What is a lead score, in one sentence?
A lead score is a single number, usually on a 0 to 100 scale, assigned to each lead based on behavioral and demographic signals, used to rank which leads sales should contact first.
Answer · Lead Scoring
What is lead scoring?
The short version: a score is a ranking signal, not a verdict. It tells a rep which five of today's thirty inbound leads to call first, in what order, and based on which specific signals.
Short answer
Lead scoring is a methodology that assigns numeric values to leads based on behavioral and demographic signals. Behavioral signals include website visits, email opens, demo requests. Demographic signals include company size, industry, title, funding stage. Marketing uses lead scores to prioritize which leads sales should contact first and when a lead crosses an MQL threshold. Modern lead scoring is model-driven, often with AI-trained weights that update based on closed-won conversion data.
What matters most.
Six things to understand before building or buying a lead scoring system. Each one is the difference between a model reps use every day and a model reps route around within a quarter.
Two kinds of signals
Behavioral plus demographic.
Every lead scoring model combines what the lead did (email opens, pricing-page visits, demo requests, webinar attendance) with who the lead is (title, department, company size, industry, geography, funding stage). Behavioral signals answer "are they paying attention right now." Demographic signals answer "would they ever buy from us." You need both.
The output is a number
A ranked sort order for sales.
The score is a single number per lead, usually on a 0 to 100 scale. Its only real job is ranking the inbound queue. The highest-scored lead gets called first. The next one gets called next. Treat the score as a sort order, not a prophecy about whether the deal will close.
MQL threshold
The handoff line from marketing to sales.
When a lead's score crosses a configured threshold, the lead is marked a Marketing Qualified Lead and routed to a sales rep. Below the line the lead stays in nurture. Above the line a flow fires, an owner is assigned, and the clock starts on first-touch SLA.
Rule-based vs predictive
Two approaches, usually run side by side.
Rule-based scoring uses weights a human writes: VP title worth 10 points, pricing-page visit worth 12. Predictive scoring uses a model trained on historical closed-won versus closed-lost data to find signal combinations rules never covered. Modern systems run both and show the rep why.
Score decay
A lead that goes cold stops being hot.
Engagement points decay on a half-life. A pricing-page visit from 60 days ago is worth less than one from yesterday. Without decay, the top of the queue slowly fills with leads that were interested two quarters ago and have ghosted since. Decay keeps the score honest in real time.
Explainability is non-negotiable
Reps ignore any score they cannot read.
A black-box number that says this lead is an 87 but will not say why gets routed around by the AE within a week. The score becomes decoration. The useful models show the top three contributing signals and the points each one added, visible in the UI the rep already works in.
From raw signal to ranked queue, step by step.
Lead scoring is less a feature and more a small pipeline that runs on every lead in the system, continuously. The pipeline has four stages: collect the signals, apply the weights, combine the component scores, and route on threshold. Each stage can be rule-based, model-driven, or a mix of the two. The sections below walk the full pipeline the way a scoring system has to work in production, not the way it is drawn on a vendor slide.
Collect the signals
Capture every lead event in one record.
A scoring model is only as good as the data feeding it. The CRM has to log form fills, email opens, email clicks, page views, pricing-page visits, demo-video views, chat sessions, webinar attendance, inbound replies, and inferred firmographic data on the account. One lead record, one timeline, one source of truth. Scattered data equals scattered scores.
Apply the weights
Each signal earns a point value.
Each signal carries a weight. Rule-based: a VP of Sales title is 10 points, pricing-page visit is 12 points, a demo request is 25 points. Predictive: an ML model assigns weights based on which signals correlated with closed-won deals in the last 12 months. Rule weights are human-set. Predictive weights are learned and updated automatically.
Combine the components
Fit score plus engagement score plus AI.
Most modern systems maintain two or three component scores and combine them into one overall score. Fit score covers demographic and firmographic signals. Engagement score covers behavioral signals. Predictive score covers the model-learned combinations. Combining gives the rep a single number plus a breakdown of why that number is what it is.
Route on threshold
Score crosses a line, a flow fires.
When any lead crosses the configured MQL threshold, a flow fires. Common patterns: score above 80 routes to an AE in a Slack notification, score above 60 drops to a nurture track, score below 20 flips to disqualified. Routing is where the score becomes a workflow instead of a dashboard metric.
