Engagement scoring: what to measure and what to ignore

Engagement is the dimension of lead scoring most teams get wrong, and it fails in the same way every time: too many signals, weighted by volume instead of quality, never decayed, never validated.

This post is the first in a series of deep dives on the four dimensions of a working B2B lead scoring framework. Engagement scores recent, lead-side activity. The signals that matter are the ones that mean something specific about buying intent. The signals that don’t get scored as zero, or deliberately scored down.

The core principle

Engagement should weight event quality, not event count.

A whitepaper download means hours of digestion and warrants weeks of warmth. A pricing page view means thirty seconds of curiosity and decays in days. A demo request is a near-binary signal of intent and should land near the top of the scale. A homepage bounce is noise and should score zero or close to it.

If your engagement score gives “any pageview” the same point value, your model rewards activity volume from researchers, competitors, and curious developers, who collectively generate a lot of activity and never buy. The result is a top decile of leads dominated by high-volume-low-fit accounts that close at near-zero rates.

Signals to score (and how)

The signals worth scoring, ranked roughly by predictive value:

High-intent actions (heavy weight)

• Demo request via form (15–25 points)
• Free trial signup (15–25 points)
• “Contact us / talk to sales” form with a real message (15–20 points)
• Replied to a sales sequence email (10–15 points)
• Booked a meeting via Chili Piper, HubSpot Meetings, etc. (15–25 points)

These are direct, lead-initiated buying signals. They should be the heaviest contributors to engagement.

Quality page visits (medium weight)

• Pricing page (5–10 points, with decay)
• Product / feature comparison page (5–10 points)
• Integrations page (3–5 points)
• Documentation deep reads, defined as 3+ pages in a session (3–5 points, with caveats)
• Webinar registration or attendance (5–10 points)

These represent real consideration, not just casual browsing. The decay matters here: a pricing page view from yesterday is worth more than the same view from three months ago.

Asset engagement (medium weight)

• Whitepaper or guide download (8–12 points, slow decay)
• Case study deep read (5–8 points)
• ROI calculator use (8–12 points)
• Video watch >50% (3–5 points)

Heavy assets that take real time to consume warrant slower decay than quick page visits.

Soft signals (low weight or zero)

• Generic page visits (homepage, about, blog): score these at 0 or 1
• Email opens: useful for trends but not for individual lead scoring
• Single-page sessions under 30 seconds: score these at 0
• Newsletter clicks: score as fit/affinity, not intent

Signals to score down (negative)

Most teams skip negative engagement scoring and lose model accuracy as a result.

• Email unsubscribe (−15 to −25)
• Hard email bounce (−5 to −10)
• “Not interested” form response (−15)
• Duplicate detection on a company already disqualified (−10)

Negative engagement scoring prevents scores from drifting upward forever. Combined with positive scoring, it produces a score that can move both ways with new information.

Decay configuration

Decay is the part of engagement scoring that most teams don’t think about until the model has already decayed silently. Scores that only ever climb are one of the most common reasons HubSpot lead scoring stops working within a couple of quarters of going live.

A practical decay rule: events should retain ~70% of their weight at the half-life, then continue to decay exponentially. Half-lives by event type, as a starting point:

• High-intent actions (demo request, trial signup): 30–45 day half-life
• Quality page visits: 7–14 day half-life
• Asset engagement (whitepaper, ROI calc): 14–30 day half-life
• Soft signals: 3–7 day half-life

HubSpot’s modern scoring tool supports decay per event group at a flat percentage step, not exponential half-life curves. To approximate the above, you’d split events into groups by half-life category and configure decay separately per group. The configuration cost grows quickly with the number of distinct event types, which is one of the reasons we built per-event half-life decay into kenbun natively.

How to validate

Engagement scoring is the easiest dimension to validate because the data exists in your CRM.

Pull every closed-won deal from the last quarter. For each one, look at the engagement events on the contact in the 90 days before close. Build a frequency distribution: which event types appeared most often in winning deals? Which appeared but didn’t predict close? Which never appeared?

If pricing page visits show up in 80% of wins, that’s a strong signal worth heavy weight. If whitepaper downloads show up in 5% of wins despite being scored as 12 points each, the signal is overweighted in your model and worth less than your rules suggest.

This is the calibration loop that keeps engagement scoring honest. Run it quarterly. Tune the weights toward what your data actually shows.

Common mistakes

A few traps:

Scoring email opens. Email opens are a deliverability and engagement-trend signal, not a lead-scoring signal at the individual level. Image-blocking, prefetching, and aggressive privacy proxies make individual opens unreliable. Use email opens for cohort analysis, not lead scoring.

Scoring activity from anonymous traffic. Until a contact is identified, scoring activity is speculative. Score the activity once the contact is known, not before.

Stacking enrichment events as engagement. ZoomInfo enrichment, Clearbit reveals, third-party intent signals are fit signals, not engagement signals. They tell you who the lead is, not what they’re doing. Don’t conflate the two.

Scoring what you can measure instead of what matters. If you have detailed product analytics, every product event will look scoreable. Resist. Score only the events that correlate with closed-won in your retrospective analysis, not every event you happen to capture.

Where this fits in the framework

Engagement is one dimension of a four-dimension framework. The others are profile fit (the person), account fit (the company), and deal context (open opportunities). Engagement alone is a partial picture; a high-engagement low-fit lead is a researcher, not a buyer.

The framework lives in another post. The point of this one is to make engagement scoring honest before you start composing it with the other three.

A note on tooling

If you want engagement scoring with per-event half-life decay (not group-level percentage steps), explainable rule-by-rule breakdowns on every score, and calibration that compares your engagement signals to actual closed-won data, that’s what kenbun ships. See it on your HubSpot data.