrong.
10. 10. Outcome reviewQuery rates at 8 weeks. Error rates at the end of the next peak season, compared with last year’s. Done by the managing partner with someone from outside the partner group, so the review is not marked by the people who made the change.
11. 11. EthicsWho loses? Juniors, if the findings are used to blame them. The query log and probe are used for the review only, with consent, and no individual is named. Partners are asked how they react to questions, and that feedback has to be survivable for them as well. Whoever raised the doubts about overload and about culture should not be penalised. A conflict of interest: the managing partner commissioned the review of a team they lead.
12. 12. Complexity budgetOne spreadsheet, twenty files, one probe. No dashboard, no psychometric instrument, no model of individual juniors. If the first cheap step explains most of the variance, we stop there.