What is the difference between cohort-based and pipeline-based forecasting?
Pipeline forecasting projects revenue from deals that exist right now. Cohort forecasting projects it from how customers who joined at a given time have behaved since. One reads the CRM. The other reads the billing history.Pipeline methods need an opportunity record with an amount, a stage, and a close date. Take that record away and the method has nothing to work with. Cohort methods need none of that. They look at what the January 2025 cohort was worth at month three, month six, and month twelve, and apply the same shape to the cohort currently at month three.
For most B2B SaaS companies both methods are correct at the same time, applied to different halves of the revenue base. The mistake is picking one and forcing the whole business through it.
Which revenue streams belong to each method?
Assign the stream by asking whether a rep has to do something for the revenue to arrive.| Revenue stream | Right method | Why |
|---|---|---|
| New logo bookings | Pipeline | Every dollar has a deal record and a close date |
| Competitive expansion or upsell | Pipeline | Worked as an opportunity with an owner |
| Contractual renewal | Cohort | Arrives on schedule unless something breaks |
| Seat and usage growth | Cohort | Accumulates without a sales cycle |
| Contraction and churn | Cohort | Shows up as a behavior pattern, not a deal |
| Price increases at renewal | Cohort with an override | Predictable timing, deliberate size |
Running the wrong method on a stream produces a specific kind of error. Forecasting renewals through pipeline turns a predictable base into a stack of low-confidence opportunities that clog every deal review. Forecasting new business through cohorts assumes next quarter looks like last quarter, which is only true until pipeline creation moves.
How does a cohort forecast actually get built?
Group customers by start period, measure what each group is worth at each month of age, then apply those curves forward to younger cohorts.The mechanics run in four steps. Group every customer by the month their contract started. Measure retained and expanded ARR for each group at every subsequent month. Average across groups to get a survival and expansion curve by age. Apply the curve to cohorts that have not reached that age yet.
The output is a monthly ARR waterfall, and the waterfall is where the method pays off. At ORM we reconcile it by month: beginning ARR, then the contraction lines for churned customers, churned products, and product decreases, then the expansion lines for new customers, new products, and product increases, ending in the ARR that becomes next month's opening balance. Net revenue retention and gross revenue retention sit on the same chart, calculated from those same lines rather than from a separate query.
That reconciliation is what makes the forecast trustworthy. Every dollar of movement has a named cause, so when the forecast is wrong you can point at which line broke.
What does pipeline forecasting see that cohorts cannot?
Change. A cohort curve is a statement that the future will rhyme with the past. It is a strong assumption for a subscription base and a weak one for anything competitive.Pipeline forecasting sees a competitor entering an account, a renewal negotiation going sideways, a champion leaving, a deal that has been resized twice. Those are visible in opportunity records before they are visible in a retention curve, because a cohort only reveals the damage once the revenue has already failed to arrive.
The reverse is also true, which is the honest case for cohorts. Pipeline forecasting sees nothing until an opportunity is created. Expansion revenue that arrives through usage growth never becomes a record, so a pipeline-only view of the business understates the quarter by exactly the amount that happens without a seller.
Which leading indicators improve a cohort forecast?
Product and support behavior, because they move before the renewal date does.Support ticket volume is one of the more useful signals, and it does not move in the direction most people expect. A customer with no support cases at all is at risk, since silence normally means nobody is using the product. A customer with seven or more cases in a year is also at risk, because that volume usually means the implementation is fighting them. The healthy middle sits around three to five cases, typically tier two or three rather than severe, which describes a customer who is engaged and getting help.
Feed those signals into the cohort model as segment splits rather than as a single blended curve. A cohort of accounts in the healthy support band and a cohort in the silent band do not retain at the same rate, and averaging them produces a curve that describes neither.
How do you run both methods without double counting?
Give every dollar exactly one home, and reconcile the two forecasts to a single total before anyone sees it.The double-count risk is concentrated in expansion. An upsell can appear as an open opportunity in the CRM and also sit inside the expansion curve of its cohort, and both models will claim it. Fix the rule at the source. Any expansion carrying an opportunity record is forecast through pipeline and is excluded from the cohort expansion term. Everything else is cohort.
Then reconcile. New business from pipeline plus renewal and organic expansion from cohorts equals total forecast ARR. The reconciliation should tie to the same monthly waterfall you use for retention reporting, so finance and RevOps are reading one set of numbers.
Both halves still need the same discipline. A cohort curve fitted before a pricing change describes a company that no longer exists, and a pipeline forecast built on close dates inherits every slipped date in the CRM. Refit the curves each quarter, keep the sales forecasting side decomposed rather than aggregated, and read how to forecast revenue for the order the pieces go together.
Frequently Asked Questions
What is cohort-based revenue forecasting?
Cohort-based forecasting groups customers by when they started, then projects future revenue from how earlier cohorts behaved at the same age. If customers who signed in month one retained 88 percent and expanded 12 percent by month twelve, that curve becomes the forecast for the cohort currently at month six. The forecast comes from customer behavior over time rather than from open opportunities.
When should you use pipeline forecasting instead of cohorts?
Use pipeline forecasting for new business and for any expansion that runs through a real sales process with a named opportunity, an owner, and a close date. Pipeline methods work when a deal exists as a record you can inspect. Cohort methods work when revenue arrives through renewal, usage growth, or seat expansion that no rep is actively working.
Can you use both methods at once without double counting?
Yes, by assigning each revenue stream to exactly one method. New logo and any expansion carrying an opportunity record belong to pipeline. Renewal, contraction, churn, and automatic usage expansion belong to cohorts. The failure case is an upsell forecast by a cohort curve while the same upsell also sits in the pipeline as an open deal, which inflates the total by counting the same dollars twice.
How many periods of history do cohort forecasts need?
You need cohorts old enough to cover the horizon you are forecasting. Predicting month twenty-four behavior requires cohorts that have reached month twenty-four, so a company with two years of history can forecast a year out with confidence and is extrapolating beyond that. Recent cohorts also behave differently from old ones after a pricing or packaging change, so weight recent curves more heavily.
What signals predict churn before a renewal cohort shows it?
Support ticket volume is one of the most useful. A customer filing no support cases at all is at risk, because silence usually means nobody is using the product. Seven or more cases in a year is also a risk signal. Three to five cases, typically tier two or three rather than severe, tends to mark an engaged customer who is getting help and is less likely to leave.
See how ORM turns these insights into action
ORM builds custom revenue forecast models for B2B SaaS companies. Not dashboards. Prescriptive analytics that tell you what to do next.
Schedule a Demo