HubSpot lead scoring fails socially before it fails technically: marketing builds a score, sales does not believe it, and within a quarter everyone is back to gut feel. The fix is a score sales helped set, split into fit and intent, back-tested against real deals, and allowed to decay. Here is that build.
Why most scores get ignored
Three defects repeat in almost every portal I audit in HubSpot consulting work: one blended score that mixes company fit with click enthusiasm; points assigned by intuition and never tested against outcomes; and no decay, so a webinar attendee from last spring still looks hot. Each defect produces false positives, and every false positive spends sales trust the model cannot refund.
Fit and intent: two scores, not one
- Fit score: industry, employee count, geography, tech stack, role seniority. Static-ish; from CRM properties and enrichment.
- Intent score: pricing page views, demo requests, repeated visits, product signups, email replies (not opens). Behavioural; time-sensitive.
The most expensive routing error is low fit + high intent: enthusiastic clickers who will never buy, consuming sales hours. A blended score hides them; the grid catches them.
The HubSpot build
- Two custom score properties: Fit and Intent, not one blended number.
- Weight decisive signals decisively: pricing page and demo request should dwarf content downloads. Back-test against your last fifty won and lost deals until the score separates them.
- Negative scoring: competitors, students, careers-page visitors, free-mail domains where relevant.
- A lifecycle automation: both scores over threshold → MQL, owner assigned, SLA timer started: the handoff machinery from the lead generation system.
Decay and thresholds
Intent expires. Add decay (negative points after 30/60/90 days of inactivity) so the queue reflects buyers, not history. Set the initial threshold from the back-test, then review quarterly with one question: of the leads that crossed it, how many did sales accept? Under ~60% acceptance, tighten fit weights; near 100% with a starving pipeline, loosen intent weights. The threshold is a dial, not a monument.
The sales feedback loop
Every rejected MQL gets a reason (wrong size, wrong timing, wrong person) picked from a short list, and the reasons feed the next scoring review. This loop is the difference between a model that improves and a dashboard that ages. It is also the honest limitation: scoring cannot fix a weak ICP or thin pipeline volume. It only routes what exists. Upstream problems belong to strategy, not the score.
FAQ
How does lead scoring work in HubSpot?
HubSpot score properties add and subtract points based on contact attributes and behaviour: form fills, page views, email engagement, firmographics. The build that works separates fit (does the company match the ICP?) from intent (are they behaving like a buyer?), sets a threshold agreed with sales, and automates the handoff when both scores clear it.
What is a good lead scoring threshold?
One that sales signed. The number itself is arbitrary; the agreement is not. Start by back-testing: score your last fifty closed-won and closed-lost deals, find the level that would have separated them, and set the initial threshold there. Review quarterly against what sales actually accepted.
Why does sales ignore our lead scores?
Because the score was built by marketing alone, never back-tested, and mixes fit with enthusiasm: a student downloading three PDFs outscores a VP who visited pricing once. Rebuild with sales in the room, separate fit from intent, add score decay, and publish the definition. Trust follows accuracy, not dashboards.
Key takeaways
- Split fit from intent; route on the pair, never the sum
- Back-test weights on real won/lost deals before going live
- Decay intent after 30/60/90 days: old clicks are not buyers
- Threshold = whatever sales signed; review acceptance quarterly
- Rejected-lead reasons feed the model: that loop is the product
