Capacity Is Arithmetic, Not Advocacy
Memo · 4 min read
To: the production leader who approves training headcount Re: why the backlog is the size it is, and what clears it
A learning function's staffing conversation should run on a model, not a feeling, and the model needs three terms that most training plans leave out: effort hours instead of courses, a maintenance load that grows with the catalog rather than the team, and a threshold below which the backlog stops being slow and becomes structurally unclearable. With those three terms on the table, a headcount request turns into arithmetic the business can inspect.
The wrong unit
A course is not a unit of work. A short online module and a multi-day technical curriculum differ by an order of magnitude, and a model that counts them as one thing each will lie at exactly the moment the answer matters. The function measures in effort hours: each deliverable type, at each complexity, carries an estimated build effort, and the backlog is the sum of those hours, not a count of tickets. That one change is what makes everything below defensible.
Capacity
Capacity is headcount multiplied by productive hours. Not gross hours: 2,080 a year is the ceiling, and nobody reaches it. Subtract leave and holidays, then subtract the interruption load every real job carries (meetings, support, the request that arrives at the desk), and a realistic utilization for project work lands near 78 percent. Ten people at that rate give about 16,200 effective hours a year. The number is lower than most plans assume, and it is the honest one.
Demand has three terms, not one
Most plans model demand as the backlog. There are three terms.
The backlog is the queue of approved work not yet started. The second term is recurring intake: new requests arrive every week, and a model that clears the backlog while ignoring intake describes a function that has stopped taking requests. The third term is the one that changes the conversation. Maintenance is the share of capacity spent sustaining what the function has already built: updating a course when the process changes, re-recording when a tool version moves, retiring what is stale. It is absent from most training plans, and it scales with the catalog, not the team. The more a function has built, the more of its hours are consumed before any new work begins. A mature catalog can carry a sustainment load approaching a full team's worth of hours, none of it on a roadmap.
Net progress, and the threshold
Net progress is capacity minus intake minus maintenance. Clearance time is backlog divided by net progress.
Illustrative figures, chosen to show the shape of the mechanism. Ten people at 78 percent utilization: 16,200 effective hours. Maintenance at 40 percent of capacity consumes about 6,500. Recurring intake takes 2,600. Net progress is roughly 7,100 hours a year, and an 11,000-hour backlog clears in about a year and a half.
Now move one lever at a time. Drop to seven people and net progress falls to about 4,200 hours; clearance stretches past two and a half years. Hold ten people but let the maintenance share rise to 60 percent as the catalog matures, and net progress falls to about 3,900; clearance is nearly three years. Do both, and net progress is under 2,000 hours against the same 11,000-hour backlog: nearly six years, which is to say never, because intake will not wait.
Push far enough and net progress turns negative. At that point clearance is undefined. The backlog is not slow to clear. It is structurally unclearable, and no prioritization scheme crosses that line, and neither does working harder. A model that locates the threshold is the difference between a staffing conversation and a staffing argument.
Pull the cheap lever first
Before this model was used to argue for people, it was used to argue against needing them. AI-assisted production, held behind the same review gates as hand-built work, raised drafting speed on the draftable share of build work. The leverage is bounded (a doubling of drafting speed yields about one and a half times the capacity, never twice, because review and analysis do not compress), but it was pulled first, and it closed part of the gap at almost no cost. Only the remainder went to a headcount request, with the model attached.
What changed in the room
Every deferral became a visible tradeoff instead of a judgment. When a requester hears "not yet," the model shows which lever would change the answer, and what it would cost. When the catalog grows, maintenance appears on the plan as a number rather than as a surprise. People argue with a decision. It is much harder to argue with arithmetic they can inspect.
The ask
Three things. Approve effort hours, utilization, and the three-term demand model as the basis for staffing decisions. Fund maintenance explicitly, as a line, so the catalog's upkeep stops being invisible. And hold the function to the model in both directions: when the numbers say a request can be absorbed, it is absorbed; when they say the backlog has crossed the threshold, the conversation is about capacity, not about effort.
