What a Healthy Lead Scoring Model Actually Looks Like: Behavioral Signals, Demographic Fit, and the Decay Logic Most Teams Skip

If your lead scoring model only assigns points, it’s not a scoring model — it’s a leaderboard. Scores go up when leads engage, and they never come down. Over time, every lead in your database with more than six months of history will have accumulated enough points to look like a hot lead, regardless of whether they’ve been active recently or fit your ideal customer profile.

This post covers the full architecture of a lead scoring model that actually works: the behavioral and demographic components, the decay logic that makes scores meaningful over time, threshold calibration, and the sales feedback loops that prevent scoring from becoming disconnected from commercial reality.

The two dimensions of a complete scoring model

Lead scoring has two fundamentally different components that are often conflated but need to be designed and maintained separately: fit scoring and behavior scoring.

Fit scoring answers the question: does this person match the profile of someone who could become a customer? It’s based on demographic and firmographic data — job title, seniority, company size, industry, geography. A VP of Marketing at a 500-person SaaS company in your target vertical is a high-fit lead even if they’ve never opened an email. An intern at a company in the wrong industry is a low-fit lead even if they’ve attended three webinars.

Behavior scoring answers the question: is this person actively engaging with content in a way that suggests they’re in a buying process? It’s based on activity data — email clicks, form submissions, page visits, event attendance, content downloads. Behavior scoring can change daily as a lead engages or goes quiet.

The best scoring models use both dimensions and combine them meaningfully. A high-fit lead with high engagement is your best candidate for sales outreach. A high-fit lead with low engagement is worth nurturing but not calling. A low-fit lead with high engagement may be worth a self-serve path but probably isn’t worth sales time. Combining fit and behavior into a single score — rather than maintaining them as separate dimensions — loses this nuance and makes the model significantly less useful.

Behavioral signals: which activities to count and how much

Not all engagement signals are equally meaningful. The scoring model needs to reflect this with differentiated point values.

High-value signals — form submissions, pricing page visits, demo requests, bottom-of-funnel content downloads, event registrations — should carry significant point values because they indicate active consideration. A lead who fills out a contact form or downloads a pricing guide is sending a qualitatively different signal than a lead who opened an email.

Medium-value signals — email clicks (to specific content types), webinar attendance, multiple visits to product pages — are engagement indicators but not necessarily buying signals. They should contribute to the score but not drive it to threshold on their own.

Low-value signals — email opens, single page visits, social media engagement — are noise more than signal at the individual lead level, and many teams score them too heavily. Email opens in particular have become less reliable since iOS privacy changes affected open tracking. If you’re scoring email opens at more than one or two points, recalibrate.

Score decay: the mechanism most models skip

Score decay is the mechanism by which behavioral scores decrease over time in the absence of new engagement. It’s what makes your scoring model an accurate representation of current engagement rather than a cumulative history of all engagement ever.

In Marketo, decay can be implemented in several ways. The simplest is a scheduled smart campaign that runs weekly or monthly and applies negative point adjustments to leads who haven’t engaged in a defined period. A lead who was highly engaged six months ago but has been silent since should not still be sitting at a high behavioral score — they should have decayed to a score that reflects their current engagement level.

The decay logic needs to be calibrated to your sales cycle. If your average sales cycle is 90 days, meaningful decay should start happening around the 30-45 day inactivity mark. If your cycle is 12 months, the decay window should be longer. The goal is for your scoring model’s behavioral component to reflect recency-weighted engagement, not lifetime engagement.

A common implementation: at 30 days of inactivity, remove 10 points. At 60 days, remove 20 additional points. At 90 days, remove all remaining behavioral score. This forces recency into the model and means that a lead’s score will drop significantly if they go dark — which is the correct signal for sales to deprioritize them.

Threshold calibration: what score actually means “ready for sales”

The MQL threshold — the score at which a lead is routed to sales — is the most important and most frequently miscalibrated number in the scoring model. Set it too low, and you’re flooding sales with leads that aren’t ready. Set it too high, and you’re holding back genuinely interested prospects.

Calibration requires data. Pull a sample of leads that converted to opportunities in the last 12 months and look at what their scores were at the time of conversion. Pull a sample of MQLs that were rejected by sales as “not ready” and look at their scores. The threshold should sit at the inflection point where conversion rate to opportunity starts becoming meaningful — not at an arbitrary round number that was set three years ago and never revisited.

Revisit threshold calibration at least twice a year. As your database grows, as your content mix changes, and as your ICP evolves, the score distribution will shift and your threshold needs to shift with it.

Sales feedback loops: the governance mechanism the model depends on

A scoring model without a sales feedback loop is operating on assumptions. The loop works like this: when sales acts on an MQL — accepts it, rejects it, or converts it — that outcome data needs to flow back into your model calibration process. If sales is consistently rejecting a certain demographic segment as low-quality despite their scoring above threshold, that’s a fit scoring problem to fix. If leads from a specific content type are consistently converting at high rates, that’s a behavioral signal to score more heavily.

Build the feedback mechanism explicitly. A monthly review with your sales development lead that covers MQL acceptance rate, rejection reasons, and conversion outcomes is the minimum. If your CRM supports it, track MQL disposition codes so you have structured data on why leads are being accepted or rejected — not just anecdotes.

Scoring that reflects commercial reality is the goal. The feedback loop is how you keep it there.