Refuse the First Brief

The first brief is almost never the decision. It is a problem, a deadline, a deliverable, and a report on the political weather, assembled by someone who is genuinely busy and reasonably expects you to work out the rest.

Treat it as a first offer rather than a specification. Absorbing it is the path of least resistance and it is the single most expensive habit in technical work, because everything downstream inherits the ambiguity. Refusing it does not mean declining the work. It means offering the owner a better object, which they can accept, correct, or overrule. Each of those answers is information. Opening the extract before you have one of them is how the default week begins.

Five shapes the first brief actually arrives in

Naming the shape is already work, because it tells you what kind of translation you owe.

A symptom. Inventory is high, fill is low, figure out what is happening. A problem in search of a decision. Analyze it as a report and you produce the forty-seven slides.

A feared metric. Fill, turns, forecast error, utilization, already sitting in somebody’s objectives. An outcome-proxy in search of a mechanism. Take the metric as the specification and you will optimize a dashboard.

A solution somebody already likes. We need machine learning. We need a dashboard. We need an agent. Offered as though it were the choice. The translation is three questions: who will use a prediction to change an action, what are they choosing among without a new model, and when does that choice go stale. For an agent, ask whose work changes, what it may do without a human, and what a failure costs. Until then, a demonstration is a demonstration, and it should not consume a freeze date.

A compromise. Two executives covering incompatible interests with a joint request: improve fill and cut inventory. The blend, now with a sponsor. Model both halves as constraints and you will announce either infeasibility or something implausible.

Reassurance, or ammunition. They already know what they will do and want a chart that makes the meeting survivable. An informing job presented as discovery.

The four questions

You need four answers before you spend serious time. If you cannot get them in a fifteen-minute conversation, that itself is information: either the choice is still unspecified, or nobody who can specify it will talk to you. In both cases, opening the warehouse is a stall.

1. Who will use this to change an action? Not who is interested. Who can commit, or who will run a rule. If the answer is a standing meeting, you have an audience and not yet an owner.

2. What are they choosing among, including the status quo? If they cannot name two commitments of the same kind, you are not yet on a decision. “Do a study or do nothing” is not a pair of the same kind. “Keep the current rule or change it” is.

3. When does the choice go stale? The meeting time is usually the wrong date. The freeze, the contract window, the promotion start, the quarter-end cash event: those go stale. A date that is only a meeting produces a deliverable a meeting can consume.

4. What would they do if you came back empty-handed? Blunt and invaluable. If finance will take the cut anyway, the work is support for a decision already made, not a ranking. There may still be a real job, such as designing the least harmful version of the cut, but it is not discovery. Asked as a test of the owner’s resolve, the question is rude. Asked as protection of the owner’s date, it is ordinary.

The four questions are the choice sentence asked aloud. The point is to check whether its blanks are real.

The translation goes back as a proposal

The form that works is a short reply that names the decision, the alternatives, the owner, and the date, and asks whether that is what the sponsor needs. It answers the four questions before asking the owner anything, so what they receive is a proposed object rather than a request for clarification.

At Meridian Motors, Priya Shah asked for a dashboard by Friday. The translation fits in four sentences. You are choosing a replenishment rule for the April freeze, not standing up a dashboard. The alternatives are to keep the current rule; to take finance’s cut, which brings every item down 20 percent by the freeze and then runs the current rule at targets 20 percent lower; or to set stock by class. Correct me if that is wrong. If we cannot rank those before Thursday, will finance take the cut anyway?

Most good owners will correct a note like that, and each correction is a better brief than the email. Priya’s correction accepted the three rules, said she still needed something in Friday’s room because finance would be there, and said she would not put her name to a class-based rule she had not seen survive the warehouse. That is the translation working. The correction surfaced a real interest the email had hidden, and it turned “figure out what is happening” into the far more answerable question of what would have to be true for the commitment to become possible.

The translation finds the decision without discarding the rest of the brief. If the owner insists on the dashboard, build a scorecard for the rule: a named owner, thresholds, and the few numbers that would change the ranking at the next freeze. What should not be built is a record of last quarter with no rule attached.

When not to analyze

Further analysis costs calendar time, attention, and the option to act while the system is still in a state you understand. So refusal sometimes applies to the request for work itself.

When the owner already knows which alternative they will pick, and says so, the work is informing or implementing. A six-week study whose only function is to bless a number is not evidence.

When nothing that could be learned in time would change the pick, refining the inputs is wasted effort. Record what was skipped and why it could not change the ranking, so that skipping is visibly part of the method.

When the dispute is about who loses rather than about what is true, analysis cannot settle it. Do not treat a political stalemate as a data-quality problem.

When a reversible test is cheap, run the small trial and watch it. A staged class assignment at two distribution centers is often better evidence than a platform.

Delay is sometimes the right choice, provided it is written as one, with a cost and a date. “We should study this further,” offered without either, is how freeze dates pass under the incumbent rule while everyone feels responsible.

Refusing without becoming the problem

A translation already carries a mild accusation, that the email did not name the work. Perform that accusation and you will be remembered as difficult, and the next request will go to someone more accommodating.

The technique is to refuse by supplying, never by withholding. Name the owner’s deadline in the first sentence, offer a one-page ranking as the thing the meeting can actually use, and say plainly which question will not be answered this week. Do not mention method, frameworks, or books. They are now editing a concrete sentence rather than defending a vague one, and the exchange costs ten minutes.

There is also a version of refusal that is itself a stall: endless clarification, no ranking, and a standing meeting called problem framing. Limit the translation to fifteen minutes, one page, and the four questions. If that does not settle the object, proceed on an explicit assumption (“we are ranking replenishment rules for Priya, for the April freeze, unless you tell us otherwise”) and send that sentence with the work. Silence is then consent to a named object, which is more than the original email provided.

“You will look uncooperative”

Sometimes you will, and sometimes that is the job. Standing is uneven. A vice president can decline a brief in one sentence. Somebody two levels down cannot. What remains available from every position is translation. Confrontation does not.

Both extremes fail. A team that never questions a brief spends its life supporting other people’s meetings. A team that questions every brief stops being asked. The test is not whether anyone was made comfortable. The test is whether the object of the week can be installed.

If the owner is any good, translation reads as competence: they asked for a dashboard because that was the word they had. If the owner is not good, or is not really the owner, you face a different decision: do the informing job honestly, or decline the work. Presenting it as discovery is the one option the method does not allow.

A test you can run in five minutes

Take the request on your desk and write the four-sentence reply: the decision, the alternatives including the status quo, the date that goes stale, and the empty-handed question. If you cannot fill a blank, that blank is the first conversation, not the first query.