What Assumptions Belong in a Revenue Forecast Model

Every forecast is an assumption set with arithmetic wrapped around it. The arithmetic is rarely the problem. The problem is that half the assumptions are undocumented, most were set once and never revisited, and nobody remembers which of them the number actually depends on. This guide covers the assumption set a B2B SaaS forecast needs, how to source each value, and which ones decay fastest.

Which assumptions does a revenue forecast model need?

Nine, covering conversion, value, timing, capacity, and retention. Two entries here cause most of the trouble. Average deal size must come from closed-won values, because pipeline amounts run high. A business showing $80,000 average pipeline deals against $40,000 average closed-won deals will model twice the new business it can deliver if it uses the pipeline figure, and nothing in the output signals the error.

Per-rep productivity must come from actual attainment. Building the model on quota assumes every rep hits plan, which encodes the miss before the quarter starts. If the team delivered 80 percent of quota last year, the assumption is 80 percent of quota.

Where should assumptions live in the model?

On their own tab, one cell per assumption, referenced by every calculation and typed into none.

The test is mechanical. Search the calculation layer for hardcoded numbers. Every one you find is an undocumented assumption that will outlive the person who typed it.

Each assumption cell needs four attributes beside it: the value, the calculation or query that produced it, the date range of the underlying data, and the date it was last refreshed. That last field is the one that gets skipped and the one that matters most in month seven, when someone asks whether the win rate in the model reflects the pricing change from April.

Structure this way and updating an assumption is one edit. Structure it any other way and updating an assumption is a project, which means it does not happen.

How do you set conversion and timing assumptions?

Calculate both from closed opportunities only, segmented, and check the trend rather than accepting the average.

For win rate, divide won opportunities by the sum of won and lost over the trailing twelve months. Open deals are excluded, which is where most win rate calculations break. Run it separately by segment, because enterprise and SMB conversion behavior are unrelated, and a blended rate describes neither.

Then look at the four quarters inside that twelve months. An assumption of 22 percent built from quarters that ran 26, 24, 20, and 18 percent is describing a business that no longer exists. Use the recent trend, and note in the documentation that the rate is declining.

For cycle length, measure created date to closed date on won deals and use the median. Means get destroyed by the single enterprise deal that took two years. Segment this too, and use it as the timing gate on pipeline creation. If mid-market deals take 90 days, pipeline that must close this quarter had to exist last quarter, and no assumption about heroic effort changes that arithmetic.

Which assumptions go stale fastest?

Win rate, deal size, and cycle length, and they usually move together.

Forecasts miss because something changed in the business or the market and the model was still running on old assumptions. The changes arrive in recognizable shapes. A new competitor enters and creates pricing pressure, so average deal size falls. Interest rates rise, private equity capital slows, valuations compress, buyers cut costs, and win rates decline. Broad uncertainty makes buying committees hesitate, so deals stretch from qualified to closed.

Each of those events touches more than one assumption. That is why flexing assumptions independently understates the damage. A model that assumes win rate can fall 10 percent while deal size and cycle length hold constant is describing a scenario that does not occur.

One more assumption ages quietly: territory stability. When territories get redrawn, sellers are distracted and execution suffers even while pipeline looks healthy. Coverage holds, the model looks fine, and the quarter still misses. Track deal slippage through the transition, since close date changes surface the problem before the revenue does.

How do you handle assumptions you cannot source?

Label them as estimates, bound them with a range, and set a date to replace them.

Some assumptions have no historical basis. A new segment, a new product, a new region. Borrow a value from the closest analogous part of your business rather than from an industry benchmark, since your own adjacent segment shares your sales motion and your pricing.

Attach a replacement trigger. After 30 closed opportunities in the new segment, the borrowed rate gets replaced with the observed one. Thirty is enough to be directional even though it is not statistically comfortable, and waiting for comfort means running on a borrowed number for a year.

What happens when an assumption is wrong?

Update it, restate the affected periods, and log the correction with its size and direction.

The log is the point. One correction is noise. Four consecutive corrections in the same direction on the same assumption is a structural bias, and structural bias is fixable once you can see it. Teams that quietly overwrite assumptions never accumulate the evidence, so they keep making the same error with a fresh number each quarter.

Review the assumption set on a fixed cadence rather than only when a forecast misses. Reviewing after a miss guarantees you fix the assumption that was wrong this time and leave the others aging. The wider discipline around this is covered in sales forecasting best practices, and the measurement side in forecast accuracy.