A book in progress
The analysis is
not the product.
Every decision you are asked to analyze already has a rule running, and your analysis is competing with it. This is the working reference for the other job: find the rule that runs today, earn a challenger that beats it, and install one that still runs when you are not in the room.
Free and complete. Nothing on the method pages is gated.
The method
Three parts, and a study is what you get when one is missing
The first three questions produce the choice, the fourth the evidence, and the fifth the rule. Ordinary words, used precisely. They are never abbreviated into an acronym, because a method of this kind should be readable in a meeting without a legend.
01
Choice
The rule running today, who can change it and by when, and the challengers that could replace it. Once, or repeating.
When it is missingA missing choice looks like a tour: no named incumbent, no owner, no freeze date, a dashboard that can be admired.
02
Evidence
Only what could make the challenger lose. Not what would be interesting.
When it is missingA missing piece of evidence looks like a sales document: a pick with no killable reasons, or reasons that could not change the ranking if they were false.
03
Rule
What the operating process does now, who runs it, when it is revisited, and what would stop it.
When it is missingA missing rule looks like a good meeting: agreement in principle, and the incumbent still firing on Tuesday.
The premise
The failure is systematic, which is what makes it fixable
A vice president asks for a dashboard by Friday. A competent team extracts, explores, models, charts, and presents. Forty-seven slides later someone asks the only question that mattered at the beginning: what, exactly, are we supposed to do? The meeting agrees the work was thorough. The replenishment rule does not change.
On Monday the replenishment system ordered under the same rule as the week before, with the pad planners add by hand on top. The analysis explained last quarter and never proposed a rule to take that one’s place, so the incumbent won without anyone having to defend it.
Evidence assembled against no candidate ranking has no stopping rule, which is why exploratory phases expand without bound. Write down what you would pick and what would prove you wrong, then go looking. A candidate that dies in week two is a successful week two.
No named owner who can change a parameter. No cadence. No specification of what happens when an input is missing. No condition for switching it off. The recommendation was accepted and the operating process never heard about it, so the incumbent kept firing.
None of this is new, and none of it is going away. If anything the pattern is accelerating, because producing the forty-seven slides is now close to free. Why that raises the value of this work
Free, and complete
The method
20 entries across the book’s fifteen chapters. Each gives the definition, the mechanics, a worked example from operations, and the failure mode that most often breaks it. The book carries the cases and the politics; these carry the method.
- 01The Analysis Is Not the ProductThe deliverable is a challenger rule that beats the incumbent, gets installed, and can be stopped, not an explanation of what happened.
- 02Every Decision Already Has a RuleThe incumbent is the rule that actually fires, not the one in the manual. Every recommendation is a comparison against it, whether or not the write-up says so.
- 03Why the Incumbent WinsBeing better on the merits is necessary and not sufficient; a challenger must be designed so the forces favoring the incumbent stop working against it.
- 04Name the ChoiceA single sentence naming who will choose among which alternatives, by when, and to what end, knowing what will still be uncertain.
- 05Problem, Decision, OutcomeThe separation of what is wrong, what someone can change, and what should improve as a consequence, joined by a stated mechanism that can be wrong.
- 06Refuse the First BriefThe practice of returning an underspecified request as a proposed choice the owner can correct, rather than absorbing it into analysis.
What it argues
Every decision you are asked to analyze already has a rule running. The product of the analysis is a challenger that beats that rule, gets installed where it ran, and carries a built-in way to tell whether it was wrong.
- 1The incumbent starts ahead, so a challenger that is merely better on average will usually lose.The rule already running is trusted, its costs are familiar, it is the default on someone’s screen, and nobody has to sign anything to keep it. Its working parts are often undocumented: a pad, a spreadsheet, a promise between two supervisors. A challenger measured against the documented rule is credited with gains the pad already captures and blamed for none of the failures it prevents.
- 2Five questions define the work, and their answers fit on one page.What rule runs today; who can change it, and by when; what could replace it; what would make the challenger lose; and how it will run, with the condition that would stop it. The first three produce the choice, the fourth the evidence, and the fifth the rule. A page that answers all five can be refused line by line, which is the point of it.
- 3The work is finished when the challenger runs, not when the analysis is correct.A challenger that has not been installed has replaced nothing, and one installed without a way to tell whether it is failing can only be defended or abandoned on opinion. The way to stop it belongs to the product as much as the way to start it.
Writing
Recent articles
Forty-Seven Slides and the Rule Did Not Change
The usual outcome of competent technical work is a meeting that agrees it was thorough and a process that runs exactly as before. Effort will not fix it.
Read
The Solver Was Right and the Plant Overrode It
The failure mode of applied optimization is not an infeasible model or a slow solve. It is a correct model that a planner silently overrides.
Read
Your Model Is Not the Answer
Model quality is an intermediate artifact. Presenting it as the deliverable is the structural error in applied data science, and clearer writing is no remedy.
Read
What Would Change the Ranking
One note every other Tuesday.
One piece of the method, one worked example from real operations, and one thing to try before you open a notebook. No news, no roundups, no launch announcements.
Work with me
When the method needs to land in your organization, not just your inbox
Individual coaching for technical people whose analysis is correct and whose recommendations are not getting run, and team engagements where the problem is the operating process rather than any one analyst.