TL;DR Lead scoring fails when it's built as a marketing exercise instead of a sales tool. The model that actually gets used separates fit from engagement, decays stale signals, subtracts points for disqualifiers, sets thresholds that mean something to the people making calls, and recalibrates quarterly based on closed-won data.
Every HubSpot portal above the Professional tier has lead scoring available. Very few portals have a scoring model that sales actually checks before making a call.
The gap is almost always the same: the model was built by marketing, for marketing, using marketing logic. It counts form fills and email opens. It doesn't distinguish between a VP of Finance at a target account and an intern downloading an ebook for a class project. Both get the same score. Sales notices. Sales stops looking.
The deeper issue is architectural. A single number that blends "who someone is" with "what they've done" produces scores that are impossible to interpret. A 75 could mean a perfect-fit contact who hasn't engaged yet, or a terrible-fit contact who's clicked everything you've ever sent. Sales can't act on ambiguity.
Lead scoring isn't a marketing feature. It's pipeline infrastructure. If sales doesn't trust it, it doesn't exist.
Fit measures who they are: job title, company size, industry, revenue, region. Static, no decay. Engagement measures what they've done: page views, form submissions, email clicks, meetings. Dynamic, decay-enabled. HubSpot's combined score keeps both legible in an A1 to C3 matrix, so a rep reading A2 knows strong fit, moderate engagement, worth a call.
HubSpot's current lead scoring tool supports three score types: engagement-only, fit-only, and combined. The combined score is where the model gets useful. It evaluates fit and engagement as separate dimensions and maps them into an A1–C3 matrix.
Fit measures who they are. Engagement measures what they've done. The combined score tells you both at once, but keeps them legible.
Evaluates demographic and firmographic properties: job title, company size, industry, annual revenue, region. Answers the question: "Is this someone we should be talking to?" Uses property groups. Static. No decay.
Evaluates behavioral events: page views, form submissions, email clicks, meeting bookings, CTA interactions. Answers the question: "Are they paying attention right now?" Uses event groups. Dynamic. Decay-enabled.
This separation is what makes the model actionable. A sales rep glancing at the CRM card can immediately see: A2 means strong fit, moderate engagement, worth a call. C1 means poor fit, high engagement, probably not a buyer, possibly a researcher or competitor. That signal clarity is what earns trust.
The fit score is your ICP encoded as a scoring model. It should reflect the characteristics of contacts who have actually become customers, not the contacts you wish you had.
A key distinction: negative scoring and exclusion lists serve different purposes. Negative scoring adjusts the score for contacts who might be a fit but have disqualifying attributes. Exclusion lists remove contacts who should never be evaluated at all. If someone should never qualify regardless of their behavior, exclude them. Don't try to subtract their way to zero.
Score behavioral events by intent, not volume: a demo request outweighs a newsletter signup, pricing page views outweigh blog reads, and email opens don't count at all. Enable decay on every event group so points fade over one to six months, and cap how much any single event type can contribute. Without decay, the score is a museum of past activity.
Engagement scoring captures behavioral signals. But here's where most models go wrong: they treat a form submission from last week the same as one from eight months ago. Without decay, your scoring model is a museum of past activity, not a signal of current intent.
HubSpot's lead scoring tool lets you enable score decay per event group, with intervals at 1, 3, 6, or 12 months. Decay follows linear logic. If you award 10 points for a form submission with 50% decay at 3 months, those points drop to 5 after 3 months and hit zero at 6 months.
Two things to watch. First, use HubSpot's "Limit to" feature on event groups to cap how many points a single event type can contribute. A contact who clicks 40 emails shouldn't score 120 on engagement alone. Second, don't score email opens. Apple's Mail Privacy Protection inflates open rates to the point where they're unreliable as intent signals.
A score is just a number until you define what it triggers. HubSpot's combined score maps fit and engagement into a matrix: A/B/C for fit quality, 1/2/3 for engagement level. The intersection is where the handoff lives.
The starting framework: 100-point combined scale, split evenly (50 fit, 50 engagement). MQL trigger fires at A1, A2, and B1. Everything else stays in marketing's domain.
But the numbers only work if they're validated against reality. Pull your closed-won data and look at the score distribution. If contacts above your MQL threshold don't convert at a meaningfully higher rate than those below it, the threshold is wrong, or the model is.
Trigger on the combined score reaching A1, A2, or B1. Set lifecycle stage to MQL, create a task for the owner with fit score and last engagement, notify them, and round-robin unowned contacts. Add re-enrollment so a contact who decays out and re-qualifies alerts sales again, and a branch that routes contacts with an open deal to a deal-level notification instead.
The score triggers the handoff. The workflow makes it operational. Without the workflow, you have a number in a field that nobody checks.
Two additions most teams skip. First, add a re-enrollment condition. If a contact drops below MQL threshold (due to decay) and later re-qualifies, sales should be notified again. This is a re-engagement signal. Second, add a branch for contacts who reach MQL but already have an open deal. These shouldn't re-enter the top of the sales process; they should trigger a deal-level notification instead.
A scoring model is a hypothesis. It encodes your best guess about which attributes and behaviors predict conversion. That hypothesis needs to be tested, and most orgs never test it.
Every quarter, pull two reports:
The model that wins isn't the one that's most sophisticated on launch day. It's the one that gets recalibrated on the 90th day.
Build the fit score from the client's closed-won data, not their wish list, and get sales to sign the MQL threshold before the workflow goes live. Use the same structure across portals (fit and engagement separated, decay on, exclusion lists for employees, customers, and competitors) so the quarterly review is one process rather than one per client. Report threshold validation to the client every quarter.
How does score decay work in HubSpot lead scoring?Score decay automatically reduces an engagement event's point value over time. You can set decay intervals of 1, 3, 6, or 12 months. Decay applies per event and follows linear logic: 10 points with 50% decay at 3 months drops to 5 at 3 months and 0 at 6. It ensures your scores reflect current engagement rather than historical activity.
Should you use negative scoring or exclusion lists in HubSpot?Use both, for different purposes. Negative scoring works for contacts who might be a fit but show disqualifying attributes or disengagement signals. Exclusion lists work for contacts who should never enter the scoring model at all: internal employees, existing customers, known competitors. If someone should never qualify regardless of behavior, exclude them rather than subtracting their way to zero.
How often should you recalibrate a lead scoring model?Quarterly is the standard cadence. Pull two reports: conversion rate for leads above vs. below your MQL threshold, and the score distribution of closed-won vs. closed-lost deals. If the rates are similar or the distributions overlap heavily, adjust point values and threshold. Major changes to your ICP, product, or pricing warrant immediate recalibration.
We build scoring models that sales actually uses: fit and engagement separation, decay, and the quarterly feedback loop most teams skip.
Let's Build It →