There are two kinds of HubSpot lead scoring models: the ones sales trusts, and the ones sales has learned to ignore. The second kind is far more common, and it fails for a reason that has nothing to do with HubSpot: the model was built to produce MQLs, not to predict what sales will accept.
When the score exists to hit a lead-volume target, thresholds drift downward, junk signals creep in, and within a couple of quarters the score is a number reps scroll past. The whole point of scoring — concentrating sales effort on the leads most likely to convert — quietly dies while the dashboard stays green.
Here’s how to build (or rebuild) a model that earns trust instead.
Start from acceptance, not activity
The only success metric that matters for a scoring model is this: of leads that cross the threshold, what share does sales accept and work? Not how many cross. If acceptance is high and volume is low, raise budget or widen the ICP deliberately — but don’t fix a volume problem by quietly cheapening the score. That trade is how definition drift starts.
So before touching HubSpot settings, pull two lists from the last two quarters: leads sales accepted and converted to opportunity, and leads sales rejected or ignored. Your scoring model’s job is to separate those two lists. Every criterion you’re about to set should be tested against them.
Fit and intent are different scores — treat them that way
The classic failure is a single number blending who someone is (fit) with what they’ve done (intent). A perfectly-fitting CTO who’s read one blog post and a student who’s downloaded twelve ebooks can land on the same score — which is exactly how sales learns the number means nothing.
In HubSpot, run two properties:
- Fit score — firmographic and demographic: company size, industry, seniority, geography, tech signals. This should be stable; it changes when the company’s reality changes.
- Intent score — behavioural: pricing page visits, demo requests, high-intent content, product signals. This should be volatile and, crucially, it should decay (more below).
Qualification is then a matrix, not a threshold: high fit + high intent routes to sales immediately; high fit + low intent goes to nurture; low fit + high intent gets a lightweight self-serve path or a polite pass. HubSpot’s score properties handle this cleanly with two custom score properties and lifecycle workflows keyed to both.
The five scoring sins
1. No negative scoring. Competitors, students, job seekers, and existing customers all inflate scores unless explicitly penalised. Add negative attributes for freemail domains on enterprise plays, competitor domains, careers-page visits.
1. No decay. A pricing-page visit from January is not intent in August. Use decreasing score over time on behavioural attributes (or a workflow that decrements a decay property) so intent reflects current interest.
1. Email-open worship. Opens are the weakest signal in the model — inflated by privacy proxies (Apple Mail pre-fetches images) and barely correlated with buying. Score clicks and page depth, not opens; better yet, weight the two or three pages that your accepted-lead list actually visited.
1. Every asset scores the same. A careers-page visit and a pricing-page visit are not equal events. Weight by proximity to purchase, using the accepted-lead list as evidence.
1. Set and forget. Scoring models decay like routing rules do. Quarterly, re-run the acceptance test: of last quarter’s threshold-crossers, what did sales accept? Below ~70%, adjust. This is a 30-minute ritual that keeps the score meaning something.
The classic broken model rewards engagement and calls it intent — opened three emails, visited pricing once, add fifteen points — so it happily scores a curious competitor or a job-seeker the same as a buyer. Sales stops trusting the number, works leads by gut, and the score becomes decoration. The change that earns trust back is almost always structural, not cosmetic: split the single score into fit (are they the right kind of company) and intent (are they showing buying behaviour now), and let intent decay so last quarter’s interest stops masquerading as this week’s. When a high score has to mean both, acceptance climbs — because the number finally means what sales assumed it meant.
Ship it with a contract, not an announcement
A rebuilt score fails at rollout if it arrives as a marketing announcement. It should arrive as a contract: here’s the new definition of a threshold-crossing lead, here’s the acceptance rate we’re committing to, here’s the quarterly review where sales gets to challenge the criteria. Pair it with the SLA so both sides know what happens after the score fires.
Trust in a scoring model isn’t built by the model. It’s built by the review cadence around it — the visible willingness to change the criteria when the evidence says so. HubSpot makes the mechanics easy; the discipline is the product.
—
Scoring sits inside lens two of the Pipeline Leak Audit, tested against your actual acceptance data. Unsure if scoring is your leak? Book a free health check.