# Predictive CRM Funnel Attribution > CRM Funnel Attribution shows which marketing touches create pipeline. Predictive CRM tells you how much of that pipeline will close — and when. CRM Funnel Attribution shows which marketing touches create pipeline. Predictive CRM tells you how much of that pipeline will close — and when. *10 min read · March 2026* ## 01 — The Deepest Delay in Marketing Measurement [CRM Funnel Attribution](/measurement-engine/crm-funnel-attribution) connects every pipeline stage — from Lead to Closed/Won — back to the original marketing click. It answers which channels create pipeline. But pipeline is not revenue until deals close, and B2B sales cycles run 6 to 12 months. The delay problem from [Predictive Attribution](/measurement-engine/predictive-attribution) is worse here — far worse. In e-commerce, a click converts in days or weeks. In B2B, a click becomes a lead in hours, an MQL in days, an opportunity in weeks, and a deal in months. Every stage has its own lag. Recent cohort data is not just incomplete — it is deeply incomplete. ### The compounding gap Consider a January LinkedIn campaign that generated 50 leads. By March, 20 have become MQLs. By May, 8 are opportunities. By September, 3 have closed at an average deal size of $45K. The campaign produced $135K in revenue — but in January, when you were deciding whether to keep running it, you saw 50 leads at $80 CPL and nothing else. Leads from recent clicks have not become opportunities yet. Opportunities have not become deals. Each stage takes weeks or months to materialize. If you evaluate channels on recent cohort data alone, every campaign looks like it produces leads and nothing else — because the downstream stages have not had time to happen. You cannot wait 9 months to evaluate a campaign. You also cannot optimize on leads alone — the [CRM Funnel Attribution](/measurement-engine/crm-funnel-attribution) page shows why channel rankings flip at every stage. The channels that produce the cheapest leads are not the channels that produce revenue. > CRM Funnel Attribution shows which marketing touches create pipeline. Predictive CRM Attribution tells you how much of that pipeline will close — and when. --- ## 02 — Multi-Stage Prediction Unlike e-commerce prediction, where a single maturation curve projects click-to-conversion, CRM prediction models a sequence of stage transitions. Each transition has its own timing, its own conversion rate, and its own model — because progression patterns differ at every stage. ### Stage-by-stage models The system trains separate ML models for each funnel transition: - **Lead → MQL** — typically days. High conversion rate, fast feedback. The model learns which lead characteristics (source channel, company size, form type) predict qualification. - **MQL → Opportunity** — typically weeks. Conversion rate drops. The model incorporates sales engagement signals alongside marketing data. - **Opportunity → Closed/Won** — typically months. Lowest conversion rate, highest value. The model weighs deal stage progression speed, deal size, and competitive dynamics. > **[Illustration: CRM Pipeline Prediction]** > Three-stage prediction diagram. Each funnel transition has its own model with distinct progression rates and timing. Each model produces two outputs: a probability of progression (will this lead become an opportunity?) and a predicted value (what will the deal be worth?). Combined, they generate expected pipeline value per channel, per campaign — weeks or months before the deals actually close. ### Why separate models matter A single end-to-end model (lead directly to Closed/Won) would be simpler but worse. The signals that predict MQL qualification are different from the signals that predict deal close. Company size matters for opportunity creation. Deal stage velocity matters for close prediction. Collapsing them into one model blurs distinct patterns. Separate models also give you visibility into where the funnel is breaking. If Lead → MQL conversion is strong but MQL → Opportunity is weak, the problem is not marketing — it is sales qualification or handoff. A single model hides this. --- ## 03 — 365-Day Attribution Window Standard attribution windows are 30 or 90 days. For B2B pipelines, this is not enough. A deal that closes in September needs to attribute back to a click in January. A 90-day window anchored on the close date cannot reach the original marketing touch. ### Why long lookback matters The [dual-date system](/measurement-engine/crm-funnel-attribution#dual-date) in CRM Funnel Attribution partially solves this by anchoring the attribution window on the lead creation date, not the close date. But even with dual-date anchoring, a 90-day window can miss early touchpoints for leads that took weeks to fill out a form. Predictive CRM Attribution extends the attribution window to 365 days. A deal won in Q4 can attribute back to awareness campaigns from Q1 of the same year. Enterprise sales cycles that span the full fiscal year are covered without window compromises. > **[Illustration: CRM Attribution Window]** > Timeline showing how a 90-day window misses the original marketing click for long sales cycles, while a 365-day