Ten thousand years of demand.

Commercial terms are agreed in advance; demand arrives later. A fixed price per project means the hub carries the risk of projects running larger than planned, of automation gains arriving slower than hoped, and of quiet months with a full team.

A single forecast hides all of that. So the question was never “what will next year look like?” but “across the years that could plausibly happen, which terms keep the team funded and the client fairly charged?”

Each of the ten thousand years is assembled from the same five steps. Where the data could speak, it does; where it could not, the assumption is explicit, named and editable.

1. 01

   Arrivals

   For every project type and month, the number of new projects is drawn around an expected rate: the historical rate × a seasonal index × a growth multiplier drawn once per year—so a good year is good across the board, as it is in reality.

2. 02

   Dispersion

   Real monthly counts scatter more than pure chance allows. A negative binomial captures that, with over-dispersion estimated per project type by the method of moments; sparse types borrow a pooled estimate, and types with no excess scatter fall back to Poisson.

3. 03

   Project size

   Each simulated project copies a real past project of the same type—a bootstrap of actual hours—with an optional small jitter so the future is not limited to exact repeats. Where only totals existed, sizes came from PERT distributions instead.

4. 04

   Effort after AI

   Automation cuts are drawn per project type and year and reduce production work only. Coordination and leadership effort comes from an activity-level catalogue, so the model knows which hours shrink with automation and which do not.

5. 05

   Capacity and price

   Worked hours are set against team capacity month by month, and every pricing structure is evaluated on the very same simulated years—common random numbers—so differences between options are real, not noise.

Each option produces a distribution of outcomes. Three points on it carry most of the decisions:

P20 · the floor

Exceeded four years in five. It sizes the committed core of the service.

P50 · the centre

The central estimate, used for planning and pricing.

P80 · a busy year

It sizes peak capacity: what the team must absorb in a busy year.

Alongside them: the probability of a loss under each option, the headcount the work needs against the headcount the revenue can carry, and how often a peak month would exceed the team.

Several commercial options were evaluated on the very same simulated years. Last year’s real projects were also replayed under each option—same volume, new terms—the cleanest comparison of old and new.

Two options can earn the same on last year’s volume and still behave very differently over ten thousand simulated years: one holds steady while another moves with every swing in demand. That gap is the risk—and the model shows who holds it.

A bridge decomposes the change from one year to the next into four steps—terms, rate, automation and volume—that add up exactly to the total, so no single number has to carry an unexplained story. Scenarios stress the structure further: a market halving or withdrawing, a seasonal year, or a few large clients coming and going, simulated by resampling the clients themselves.

Assumptions in plain sight

Every input lives in Excel with a description, row by row. Changing an assumption needs no code, and every change is logged with a date and a reason.

Reproducible by design

One seed, version-controlled code, outputs regenerated rather than edited. Anyone can rerun the analysis and get the same numbers.

Checks before charts

The run stops with a list of problems on invalid inputs, warns if the two halves of the simulation disagree by 2% or more, and reconciles the replayed year with what was actually invoiced.

Data stays where it belongs

No client data in the repository—only synthetic example files, so the method can be shared without the numbers.

The result is a decision-support tool for the collaboration model and pricing across markets: which terms to offer, what core team they can carry, and how much risk sits with whom.

Its limits are stated as plainly as its results. With one year of project history, seasonality is treated as known; the link between a project’s size and its month is lost in resampling; automation savings are team estimates, not measurements. Each is named in the documentation, together with the data that would remove it.