How to Forecast Revenue Bottom-Up (Instead of Guessing a Growth Rate)

Most founder forecasts start with last year's revenue and a growth rate. It takes a minute to build and tells you nothing, because when the number is missed you cannot say which assumption broke. A bottom-up forecast starts from the handful of drivers you actually influence, so a miss points straight at the cause.

The driver stack

For a DTC or e-commerce brand:

New revenue = sessions × conversion rate × average order value Repeat revenue = active customers × orders per customer per period × average order value Total = new + repeat − refunds and discounts

For subscription or SaaS:

Revenue = opening MRR + new MRR + expansion − contraction − churned MRR

Every line above is a driver someone owns. Sessions belong to marketing, conversion to the site, AOV to merchandising and price, repeat rate to lifecycle and product.

Build it in this order

1. Split new from repeat. Blending them hides the only trend that matters – whether your existing customers are coming back more or less often than a year ago.

2. Drive paid traffic from spend, not from a growth rate. Sessions from paid = spend ÷ cost per click, then apply the conversion rate that channel actually produces. That way the forecast breaks correctly when you cut the budget.

3. Use cohorts for repeat revenue. Take the last 12 monthly cohorts, compute orders per customer by month since first purchase, and apply that curve forward. It is unglamorous arithmetic and by far the most reliable part of the model.

4. Keep seasonality explicit. A monthly index by month of year, based on two or three years of your own history, applied as a multiplier. Never bake seasonality into the growth rate where nobody can see it.

5. Model price and discount separately. Gross price, discount rate, then net. Blending them means a discount-driven revenue rise looks like demand.

The review loop that makes it useful

Every month, put forecast next to actual for each driver – not just revenue – and record the variance. Two or three cycles of that and you know which assumptions are optimistic and by how much. That correction factor is worth more than any refinement to the model structure.

A forecast is not a promise about the future; it is a statement of what has to be true for the plan to work. Built bottom-up, it tells you within a month which of those things is not true yet – while there is still time to do something about it.