THE BOTTLENECK
The FY27 plan is done except for one tab. Revenue is filled in, up 25 percent, because two accounts are ramping and the pipeline supports it. Now the headcount tab. Two more CSRs, an estimator, a scheduler in the second half. Four boxes, four salaries, a number the CFO can add up.
That tab is a capacity forecast. It is the only place in the entire budget where next year's capacity gets written down, and it is written in the one unit a budget understands, which is people.
Now notice what nobody does to it. You would never accept a revenue forecast that assumed every deal closes. Somebody would ask about win rate, about timing, about which two accounts carry the second half. The headcount tab gets none of that. It assumes four reqs, four hires, roughly on schedule, at roughly the wage in the cell.
Check that assumption against the market it depends on. NFIB's August survey, published September 3, found 35 percent of small business owners with openings they could not fill, 31 percent of them for skilled workers against 13 percent for unskilled. Among owners actually hiring, 82 percent reported few or no qualified applicants. In the same survey, a net 17 percent plan to add jobs in the next three months.
Read those last two together, because that is the whole thing. Plenty of companies are planning to hire. Most of the ones already trying cannot find anybody. Your plan lives in the first number. Your outcome lives in the second.
So here is the growth ceiling, stated plainly. If the estimator req does not land by May, the volume you have already promised the bank arrives at a process that can only quote as fast as two people can quote. The revenue tab says 25 percent. The headcount tab is supposed to say how. Right now it says how only if four strangers say yes.
THE FIX
Turn the headcount tab back into work.
For each box on it, answer one question: what countable thing rises next year that made you draw this box? Not "we need another estimator." Closer to: 78 RFQs a week instead of 60. Another location's worth of payroll cycles and safety numbers. If you cannot name the count, the box is a hunch. Hunches are allowed. They should not be load-bearing.
Then find where that count lands, step by step. That is the part that gets skipped, and where the surprise is. The step that runs out first is usually not the seat you were about to post. It is a signature that batches on Thursday, or a quote file only one person can have open at a time, or one engineer turning estimates into routings at 90 minutes each with nobody else qualified.
That last one is not an argument for fewer engineers. You have one, the work is growing, and the real question is whether the next 30 percent of it needs a second person or needs the first one to stop re-keying. A Census working paper released in April, built on the Business Trends and Outlook Survey AI supplement, found 66 percent of AI-using firms use it only to augment tasks, with AI-related employment decreases at 2 percent of firms.
When this tells you to hire, hire. If the rising count is mostly judgment or mostly hands on material, you have a req, and now you can write it knowing which one and why. Skip it on a process you already plan to change.
Three versions to get the prompt below right. Version one divided by "works 40 hours a week" as though a scheduled shift were a measurement. Version two would not rank a step whose hours I had not written down, so it quietly dropped the heaviest load in the test data, 90 minutes an order on one engineer, out of the answer. The third fix was not arithmetic: version one recommended I time-study a named employee. Not what this is for, so it counts the work now, not the person.
I run operations at a company of about 140 people. Below is how one
process runs today, step by step, plus the volume we are planning for
next year. Tell me which step runs out of capacity first as volume
rises, and what kind of constraint it is.
Ground rules.
Use only what I wrote. No industry benchmarks, no typical durations, no
normal staffing levels. Where I gave you nothing, write "not provided."
Scheduled hours are not capacity. Where a step's owner also does anything
I have not quantified, label that capacity "upper bound" and treat every
percentage derived from it as a ceiling, not a value.
Where I give a range, run both ends and report both. Never the midpoint.
Where I give a mix (how many of X are Y), say whether you held that mix
constant at plan volume, because I did not tell you.
Describe the process, not the people. Role names only, never personal
names, tenure, or work habits. Do not propose measuring any individual's
time; where a step needs better measurement, ask for the work to be
counted, not the person.
Name system and license limits as constraints. Do not name products or
propose replacing anything.
Return these sections.
THE MATH. One table. Per step: the unit that drives it, units per week
today and at plan, demand hours at both ends of any range, the capacity
figure in my own words, and whether that capacity is an upper bound.
Where a step has demand hours but no capacity figure, leave capacity
blank and fill in the rest.
FIRST TO BREAK. Two lists.
A. NO CAPACITY FIGURE, HIGHEST LOAD FIRST. Every step with demand hours
but no capacity figure, ordered by demand hours at plan, largest first.
Steps with no demand hours at all go underneath those, in process order.
Head the list with this sentence: any of these could be the first
constraint and nothing I gave you can rule it out. Put this list first.
B. RANKED. The remaining steps, by the percent increase in that step's
own driver it absorbs before passing its limit. Show the division. Name
the driver on every row, because my growth rates differ. Rank on the
pessimistic end of any range and say that you did. A step already past
its limit gets a negative percent and goes at the top. For the top three,
name the constraint type: hours available, a single qualified role, a
batch or approval cadence, a system or license limit, an external party's
response time, or a work-routing rule.
Anything that does not scale with volume, such as a seat count, goes in a
short list underneath, in neither ranking.
Where one role owns two steps, show the steps separately and that role's
combined load as its own row.
QUEUES AND SHARED RESOURCES. Anything that batches on a schedule, any
resource two steps share, any rework loop, any wait between steps, and
what each wait depends on. A step that runs once a week is a cadence, not
a capacity, and belongs here.
WHERE THIS ANSWER IS WEAK. In order: what to count that would change the
ranking, any place my volume assumptions contradict each other, and every
place I was vague enough that you had to pick a reading. Name the reading
you picked.
Do not recommend software. Do not estimate savings. Do not tell me to
hire.
THE OPERATOR'S READ
On August 28 BLS published its preliminary benchmark revision to payroll employment. The headline: total nonfarm for March 2026 marked down 79,000, or 0.1 percent. Which is nothing.
Underneath the nothing, sectors moved. Manufacturing revised down 67,000, half a percent. Wholesale trade down 86,200, or 1.4 percent. Retail trade down 154,600. Transportation and warehousing up 135,100, a full 2 percent the other way.
Two things to take from that. The aggregate figure that gets reported is the one least likely to describe you. And this is a restatement of March arriving at the end of August, right as you write next year's plan.
Your own numbers are not late, and nobody revises them. How long your last three reqs sat open, and whether you closed them at the budgeted wage, is what decides whether the boxes on that tab are a plan or a wish. No national series is going to tell you that.
ONE MOVE THIS WEEK
Open next year's headcount plan. Beside each new box write two things: the countable work that made you add it, and the date you would have to start recruiting for that person to be productive when you need them. Then pull how long your last three hires really took, posting to productive. Twenty minutes.
Any box with no count beside it, or a recruiting date already behind you, is this week's real conversation.
— Rich
P.S. If you already know which box is the one that will not get filled, you did not need the exercise. You needed to say it out loud in the budget meeting.