The skill: Forecasting turns what you know today into a range of plausible futures. The goal isn't to predict the number exactly. It's to make decisions under uncertainty with your eyes open.
In a Nutshell
- Ranges, not points. "We'll be between $800K and $1.2M ARR by December" is honest. "$1M ARR by December" is a guess pretending to be a fact.
- Three scenarios minimum. Bull case (things go right), base case (things continue as-is), bear case (things go wrong). If you can only present one, present the base case.
- Linear extrapolation works for short windows. If you're forecasting next month based on the last three months, a straight line is fine. For anything beyond a quarter, you need something smarter.
- Cohort-based projections beat top-down for SaaS. Model each monthly cohort's revenue contribution over time, then stack them. This captures churn and expansion naturally instead of hand-waving them away.
- Base rates are your sanity check. If your forecast implies you'll grow 4x this year, but the median Series A company grows 2x, your assumptions need justification.
- State every assumption. "This assumes we maintain 5% monthly growth, 3% monthly churn, and hire two more salespeople in Q2." Now someone can disagree with specific assumptions instead of the whole forecast.
- Update monthly. Compare forecast to actuals. Adjust. The forecast isn't a commitment, it's a living model.
Three Methods Worth Knowing
Linear extrapolation is the simplest. Take your monthly growth rate over the last 3-6 months, project it forward. This works for short-term forecasts (1-2 months) where conditions are stable. It breaks down quickly for compounding metrics or anything beyond a quarter. If you grew 8% per month for the last three months, projecting 8% next month is reasonable. Projecting 8% for the next twelve months is fantasy, because that implies 2.5x growth and ignores the ceiling effects that every company hits.
Cohort-based projection is the right method for recurring revenue. Model each cohort of customers separately: how many signed up, what they pay, how they expand or churn over time. Stack all the cohorts together to get total revenue. This method forces you to be explicit about retention and expansion rates instead of burying them inside a single growth number. If your churn rate is 5% monthly but you're forecasting 3x ARR growth, the cohort model will expose the contradiction.
Base rate adjustment is a calibration tool, not a forecasting method. Before you finalize any forecast, check it against base rates. What does the typical company at your stage achieve? If your Series A forecast implies top-decile growth, you need a specific explanation for why you're different. Not just optimism, but a structural reason: a distribution advantage, a product moat, a market tailwind. Without that, adjust toward the base rate.
The Forecast Is Not a Promise
A forecast is a model, not a commitment. When the forecast becomes a target people are held to, they stop updating it honestly. They fudge inputs to make the number look achievable. They stop reporting bad news early.
A good forecasting culture separates the model from the target. The model says "here's what we expect given current trends and assumptions." The target says "here's what we're aiming for." The two should be close, but they serve different purposes. When actuals diverge from the forecast, that's information, not failure. The question is always: what changed, and do we need to adjust the model or the plan?
Do's and Don'ts
Written with ❤️ by a human (still)