Ranges beat certainty: calibrated decisions for founders

Somebody asks what the number will be. You say a number. Everyone writes it down. The plan is built on it, and if it lands anywhere near, you look competent.

The problem is not that the number is wrong. It is that the number carried no information about how wrong it might be, and the width was the part the decision actually depended on.

The width was the decision

Consider two forecasts that share a midpoint. Revenue next quarter of 400, plus or minus 20. Revenue next quarter of 400, plus or minus 250.

The first supports a hiring commitment. The second supports a hiring plan with a trigger. Same midpoint, opposite decisions, and a point estimate of "400" collapses them into the same sentence.

This is why ranges over false precision is not a stylistic preference. "8 to 14" and "11.3 expected" are not two ways of saying the same thing. The second asserts a confidence the underlying evidence almost never supports, and it removes the reader's ability to notice.

The two questions a range forces

Give someone a range and two questions become unavoidable.

What makes it the low end? Answering this names the risks concretely. Not "there is execution risk" but "the low end is where the two enterprise deals slip past the quarter and the second engineer starts in March instead of January".

What would narrow it? This is the question that produces work. A range narrows when an assumption gets tested, so asking it converts anxiety into a short list of things to go find out. Often the highest-value action available in a planning session is not choosing a path but buying information that narrows a band.

Neither question is askable of a point estimate, which is the quiet cost of the format.

Stating ranges without theatre

Two failure modes bracket this. One is fake precision. The other is a range so wide and so unexplained that it communicates nothing except that the author declined to think.

Three rules keep a range honest.

State the basis. A range without its assumptions is a shrug in numeric form. "340 to 430, assuming conversion holds within its trailing band and no change to headcount" is a claim you can argue with. "340 to 430" is not.

Say which end is load-bearing. For most decisions only one tail matters. A runway range matters at the bottom. A capacity range matters at the top. Naming the relevant tail focuses the conversation on the half of the distribution that can hurt you.

Attach a trigger, not a review date. "Revisit if two consecutive weeks land below the band" beats "revisit monthly", because the condition fires when the world changes rather than when the calendar does.

Calibration is a slow asset

A range is only meaningful if the ranges you produce turn out to contain reality about as often as they claim to. That property is calibration, and it cannot be asserted. It has to be measured against a record of past predictions that were written down before the outcome was known.

Which is why it takes elapsed time. You need predictions, then outcomes, then enough of both to say anything. There is no shortcut, and any system offering calibrated confidence on day one is describing a hope.

The right posture in the meantime is deliberately wide bands. Ranges stay honest-wide until calibration evidence justifies narrowing them, and a prediction that cannot be made honestly is an abstention rather than a guess. Wide bands early are unsatisfying and correct. Narrow bands early are satisfying and unearned.

The same rule applies to how a system stores its own history. Predictions carry ranges and a model version, and are preserved unchanged when reality arrives. A prediction that can be edited after the outcome is not a prediction, it is a retelling, and a company that retells its forecasts learns nothing from them.

What this looks like in a real decision

Take a concrete shape: whether to raise price on a plan.

The uncalibrated version produces a number for expected revenue change, everyone debates the number, and the debate is really about which assumptions each person is silently holding.

The ranged version produces something more like this. Revenue effect over two quarters lands between a modest decline and a material gain. The low end is the world where churn responds faster than the trailing data suggests. The high end is the world where the segment that would leave was already unprofitable. The ranking flips if churn response exceeds a stated threshold, which is measurable within about six weeks of the change.

Now the decision is not "pick a number". It is "run the change on a segment where the churn response is observable early, and hold the rest until the band narrows". That is a better decision, and it came from the width, not the midpoint.

Brainis models alternative futures before recommending a path — the mechanism is on the Strategy Engine, and the layer that produces and grades the ranges is the Business World Model.

Certainty is easier to present and worse to act on. The range is the part that tells you what to do next.