tinyhumansai Financial Model
Build the model so that somebody can disagree with it precisely, rather than generally. A model that cannot be disagreed with in detail will be dismissed wholesale.
When to use
- A recommendation depends on numbers.
- A client asks what something is worth, costs, or returns.
Steps
- Separate inputs from calculation. Every number that came from outside is an input, in one place, each with its source.
- Record every input as an
assumptionsrow — the source, and honestlyestimatewhere it is one. This is not paperwork; it is what lets a falsified input be traced to everything that depended on it. - Build the base case only. Optimistic and pessimistic cases built first are decoration; they are meaningful only as sensitivities off a base.
- Run the sensitivities that matter. Vary each input and record which ones
actually move the conclusion in
sensitivity. Usually two do, and the deck spends its time on the other nine. - Find the break-even. What would have to be true for this to be a bad idea? That single sentence is worth more than the model's output.
- Sanity-check against something external — a comparable, a market size, a published benchmark. A model that agrees with nothing outside itself is a spreadsheet.
Output
A model with its inputs named and sourced, assumptions rows for each, the two
or three sensitivities that actually matter, and the break-even case stated in a
sentence.