How to Build a Revenue Forecast With No Historical Data

The standard advice for building a forecast starts with pulling twelve months of closed opportunities. That advice is useless if you have been selling for two quarters and closed nine deals. The situation is common at early-stage B2B SaaS companies and after entering a new segment, and it does not remove the obligation to produce a number. It changes how the number gets built and how it should be presented.

Can you forecast revenue without historical data?

Yes, by borrowing rates you cannot calculate and being explicit about which ones are borrowed.

The output will be a range rather than a point, and that is correct. A forecast built on nine closed deals that arrives as a single confident number is a fabrication with a spreadsheet attached.

What makes the exercise worth doing is that a bottom-up build is falsifiable. When the quarter comes in low, a model made of deal counts and assumed conversion rates shows you exactly which assumption broke. A top-down number derived from market size and a capture rate teaches you nothing when it misses, because there is no line item to inspect.

What do you use instead of your own conversion rates?

Proxies, ranked by how much they share with the business you are actually running. The ranking is about similarity of sales motion, not sample size. Your own eleven closed mid-market deals are a better guide to your mid-market win rate than a thousand-company benchmark that blends self-serve, inside sales, and field motions into one average. Benchmarks describe a population you are not part of.

Whichever proxy you use, record it as a proxy. An assumption cell with a note reading "borrowed from SMB segment, replace at 30 closed deals" survives review. A bare percentage does not, and within a quarter nobody remembers it was a guess.

How do you build the first forecast?

Four inputs, multiplied, with a range on each.

Start with qualified opportunity count. This is the one number you probably do know, because you can count the deals in front of your sellers today plus what your demand generation has been producing per month. Count is more reliable than value at this stage.

Multiply by an assumed win rate from your proxy source, expressed as a range rather than a point. If the proxy suggests 22 percent, model 16 to 28 percent.

Multiply by average deal size, taken from actual closed contracts. Never use pipeline amounts. Deals close for less than their recorded value, and at low deal volume one optimistic pipeline entry distorts the entire model.

Then apply the timing gate. Measure created-to-closed on the deals you have won, use the median, and check it against the calendar. Pipeline that must close this quarter had to exist by a date that has probably already passed. This single check kills more early-stage forecasts than any conversion assumption, and finding out in week two beats finding out in week eleven.

For the structural walkthrough of these steps at any data maturity, see how to create a sales forecast.

How fast should borrowed assumptions get replaced?

Recalculate every month and swap each proxy the moment you have 30 closed opportunities in that segment.

Thirty is a working threshold rather than a statistical one. Below it, one unusual deal moves the rate several points. Above it, the rate is directional enough to act on, and holding out for a more comfortable sample means shipping a borrowed number for another three quarters.

Watch for stabilization rather than just recalculating. Plot the running win rate as deals accumulate. When adding five more closed deals stops moving the number by more than a point or two, the rate has settled. Until then, keep the range wide in the forecast.

Replace assumptions one at a time, in the order of their impact on the output. Win rate usually moves the number most, deal size second, cycle length third.

How do you present a forecast built on thin data?

As a range, with the assumptions that create its width named on the same slide.

The presentation format matters more here than in a mature forecast, because the audience will treat a single number as a commitment regardless of the caveats spoken aloud. Show the low and high, then show which two assumptions produce most of the spread.

That framing converts the meeting from a negotiation about the number into a decision about what to go learn. When the spread is driven mostly by win rate uncertainty, the next quarter's priority is closing enough deals to observe the real rate, which is an executable answer.

Avoid presenting probability-weighted midpoints as though they were forecasts. A midpoint from two weak assumptions carries the confidence of neither and hides both.

When does the model become trustworthy?

Once every borrowed input has been replaced and the model has been checked against actuals for two or three full periods.

The replacement is the first gate. The second is variance tracking. Compare what the model said at the start of each period against what closed, by line rather than in total, and record the gaps. A model that hits the total while overstating win rate and understating deal size is not working, and the errors will stop cancelling at some point.

Models trained on a company's own historical sales performance need real data volume behind them, which is why fully trained forecasting models take 4 to 6 weeks to build once the history exists. The constraint at an early-stage company is not modeling time, it is accumulated closed deals. Until you have them, the forecast is a structured hypothesis, and treating it as anything more is where the damage happens. Grounding in the fundamentals of sales forecasting and calculating a real win rate as soon as volume allows are the two moves that shorten the wait.