window captures the full journey. ### The tradeoff Longer windows increase the risk of false attribution — crediting a click from 11 months ago that may not have influenced the deal. The 365-day window is available, not mandatory. Each project configures the window based on its sales cycle length. A SaaS company with a 45-day sales cycle might use 180 days. An enterprise vendor with 9-month cycles might use the full 365. The window applies per conversion type. Lead attribution might use 90 days (leads arrive fast). Closed/Won attribution might use 365 days (deals close slow). Each stage gets the window that matches its timing. --- ## 04 — Seasonal Adaptation B2B pipeline progression follows business rhythms, not calendar seasons. The system adapts its predictions to these patterns automatically. ### Q4 budget pressure Fiscal year-end drives faster pipeline progression. Buyers accelerate procurement to use remaining budget. Deals that normally take 4 months close in 6 weeks. The prediction models learn this — Q4 opportunities get higher close probabilities and shorter projected timelines because historical Q4 cohorts progressed faster. ### Q1 new fiscal year New budget cycles mean longer approval chains. Decision-makers reassess vendor priorities. Deals stall. Pipeline that looked healthy in December goes quiet in January. The system learns this pattern per client — Q1 opportunities get lower close probabilities and longer projected timelines. ### Industry-specific cycles Not every business follows the fiscal calendar. Retail vendors sell to buyers in spring for fall inventory. Education companies close deals in summer for September launches. The system does not assume a universal cycle — it learns each client's progression patterns from their historical data. > **[Illustration: CRM Seasonal Patterns]** > Quarterly progression rates showing Q4 deals closing faster due to budget deadlines and Q1 deals decelerating as new budgets are approved. The adaptation happens automatically as the models retrain on new data. No manual configuration. No calendar rules. The system observes progression velocity and adjusts projections accordingly. --- ## 05 — Projected Pipeline by Channel Predictive CRM Attribution adds a forward-looking layer to every CRM funnel report. Instead of seeing only pipeline that has already progressed, you see projected pipeline value — how much of today's open pipeline is predicted to close and what it will be worth. ### Channel-level projections For each channel: current leads, current opportunities, projected opportunities (from leads that have not yet progressed), current pipeline value, and projected closed-won value. The projection answers the question every B2B marketer asks: "How much pipeline is my Meta prospecting generating that has not closed yet?" ### Campaign-level visibility Drill from channel to campaign to individual lead. A LinkedIn campaign generated 50 leads. 20 are MQLs. The system projects that 7 will become opportunities and 2.5 will close at an average deal value of $42K — projected pipeline of $105K from a campaign that currently shows zero revenue. This changes budget decisions. Without prediction, the LinkedIn campaign looks like it produced 50 leads at $80 CPL — expensive compared to display. With prediction, it shows $105K in projected pipeline at a $4K acquisition cost — a projected ROAS the display campaign cannot match. ### Time-to-close projections Beyond value, the system projects when pipeline will close. An opportunity created in March with a 65% close probability is projected to close in June. This gives finance teams a forward view of revenue — not just a backward view of what already closed. --- ## 06 — Validation Predictive CRM Attribution validates itself the same way [Predictive Attribution](/measurement-engine/predictive-attribution#validation) does: by backtesting projections against actual outcomes once enough time has passed. ### Projected vs actual closed-won revenue The most important validation: compare projected pipeline value from 6 months ago against actual closed-won revenue today. If the system projected $500K in pipeline from a Q1 cohort and $480K actually closed by Q3, the projection error is 4%. If it projected $500K and $200K closed, the model needs recalibration. ### Stage-level accuracy Each stage transition is validated independently. Lead → MQL prediction accuracy might be 85%. MQL → Opportunity might be 70%. Opportunity → Closed/Won might be 60%. Accuracy naturally decreases with pipeline depth because more unknowns accumulate. The system exposes this transparency — you see which stage the prediction is strong at and which stage introduces uncertainty. ### Continuous recalibration As deals close (or do not close), the training data grows. The models retrain on updated outcomes. A model trained on 200 historical Closed/Won deals performs differently than one trained on 50. Over time, the projections improve as the system accumulates more ground truth from your specific pipeline. > CRM prediction accuracy improves with time. Every deal that closes — or does not close — makes the next projection more accurate. The system learns your pipeline, not a generic one.