Deal count sets whether aggregate math can work. When a quarter is built from two hundred deals, the law of large numbers does the work and a statistical model can forecast the total accurately without knowing anything about any single opportunity. When a quarter is built from twelve deals and three of them are half the number, no aggregate model can help, because the outcome is decided by individual events the model cannot see.
Start with deal concentration
Take the largest open deal in a typical quarter and divide it by the quarterly target. As a working rule, under roughly 5%, statistical methods behave. Bookings become a series with a stable pattern, and time series or regression approaches will forecast it well.
When a single deal is large enough to decide the quarter on its own, the forecast is a set of individual bets. Deal-level scoring, close plans, and multi-threading analysis carry the number, and the statistical model becomes a background sanity check rather than the forecast itself.
Between those cases, run both and reconcile. Most mid-market SaaS companies sit in the middle.
Match the method to how fast conditions change
The second question is whether your market holds still long enough for history to predict it. Methods that read only the past, such as run-rate projections and time series models, work while conditions hold and break when they do not.
ORM identifies stale assumptions as the leading cause of forecast misses, and names what changes: a new competitor creating pricing pressure that shrinks average deal size, rising rates slowing buyer decisions and dropping win rates, uncertainty stretching the time from qualified to closed, or a territory redesign that leaves coverage intact while execution suffers. In each case pipeline can look healthy while the forecast quietly stops being true.
If your market moves like that, weight toward methods that read current drivers. Regression on pipeline created, win rate, and deal size registers a shift in the period it happens. A model that reads only the bookings line finds out months later.
Check what history you actually have
Method choice is constrained by data before it is constrained by preference. Seasonal statistical models need two to three full cycles, which means two to three years of clean monthly data. Deal-level scoring needs consistent stage definitions across the period being learned from, not only going forward.
Data quality is a smaller constraint than most teams believe. ORM's position is direct: everyone thinks their data is uniquely bad, and it does not matter, because consistent data produces accurate predictions even when it is messy. Inconsistency is the real blocker. A stage that meant one thing last year and something else this year corrupts any model trained across the boundary.
Run two and read the gap
Whatever you pick as primary, run a second method alongside it. The gap between a deal roll-up and a statistical projection is a diagnostic no single method produces on its own. Log both raw numbers each quarter, track forecast accuracy for each separately, and after a few quarters the weighting stops being an argument. Read both against pipeline coverage, which is an input to the decision rather than the answer to it.
Frequently Asked Questions
What is the best forecasting method for an early-stage company?
A deal-level roll-up with explicit assumptions, because there is not enough history for anything statistical to learn from. Statistical methods need two to three years of clean data to estimate a seasonal pattern, which an early-stage company does not have. Focus on consistent stage definitions instead, since that history becomes the training data later.
How many deals do you need before a statistical model works?
Enough that no single deal decides the period. As a working rule, when the largest deal in a quarter is under roughly 5% of the target, aggregate statistical methods behave well. Above that, one outcome moves the total more than the model's precision, and deal-level analysis has to carry the forecast.
Should you use more than one forecasting method?
Yes. Run a deal-level method and a statistical method side by side and the gap between them becomes a diagnostic. Agreement is confirmation. Disagreement points at either optimistic deal calls or a changed condition the history has not absorbed, and both are worth knowing before the quarter ends.
When should you replace your forecasting method?
When accuracy degrades for two consecutive quarters, or immediately after a structural change such as a territory redesign, a pricing change, or a move upmarket. Both cases mean the method is running on assumptions the business no longer matches.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like how do you choose a sales forecasting method? into prescriptive action for your team.
Schedule a Demo