Estimating capacity when you don't know enough yet

When you cannot size work confidently, give a range with an explicit confidence level rather than a single number, and treat "I need to investigate before I can estimate" as a legitimate answer that triggers a time-boxed discovery instead of a guess. A single number with no confidence attached is indistinguishable from a considered one by the time it reaches a plan, so the uncertainty is lost exactly where it mattered. Record confidence next to the figure and the estimate stays honest as it travels.

Estimating capacity under uncertainty

"How big is this?" has three honest answers. A number. A range. And "I don't know enough yet to tell you."

In most planning conversations only the first is really allowed. The second gets rounded to its midpoint by the time it reaches a slide, and the third is heard as evasion. So people produce a number, everyone treats it as an estimate, and nobody records that it was a guess.

What actually gets lost

Two estimates of "3 FTE-months" can mean completely different things. One came from a team that has built this exact shape of thing twice. The other came from someone who has never opened the service and was asked in a meeting.

Both are 3 by the time they reach the plan. The difference — the only part that would have changed a decision — evaporated at the moment it was written down, because the format had nowhere to put it.

This is the same failure as writing an assumption in the same tone as a fact: not a wrong number, a number that has lost the information about how much to trust it. And it fails in a specific direction. Confident estimates and guesses both get planned as if they were confident, so the portfolio's total error is always larger than it looks.

Attach confidence to the figure, not to the conversation

The fix is unglamorous: record a confidence level next to every capacity figure, and let it travel with it.

Three levels is enough. High — we have done this shape of work before and the unknowns are small. Medium — we understand the problem, the approach has open questions. Low — this is a placeholder so the plan has a number in it.

What makes this work is not the label but the consequence. A commitment resting mostly on low-confidence capacity should not read as ready, and a system that says so out loud is doing the arithmetic the room usually skips. "Three teams estimated, all low confidence" is a different situation from "three teams estimated" — and only the first one tells you to go and look before promising.

Ranges, and the trap in them

A range is more honest than a point estimate and easier to misuse. Give leadership "2 to 5 FTE-months" and the number that survives to the next meeting is 2 — not from bad faith, but because the low end is the one that makes the plan work.

Two habits keep a range intact:

Say what the top of the range means, not just the bottom. "5 if the data migration is in scope" is a range with a cause. Someone can resolve it. "2 to 5" alone is a mood.
Plan against the top and report against the middle. If the top of the range would break the quarter, you do not have an estimate problem, you have a scope decision to make now rather than in week nine.

"I need to look first" is an answer

The third response — genuine inability to estimate — is the one worth protecting, because it is the only one that leads to the estimate improving.

The mistake is treating it as a refusal. It should be a fork: a time-boxed piece of investigation with a stated question and a date, after which you decide again. Not open-ended research, not a "spike" that quietly becomes the project. A week to answer "is the billing schema a blocker or not", then back to the decision.

This is also the honest response to pressure. When someone asks for a number you do not have, "I can give you one on Thursday after I check two things, or I can give you a low-confidence guess now and we record it as one" is a real choice rather than a stall. Most people take Thursday.

The estimate is not the commitment

Worth separating, because conflating them is why estimates get defended rather than revised.

An estimate is a claim about size, made with what is known now. A commitment is a promise about delivery, made by someone with the authority to make it. Treating the estimate as the promise means the estimator carries a risk they never accepted, which teaches them to pad — and padding destroys the signal for everyone downstream.

Keep them apart and both get more honest. The estimate can be low-confidence without anyone panicking, because low confidence feeds into the commitment decision instead of being the commitment.

How this works in DeliverySheet

Capacity is recorded per craft, by the team that would do the work, with a confidence level attached to each row — and the total is derived from the rows rather than typed, so the headline can't disagree with its own breakdown.

Confidence has consequences rather than being decoration. Capacity that is entirely very-low-confidence reads as "needs discovery" in the commitment verdict, not as ready. And "I need to investigate before I can estimate" is a first-class option at the planning stage — it routes to a time-boxed discovery with a stated question, and the demand comes back to the decision afterwards rather than drifting.

On the lightweight paths — a small, quick demand — capacity can honestly be skipped, because a number there would be theatre. On a standard or initiative path it cannot: those are the ones where the figure is load-bearing. The reasoning for that asymmetry is in the piece on why "none known" has to be a valid answer .