Capacity planning for engineering teams

Capacity planning, at the level an engineering manager is accountable for, answers one question: what can each team honestly promise this quarter? Not how hours will be spent — that is timesheets, a different and mostly worthless exercise — and not how a sprint will be filled. What follows is the whole practice, end to end, in four steps. Tomas, who runs three teams at a logistics scale-up, will walk them with us.

The unit: FTE-months, per team, per craft

The unit is the FTE-month — one full-time person working for one month. It is coarse on purpose: comparable across teams, convertible into a cost forecast, and blunt enough that nobody pretends it is precise. It is recorded per team and per craft — backend, frontend, data — because "the team has capacity" is exactly the sentence that hides the fact that the one data engineer is the constraint on half the plan.

Step one: calculate each team's honest capacity

Gross capacity is headcount times months. Tomas's payments team of six has eighteen FTE-months in a quarter — and that number is the first lie to kill, because none of those six people has an empty quarter. Subtract what is already spoken for: the support rotation (two FTE-months), the platform upgrade nobody is allowed to skip (one and a half), holidays actually booked (one), and realistic interruption — incidents, reviews for other teams, the meetings that happen whether planned or not (two and a half; if you have no data, twenty per cent is a defensible opening claim). Payments' honest number is eleven. Planning against eighteen would not make the missing seven reappear; it would decide, in advance and in silence, that the quarter ends in overrun.

The same arithmetic as a table:

If you want this walk as an artifact, the one-sheet capacity planning template is these five columns and nothing else.

Step two: size demands, with confidence attached

Each candidate demand gets a per-craft figure with a confidence level from one to five. Four backend FTE-months at confidence five and the same figure at confidence two are different claims — the second says "nobody has looked at the migration yet" in a form that survives being written down. The figure comes from the delivering team, never from planning on their behalf; a capacity number produced for a team is a wish with units.

Two honest special cases. "I cannot estimate this yet" is a legitimate answer, and it forks to a time-boxed discovery with an end date — not to a guess extracted under meeting pressure. And a low-confidence figure on a large demand is a prompt to shape further or hold reserve, not an invitation to argue the number down until it fits.

Step three: walk the line

In priority order, walk the demands down each team's number, subtracting per craft as you go. For payments: the retention work takes four backend, leaving seven; carrier invoicing takes three backend and one frontend, leaving four; the reconciliation rewrite wants five backend — and does not fit. Stop. That is the line, and it is per-team: an organisation-level total is meaningless, because FTE-months do not transfer between a data engineer and an iOS engineer at par. The demands above the line, once each team has agreed to its figures, are the quarter's commitments. Below the line does not mean "stretch goals" — stretch goals are commitments that have pre-arranged their excuse.

Step four: the first demand past the line

The reconciliation rewrite, sitting just past payments' line, is the entire point of the exercise — the demand every planning process is secretly designed to avoid looking at. There are exactly three honest things to do with it. Record a not-now: park it with a review date, or decline it with a reason its requester can read. Displace visibly: if it matters more than something above the line, that something moves down, on the record, and its stakeholder is told. Or commit it early anyway, because a real deadline outweighs the missing capacity — at the price of a recorded exception naming why it is early, what is missing, who accepts the risk, and what would change the decision.

What is not on the list is the popular answer: "we'll try to fit it in." Trying to fit it in is a displacement performed silently, against a victim chosen later, discovered at quarter end.

During and after the quarter

The line keeps working after planning ends. A mid-quarter urgent ask is an edit to a named commitment above the line, never an addition — capacity did not grow because a request arrived. And at quarter end, the closes come back to the number: Tomas closes payments' quarter, finds the team overran by two FTE-months, and resists the reflexive conclusion. The honest correction is usually not "estimate better" — it is a lower line or a bigger interruption reserve next quarter, because the closes are the only evidence about the honest number he will ever get. Capacity planned this way survives being looked at, because it measures what teams are committed to, never how individuals spent their hours. That distinction, and the rest of the capacity layer, is at the capacity overview.