Most B2B lead scoring is theater. Marketing assigns points for "downloaded whitepaper" (10 points), "visited pricing page" (15 points), "opened email" (5 points). The numbers add up. The leads still suck.
The reason it doesn't work. Most scoring models are built backward — from what marketing wants to measure, not from what predicts closed-won. We've rebuilt scoring for a dozen B2B clients in the last year. In every case, the actual predictors of closed-won were:
(1) Company stage and headcount fit. Pre-seed prospects don't close on enterprise pricing. Post-Series B prospects don't have time for SMB-priced products. Score for fit before behavior.
(2) Pricing page time, not pricing page visits. A 3-second bounce off /pricing means nothing. 90 seconds on /pricing followed by /enterprise means the prospect is sizing budget.
(3) Multiple stakeholders from the same domain in a 14-day window. One person visiting your site 8 times is a researcher. Three people from the same company visiting once each is a buying committee.
(4) Direct traffic to deep pages. If someone shows up at /case-studies/your-best-case without going through your home page, they've been recommended. That signal is worth 50 "opened email" points.
The model we ship. A weighted score across fit (50%) + behavior (30%) + intent signals (20%). The threshold for sales handoff is calibrated by analyzing 90 days of closed-won and closed-lost. The model gets retrained quarterly.
The output: sales talks to fewer leads, closes more deals. Marketing stops being graded on MQL volume.