How to Set Up Lead Scoring That Works
A practical, step-by-step process for building a lead scoring model that actually predicts which leads will close, instead of one that just looks sophisticated.
Last updated July 18, 2026
What lead scoring actually is
Lead scoring is a point system that ranks leads by how likely they are to become paying customers, so sales spends time on the leads most likely to close instead of working every inbound lead in the order it arrived. A lead earns or loses points based on two things: how closely they match your ideal customer (fit), and how actively they're engaging with your business (behavior).
The output is a single number attached to each lead record. Cross a threshold, and the lead routes to a rep. Stay below it, and the lead keeps nurturing through marketing until it earns enough points to qualify.
Example
A CRM software company might give a lead 20 points for having the title "VP of Sales," 15 points for a company size between 20 and 200 employees, and 25 points for visiting the pricing page twice in a week. A lead with all three has 60 points and routes to sales; a lead with only the job title has 20 points and stays in nurture.
Why most lead scoring models fail
Most lead scoring models fail because they're built on assumptions instead of on the company's own closed-deal data. A team guesses that webinar attendance matters more than it does, assigns it 30 points, and ends up flooding sales with leads who watched a webinar out of curiosity but never had budget or authority to buy.
The fix is to build the model backward from outcomes: look at deals that actually closed, find what those leads had in common before they closed, and score for those specific traits. A model built from your own closed-won data will always outperform one built from generic industry assumptions, because it reflects how your specific product actually gets bought.
How to split fit and behavior scoring
Fit and behavior need separate point buckets because they measure different things and decay at different rates. Fit criteria — company size, industry, job title, budget range — rarely change once you know them. Behavior criteria — email opens, page visits, demo attendance — change constantly and should be allowed to fade over time.
Mixing the two into one score creates a problem: a poor-fit lead who happens to be very active can outscore a great-fit lead who's just getting started. Keeping them as two visible components (or at least two clearly weighted categories within one score) lets a rep see at a glance whether a high score means "this is exactly who we sell to" or "this person is currently very engaged," which changes how the first call should go.
Example
A recruiting agency might score a hiring manager at a 500-person company 40 fit points on title and company size alone, even before any behavior. A candidate-side contact browsing the same site might rack up 35 behavior points from repeat visits but score 0 on fit, because they're not a buyer at all.
How to set the threshold without guessing
The sales-ready threshold should come from scoring your own closed-won deals against the new model, not from picking a round number like 100. Take the last quarter or two of closed-won deals, run them through the point system you just built, and find the score most of them had crossed by the time they became a real opportunity. Set the threshold just below that number.
This retroactive test also catches a broken model before it goes live. If closed-won deals span wildly different scores — some at 30, others at 150 — the point values aren't capturing what actually predicts a sale, and it's worth revisiting step 3 before turning the model on.
Why negative scoring matters
Negative scoring keeps a lead's score honest as their engagement and fit information change, instead of letting scores only accumulate over time. Without it, a lead who showed strong interest six months ago but has gone completely silent still shows the same score as a lead who is actively engaging today — and a rep working a list by score alone will waste time on the former.
Common negative signals include: no activity for 30, 60, or 90 days; unsubscribing from email; a personal email domain (gmail.com, yahoo.com) on a B2B lead; and a job title indicating no purchasing authority, like an intern or student role. Each of these should subtract points rather than simply being ignored.
Example
A lead that scored 70 points after downloading two whitepapers might drop to 40 points after 45 days of no further activity, correctly signaling to sales that this lead has gone cold and shouldn't be a priority for today's calls.
How often to recalibrate the model
A lead scoring model should be reviewed every 60-90 days, because the traits that predicted a closed deal last quarter can shift as the product, market, or ideal customer changes. A model set up once and never revisited slowly drifts out of sync with what's actually closing, and sales starts to distrust the scores — which defeats the purpose of building the system at all.
The recalibration process is the same as the initial setup: pull a fresh batch of closed-won and closed-lost deals, check whether the current point values still separate the two groups cleanly, and adjust weights or the threshold where they don't. Teams that skip this step tend to end up with a scoring system everyone quietly ignores within two quarters.
Keep the model visible to sales
Show reps the point breakdown behind a lead's score, not just the final number. A rep who can see "40 fit points, 15 behavior points" trusts the system and can explain to a manager why a lead is or isn't a priority — a black-box score gets ignored the first time it's wrong.