Explaining Uncertainty to Someone Who Wants a Number

Someone asks what the campaign delivered. The honest answer has an interval, several caveats and a note about what could not be separated. Written decisions should preserve the evidence and uncertainty; this template guide covers employee write-up forms.

They want a number, and they want it because they have to justify a decision to someone else.

Hedging reads as not knowing. False precision buys credibility now and destroys it the first time the number is checked. Neither is acceptable, and the skill is the third option.

Why the demand for a single number is reasonable

Worth starting here, because analysts frequently treat it as innumeracy.

A decision is binary. Spend or do not. A range does not directly answer that; someone has to convert it into an action anyway.

They are accountable to someone who will also want one number. Passing along an interval means passing along the burden of interpreting it, which is not obviously their job.

Most of their information arrives as single numbers. Revenue, headcount, budget. An interval is an unfamiliar object in that context.

So the goal is not to teach them statistics. It is to give them something they can act on and repeat, that is also true.

Formulations that work

Give the range in money, not in percentages.

"Somewhere between £40,000 and £180,000 of incremental revenue" is understood immediately. "A lift of 3% with a 95% confidence interval of ±2.6 percentage points" is not, by most audiences, and the second is not more precise — it is the same information in a less usable form.

State the decision, not the estimate.

Frequently the range does not matter to the decision. "Even at the bottom of the range this pays for itself, so the answer is yes regardless of where in the range the truth is" is both honest and decisive.

This is the single most useful move available, and it is underused because analysts present the estimate rather than the implication.

Say what would change the answer.

"If the true effect is below £60,000 this does not pay for itself. Our estimate is above that, but not by a comfortable margin. A four-week holdout would tell us."

Use confidence language consistently, with a scale people can learn: this is measured, this is estimated, this is a guess. Three levels, used the same way every time, does more than any explanation of intervals.

Give a number and its status in one sentence. "About 400 conversions — that is our best estimate and it could reasonably be 300 or 500." Nobody needs the word "interval" for that to land.

What to avoid

Leading with caveats. By the third qualifier the audience has stopped listening and concluded you do not know. Answer first, qualify second.

"It depends." True and useless. If it depends, say on what, and give the answer for the likely case.

Precision you do not have. "£47,382 of incremental revenue" from an estimate with a six-figure range is a claim about your rigour that the next question will destroy.

Explaining the methodology when nobody asked. Have it ready; do not lead with it.

Refusing to give a number at all. This is the failure mode of careful analysts, and it is why decisions get made without them. If pressed, give the best estimate with its status attached — that is what the estimate is for.

When you genuinely cannot answer

Sometimes the honest position is that the question cannot be answered with what exists. See answering "did the campaign work" honestly.

Even then, do not stop at "we cannot know."

Say what you can see, and how unusual it is against normal variation.

Say what would be needed, specifically and with a timeline.

Offer certainty on something adjacent. Cost per acquisition, reach, delivery — observable things that are not nothing.

Commit to a date for a real answer. "Next campaign has a holdout, number by the end of the month" turns a refusal into a plan.

Building tolerance for uncertainty over time

The conversation gets easier if the groundwork is done outside the moment of pressure.

Use ranges routinely in ordinary reporting, not only when the news is uncomfortable. If intervals only appear when a result is disappointing, they will be read as excuses.

Show a forecast against what happened, repeatedly. Nothing teaches the value of a range faster than seeing point forecasts miss and ranges contain.

Record predictions and revisit them. A short log of what was expected and what occurred is the most persuasive artefact available for this argument, and it takes minutes to maintain.

Be right about the direction consistently. Credibility is built by being reliably approximately right, and it is what buys you permission to be uncertain when it matters.

When someone insists on a single number anyway

It happens, and refusing outright is usually the wrong call.

Give the point estimate, state its status once, and move on. "Our central estimate is £110,000; it could plausibly be half or double that." Then answer the next question.

Put the range in writing, in the document that follows. The verbal answer serves the meeting; the written record protects everyone when the number is revisited.

Do not privately treat the point estimate as more certain than it is. The risk of giving single numbers is that you start believing them.

The summary

Answer first, qualify second. Leading with caveats loses the room and the argument.

Put the range in money, and say what decision it supports.

"Even at the bottom of the range, this is worth doing" is often the whole answer, and it is both honest and decisive.

Use a consistent three-level scale — measured, estimated, guessed — and people will learn it without a lecture.

And build the tolerance in quiet periods, so that uncertainty in a difficult moment reads as rigour rather than as evasion. For practical language and presentation principles, see the guidance on communicating uncertainty.