Pipeline forecasting that survives the quarter
The standard pipeline forecast multiplies each open deal by a probability attached to its stage — 20% at discovery, 60% at proposal — and adds up the results. It produces a single confident number, and it is wrong in the same direction almost every quarter. The reason is simple: stage probabilities are averages across your whole pipeline, and your pipeline is not homogeneous.
Deals carry their origin with them
Two deals sitting in the same proposal stage are not the same deal. One came from a referral: the buyer arrived with trust already established, the evaluation is short, and the close rate from proposal might be seventy percent. The other came from a cold outbound sequence: the champion is real but the organization is unconvinced, and the honest close rate from the same stage is half that. A stage-weighted model prices both identically, which means it is systematically optimistic about one and pessimistic about the other.
Source is the strongest signal most forecasting models throw away. In every pipeline we have measured, conversion from qualified to closed-won varies more by originating channel than by any other observable property of the deal — more than deal size, more than segment, more than rep. Time-to-close varies just as much: search-originated deals move fast because the buyer was already in motion; content- and social-originated deals move slowly and then close at a higher rate.
Cohorts by source
The fix is to forecast the way actuaries price risk: by cohort. Group deals by originating channel and entry month, measure each cohort's historical stage-to-stage conversion and velocity, and apply those rates — not the global averages — to today's open pipeline. The forecast becomes a sum of small, well-estimated numbers instead of one large, poorly estimated one.
This has a prerequisite that trips up most teams: you have to know the originating channel reliably, all the way from the first anonymous website visit through to the CRM opportunity. If a third of your deals are labeled "Direct" or "Unknown," cohort rates are noise. Closing that loop — carrying first-touch attribution into the CRM, and CRM outcomes back into analytics — is not a marketing vanity project. It is the data foundation the forecast stands on.
What changes when you do this
Three things, in our experience. The forecast stops whipsawing mid-quarter, because cohort rates are stable even when individual deals are not. Pipeline coverage targets become channel-specific — you learn that 3x coverage of referral pipeline and 6x coverage of outbound pipeline are the same amount of expected revenue. And the marketing budget conversation changes, because you can now state, with history behind it, what a dollar into each channel returns in closed revenue and how many months later it arrives.
A forecast is a claim about the future backed by structure in the past. Stage-weighting assumes the structure lives in your sales process. Mostly, it lives in where the deal came from.