What the system produces
One number, always current: committed effort across a rolling eight-week window, expressed in whatever unit your business actually plans in — person-days is the most common and the least ambiguous. Alongside it, a threshold, and a written rule about what sales may commit to on each side of it.
That is genuinely the whole output, and the minimalism is deliberate. Capacity systems fail by producing a planning view rich enough to require interpretation, and interpretation does not happen during a call.
The second output is a forward shape: the same number broken by week, so you can see not just whether you are full but when the gap is. That is what makes a later start date a specific offer rather than a vague deferral.
The architecture, part one: what feeds the number
Stage one is estimate capture, triggered when a deal is signed. Remaining effort and the weeks it spans, recorded against the engagement. Owner: whoever will deliver it. Without this the whole system has no input, and this is where most attempts stall.
Stage two is consumption. As work is completed, remaining effort decreases. The design rule that matters: this must be a byproduct of work moving, not a separate reporting act. If updating remaining effort is its own task, it will be accurate for three weeks and then fiction. Tie it to stage transitions that already happen.
Stage three is the known-load additions everyone forgets, and it is what separates a real number from an optimistic one. Booked leave. The recurring internal commitments — the weekly review, the sales calls, the admin — that consume delivery time and appear on no project plan. A revision allowance per engagement type, based on what actually happens rather than what is scoped. In most service businesses these three add up to somewhere between a fifth and a third of nominal capacity, which is exactly the margin that makes a well-planned quarter fail.
Stage four is aggregation into the rolling eight-week figure, recomputed whenever anything upstream changes. Owner: the system. Nobody assembles this.
The architecture, part two: the threshold and the rules
The threshold is a percentage of true available capacity, set deliberately below a hundred. Eighty is a common starting point and the exact figure matters less than the fact that it was decided calmly rather than during a negotiation.
Below the threshold, sales may commit to a start date within the window. Above it, the only available commitments are a start date outside the window or a place on a waitlist. That is the entire rule and its power is that it converts a judgment call made under social pressure into a decision made in advance.
Stage five is the visibility of the number at the point of commitment. It has to be where the sales conversation happens — in the CRM, on the deal record, in the channel where the founder works. A capacity number in a planning tool nobody opens during a call is not an instrument.
Stage six is the override, which must exist and must be recorded. Sometimes you take the work anyway, for a strategic client or a reason that outweighs the rule. The system's job is not to prevent that; it is to make sure the decision is conscious and that somebody can see, a quarter later, that four overrides in a row preceded a bad month.
Stage seven is the weekly reconciliation: planned versus actual effort on completed work, which is what makes future estimates better. Without it the number stays as wrong as your estimating was on day one.
The failure edges
The first: planning to a hundred percent of nominal capacity. A business planned to full theoretical utilization fails on its first ordinary week, and the founder reads that as a people problem rather than as an arithmetic one.
The second: remaining effort maintained as a reporting task. It decays within a month, and a stale capacity number is worse than none, because decisions get made against it confidently.
The third: the threshold gets renegotiated during the call it was designed to govern. Once that happens twice, the threshold is advisory and the system is decorative. Overrides are fine; moving the line is not.
The fourth: no forward shape. A single aggregate number tells you that you are full and not when you are free, so the conversation with the prospect becomes no rather than yes, starting in five weeks — which is a materially worse commercial outcome from the same information.
The fifth: capacity measured in headcount rather than in the constraint. If one senior person is required for a phase of every engagement, your capacity is that person's availability regardless of how many people you employ. Systems that aggregate across a team hide single-resource bottlenecks completely.
The sixth: no distinction between committed and probable. A pipeline deal at eighty percent is not capacity and it is not nothing, and businesses that count it as either get the same wrong answer from opposite directions. Track it as a separate shadow figure so you can see the week where probable work would push you past the threshold.
The constraint question, and how to model it honestly
Before building anything, identify what your capacity is actually constrained by, because aggregating across a team hides the real limit in most service businesses. If every engagement needs a specific senior person for its first two weeks, your capacity is that person's availability, and a team-level number will show room that does not exist.
Model the constraint explicitly: capacity per constrained resource, not just in total. That usually means two or three tracked pools rather than one — the senior reviewer, the delivery capacity, and whoever does the specialist step. The aggregate number is still worth having for the sales conversation, but the threshold has to be evaluated against the tightest pool.
The second honesty adjustment is the revision allowance, and it should come from history rather than from scope. Look back at the last ten engagements of a given type and compare planned to actual effort. The ratio is usually somewhere between a fifth and a third over, it is remarkably stable per engagement type, and applying it turns a plan that fails every quarter into one that holds.
What done looks like, and what it takes to build
Done is being asked on a call whether you can start in three weeks and answering from a number you did not have to produce — including the case where the answer is not in three weeks, but the fourteenth is open.
The checklist: remaining effort captured at signature; consumption tied to stage transitions that already occur; leave, internal commitments and a revision allowance included; a rolling eight-week aggregate with a weekly breakdown; a threshold with written rules on each side; the number visible where deals are committed; a recorded override path; and a weekly planned-versus-actual reconciliation.
For a business with five or fewer active engagements this is genuinely a spreadsheet and an afternoon, and it will work. The build becomes real when effort is not tracked at all, when the delivery system and the sales system do not share data, or when the constraint is a single person rather than the team.
The prerequisite is a delivery process where work has stages that change state, because that is what consumption hangs off. In a business where every project runs differently, there is nothing for the number to update against — and that is the thing to fix first.
The number must already exist when a yes is needed, so it has to update as a byproduct of delivery. Rolling eight weeks, a threshold set below full, and a rule decided before the call.