Most lead scoring models measure the wrong thing well. They track form fills, email opens, and page visits with precision, then add them up into a number that's supposed to mean "sales-ready" — but activity isn't intent, and a model built only on activity tends to hand sales a pile of downloaded-the-ebook leads that go nowhere. Building a model that actually predicts readiness takes a different starting point.
Start from closed-won, not from the funnel
The common mistake is designing a scoring model around the stages marketing thinks matter — visited pricing, downloaded a guide, attended a webinar — without ever checking whether those actions actually preceded the deals that closed. The better starting point is to pull your last 12–24 months of closed-won deals and reverse-engineer what those accounts actually did before they became opportunities. Sometimes the strongest predictor is a channel nobody weighted heavily, and sometimes an action everyone assumed mattered — like a whitepaper download — barely correlates with anything.
Weight fit and behavior separately
A lead score is really two different questions mashed into one number: does this account fit our ICP, and is this account behaving like it's ready to buy? Collapsing both into a single score hides useful information. A model that scores fit and behavior separately lets you route leads more precisely — a perfect-fit account showing early research behavior gets a different treatment than a poor-fit account with high activity, even if their combined scores land in the same range.
Build in decay
Intent isn't permanent. An account that engaged heavily two months ago and has gone quiet since is a worse bet than one that engaged moderately this week. Most static scoring models don't account for this — points accumulate and never disappear, so an account can sit at a "hot" score long after it's actually cooled off. Adding time decay to behavioral points, so older activity contributes less than recent activity, keeps the score closer to reality.
Validate against what sales actually says
The model isn't finished at launch. The fastest way to tell whether it's working is to sit with the sales team monthly and compare: which "sales-ready" leads did they actually want, and which ones did they immediately disqualify? Patterns show up quickly — certain firmographic combinations that score well but never convert, certain lower-scoring signals that sales says are actually strong indicators. Feed that back into the weighting rather than treating the initial model as fixed.
The bottom line
A lead scoring model earns trust from sales by being right more often than it's wrong, and it stays right by being checked against real outcomes, not left running on assumptions from the day it launched. Score fit and intent separately, weight recent behavior over old behavior, and validate the whole thing against what closed — not just what happened along the way.
