Forecasting models

Historical, driver-based, or contract-based — choose how each category is projected forward.

3 min read

Every category is projected forward one of three ways. You can mix them freely — revenue on drivers, rent on a contract, everything else on history — in the same forecast.

Historical (the default)

Projected from the category's own booked history using the trailing 30-day average: the last 30 days summed and divided by 30 to get a daily rate, carried forward flat.

Good for anything steady and monthly. Poor for three things, predictably:

•Periodic payments — annual insurance, quarterly tax. The window either catches the payment and repeats it every month, or misses it and forecasts nearly nothing.

•Volume-linked costs — shipping, payment fees, cost of goods. These should move with revenue, and a flat rate means margin comes out flattering in any month revenue rises.

•Anything you are about to change — ad budget, headcount. History cannot know about a decision you have not made yet.

See the per-line detail on each category page.

Driver-based

Computed from operational numbers you enter or sync — orders, customers, ad spend. Use it where a line is caused by something you can measure, which is almost always more forecastable than the line itself. See Driver-Based Revenue.

Contract-based

The sum of the contracts and subscriptions filed under a category, each with its own amount, interval and dates. Use it wherever you already know the answer — leases, insurance, SaaS renewals, retainers. See Contracts & Subscriptions.

Choosing

The question to ask of each line is: do I know something the history does not? If the answer is no, leave it historical. If you know the schedule, use contracts. If you know the cause, use drivers.

Start historical

Connect data, let every line default to historical, and change only the lines that are visibly wrong. Modelling everything up front is how forecasts get abandoned.

PreviousAssets & unit economics Next Building scenarios