Working With a Client Who Has Already Decided

Sometimes the request is not for an answer. It is for support — a number, a chart, a conclusion that confirms something already settled. A useful counterweight is to make the next analytical step explicit; this guide discusses becoming more proactive at work.

Recognising this early matters, because the two obvious responses both fail. Refusing makes you an obstacle and gets you excluded from the next decision. Complying makes you an instrument and, eventually, responsible for a conclusion you did not believe.

Recognising it

The requested output is specified before the analysis. "I need a slide showing the campaign drove the uplift."

The framing assumes the answer. "How much did the campaign contribute?" rather than "did it contribute?"

Constraints appear on the method. A particular comparison period, a particular segment, a particular attribution model — chosen without a stated reason.

Previous analyses that reached a different conclusion are not mentioned.

The timeline is short and the decision is already communicated.

None of these is proof, and any two together are worth noticing.

Why it happens, without assuming bad faith

Usually not dishonesty. Usually structure.

The decision was made for reasons that are not analytical — strategic, relational, political — and those reasons may be perfectly good ones that simply do not appear in the data.

They have already committed publicly, and reversing costs credibility.

They are being asked for a number by someone else, and are passing the request down.

The analysis is a formality in a process that requires one.

Treating it as dishonesty is both usually wrong and tactically poor. The useful assumption is that the person has a real constraint you cannot see.

The response that works

Ask what the decision is and what would change it.

"Just so I produce something useful — is this decision open, or is the analysis to support a decision already made? Both are fine, I just need to know what to write."

Asked neutrally, this is not confrontational, and the answer is usually honest. Most people will say "it is essentially decided, I need something for the board", and now you both know what you are producing.

Then produce that, honestly. A supporting document is a legitimate artefact, and it can be accurate.

Producing supporting material honestly

Report what is true, and let the selection be the framing rather than the numbers.

Do not manufacture a causal claim. "Sales were 14% above the prior period during the campaign" is true. "The campaign delivered a 14% uplift" is not, unless it was measured. See answering "did the campaign work" honestly.

Include the caveat, briefly, once. In the document, not the headline. It does its work later, when someone checks.

Do not put your name on a claim you would not defend. This is the line, and it is worth holding calmly rather than as a stand.

And keep a record of what you advised and what was produced. Not defensively — because in twelve months someone will ask how the number was arrived at, and the answer should be findable.

When the decision is genuinely wrong

Sometimes the analysis says the opposite and the decision is expensive.

Say it once, clearly, in writing, early. Not repeatedly, and not in the meeting where it is being announced.

Quantify the exposure rather than arguing the method. "If the incremental effect is what the test suggests, this spend returns about a third of its cost" lands better than a discussion of attribution models.

Propose a cheaper test rather than a reversal. People reverse decisions far more readily when they can do it on new evidence than when asked to accept they were wrong. A two-week holdout gives them that route.

Then let it go. You are not the decision-maker, and being right repeatedly and loudly is a good way to stop being consulted at all.

Revisit with data when it exists. If the outcome confirms your concern, the useful move is a neutral note — not a reminder that you said so. The credibility accrues on its own, and it accrues faster without the reminder.

Building the position over time

The situation is easier to handle when it is not the first conversation.

Be involved before the decision. Most of this problem is a symptom of analytics being consulted at the reporting stage rather than the planning stage. A measurement plan agreed beforehand removes the whole dynamic. See building a measurement plan.

Be reliably approximately right, publicly. Credibility built on small accurate calls is what gets you heard on a large one.

Show your uncertainty routinely, so that a caveat on an inconvenient result is not read as an excuse. See explaining uncertainty.

Make agreeing easy. A person who has to concede publicly will resist; a person who can say "we ran a test and it told us something" will move.

The line

There is one, and it is worth being clear about where.

Selecting which true things to present is normal. Every report does it.

Choosing a favourable comparison period is questionable and common, and it should at least be disclosed.

Presenting a correlation as a causal finding is a false statement, whatever the framing.

Producing a number you know to be wrong is not a grey area.

And if that is the request, the answer is no — said calmly, once, with an offer of what you can produce instead. Most requests do not go that far, and the ones that do rarely survive being named plainly.

The summary

Ask early whether the decision is open. The answer is usually honest and it changes what you produce.

Supporting material is a legitimate artefact, and it can be accurate — the selection is the framing, not the numbers.

Say the disagreement once, in writing, with the exposure quantified, and propose a test rather than a reversal.

Do not attach your name to a causal claim that was not measured, and everything else is negotiable. For safeguards around analytical quality and decision-making, see the Aqua Book guidance.