What does a capacity plan actually assume?
In most firms of 20 to 200 people the capacity plan is one line of arithmetic. Take the number of fee earners, multiply by available hours, multiply by a target utilisation. That gives a quantity of sellable time for the quarter, and the pipeline gets sized against it.
That one line carries three hidden assumptions. It assumes the mix of jobs is stable. It assumes each grade spends roughly the same share of its week on the same activities. And it assumes the unbillable load is a fixed tax rather than something that can move. Rebuild one process properly and all three stop being true in the same month. The plan still produces a number. The number just stops describing the firm.
Where does the freed time actually land?
Never evenly, which is the part the blended plan cannot cope with. If you rebuild client reporting, the time comes back to the people who write reports, in the week the reports are due. If you rebuild bid and proposal production, it comes back to whoever drops everything when a tender lands, which in a lot of firms is one or two senior people and a bid coordinator.
So the first question is not how many hours were saved. It is: which people, at which grade, in which weeks. A gain of several hours concentrated in the last week of every month is a different asset from the same hours smeared across a quarter. The concentrated version can be planned against. The smeared version evaporates.
Firms with sharp seasonality feel this most. Accountancy practices do not have a capacity problem in July, they have one in the run up to filing deadlines. A process rebuild that takes pressure off the peak is worth far more than the annual hour count suggests, and a plan built on annual averages will miss that completely.
What becomes the constraint instead?
Capacity is set by the tightest link, not by the total. Take the slowest step out and the firm does not become uniformly faster. It becomes faster until it hits the next limit, and that limit is usually a person.
The common ones are worth naming. Review and sign-off capacity, because drafts now arrive faster than anyone can approve them. Specialist knowledge that only two people hold. Client responsiveness, where the file waits on information that has not come back. Physical attendance, for anyone doing inspections, surveys or site work. If drafting was consuming the senior hours that review needs, removing it helps. If it was not, you have just built a queue in front of the same desk. That is the pattern we set out in the partner bottleneck, and it reappears every time throughput improves upstream.
Before the build goes live, write down what you expect the next constraint to be. After it goes live, check whether you were right. If the answer surprises you, that is your next process, and what to automate first covers how to rank the candidates.
Why re-plan the pipeline before go-live rather than after?
Because work takes longer to win than a process takes to rebuild. A process can be mapped, built and live inside 30 days. A tender cycle, a referral relationship or a panel appointment runs to months. If you wait until the capacity is visibly free before you go looking for work to fill it, you have booked yourself a quiet quarter and a nervous partnership.
The practical move is to start business development in parallel with the build, sized to the capacity you expect rather than the capacity you have. That means telling the people who win work what is changing and when, so proposals and panel submissions can reflect the new delivery shape. It also means cleaning the front of the pipeline: more enquiries only help if they are the right ones. Enquiry qualification and where AI helps you win work both bear on that.
How do you re-baseline without guessing?
Stop planning in hours and start planning in units of work. Hours are an input nobody measures accurately. Units of work are countable: matters per fee earner per month, reports per surveyor per week, placements per consultant, engagements per manager, policies per broker. Every firm knows roughly what those numbers were. The job is to establish what they are now.
Do it like this. Take the last ten jobs that went through the old process and record the elapsed time and the touch time for each. Run the same measurement on the first ten through the new one. Include the awkward cases, because the easy ones always improve first and will flatter the result. Then re-state the team's realistic job count per month from the new figures, and only then change the plan. The guide to calculating unbillable hours covers the before measurement, and the capacity calculator does the arithmetic on the after.
One warning. Do not re-baseline off a single month. New processes have a settling period while people learn where the judgement calls sit, and the first cycle is rarely representative in either direction.
What does it mean for the next hire?
It rarely means hiring nobody. It usually means hiring differently. If the drafting load has fallen and the review load has risen, the firm needs more capacity at the grade that checks, advises and wins work, and less at the grade that produces first versions. That is a real change to the pyramid, and it affects how juniors learn, which we deal with in what changes for junior staff.
Cancelling a hire on the strength of a forecast is a mistake. Deferring one while you hold a measured baseline for a quarter is reasonable. Automation versus hiring sets out how the two decisions actually compare.
Who decides what the gain is for?
There are three doors and the firm has to pick one on purpose. Take on more work at the same headcount. Do the same work at better margin. Or give the hours back, as earlier finishes and less weekend catch-up, which is a legitimate choice if retention is the pressing problem. Partnerships that do not choose end up with a fourth outcome by default: the time disperses, nobody can say where it went, and the rebuild looks like it did nothing.
Whichever door you pick, say it out loud before go-live and attach a measure to it. If the answer is more work, the measure is jobs started. If it is margin, the measure is realisation. If it is hours back, the measure is recorded time per head, and your utilisation target has to move with it, as utilisation targets after the admin goes explains. Pricing is a separate decision again, covered in pricing work that AI makes faster.
If you do not yet know which process is eating the week, that is the thing to settle first. The audit estimates the unbillable hours your fee earners lose, what they cost a year at your own charge-out rate, and which process to take on first. Capacity planning after that is arithmetic. Before it, it is guesswork.