Decay nightly
Keep the queue reflecting reality now.
Every engagement point has a half-life. A pricing-page visit from 60 days ago is worth a fraction of one from yesterday. The decay job runs nightly across every lead. Without it, the top of the queue slowly fills with leads that were interested a quarter ago and have gone silent since.
Explain every score
The rep sees the number and the reason.
Click the score on any lead to see the top signals that drove it. Pricing-page visit on Tuesday worth 12 points. Title match worth 8 points. AI contribution worth 7 points, driven by "similar closed-won accounts in the last 90 days." The explanation is not a nice-to-have. It is what makes the score usable.
Two models, one answer, meant to run together.
Buyers often ask which scoring approach to pick, rule-based or predictive. The real answer is both. Rule-based scoring is transparent and easy to explain to a comp-planning committee. Predictive scoring finds the combinations rules miss. Running them in parallel gives the rep a single number and the reasons behind it, and it protects the team from either model drifting too far on its own.
Rule-based
Humans write the weights.
A revops lead sits down and decides: VP title is worth 10, Director is 6, Manager is 3. Pricing-page visit is 12. Demo request is 25. The weights are visible, auditable, and defensible. The downside is they get stale quickly if nobody revisits them, and they can only encode patterns a human already noticed.
Predictive
A model learns the weights from closed-won data.
A machine-learning model trains on the last 12 to 24 months of closed-won versus closed-lost deals and learns which signal combinations correlated with revenue. The weights are learned, not written. The upside is finding non-obvious combinations. The downside is the model needs enough history to train on, and explainability has to be built in.
Hybrid in practice
Rules for transparency, AI for coverage.
Most teams in 2026 run both. Rule-based fit and engagement scores run for transparency. A predictive model runs on top and nudges the overall score by a bounded amount, say plus or minus 15 points, based on learned combinations. The rep sees three component scores and the top signals for each one.
Explainability
Both approaches have to show their work.
Rule-based scoring is naturally explainable because the weights are written down. Predictive scoring has to work harder: it has to surface the top contributing features for every individual score. Modern predictive engines expose the top three signals and their point contribution for every lead, every time.
Refresh cadence
A scoring model is a living document.
Rule-based weights need a quarterly review the same way a comp plan does. Predictive models retrain on new closed-won data every 30 to 90 days. A model that was accurate in Q1 can go stale by Q3 if the product shipped new features, the market shifted, or the ideal customer profile changed. Treat refresh cadence as part of the system.
Common failure modes
What goes wrong in production.
Rules written on day one and never revisited. Predictive models with no explainability that reps ignore. No decay, so the queue fills with cold leads. No threshold routing, so the score is a dashboard metric rather than a workflow. Any one of these turns scoring into decoration. All four of them together is why most teams silently stop using scoring within a year.
See lead scoring that actually tells reps who to call.
Rule-based fit and engagement scoring plus Strkr AI predictive scoring on every paid tier. Score decay, threshold routing, model explainability, no data-volume floor.
Related questions.
What is the difference between lead scoring and lead grading?
Lead grading usually refers to the fit component alone, often expressed as a letter grade from A to D, based on how closely the lead matches the ideal customer profile. Lead scoring usually refers to the combined number that includes both fit and engagement. Some vendors use the terms interchangeably, so always check the definition in the specific tool.
What is an MQL and how does lead scoring relate to it?
A Marketing Qualified Lead is a lead whose score has crossed a configured threshold, signaling that marketing has done enough nurturing and the lead is ready for a sales conversation. Lead scoring is the mechanism that defines the MQL line. Below the threshold the lead stays with marketing. Above it, routing kicks in and a sales rep is assigned.
What signals should a lead scoring model use?
Behavioral signals include email opens, email clicks, page views, pricing-page visits, demo-video views, form fills, chat sessions, webinar attendance, and inbound replies. Demographic signals include job title, seniority, department, company size, industry, revenue band, geography, tech stack, and funding stage. Most models combine 15 to 40 signals in total.
What is predictive lead scoring?
Predictive lead scoring uses a machine-learning model trained on historical closed-won and closed-lost deals to assign weights to signals automatically. Instead of a human writing that a VP title is worth 10 points, the model learns which signal combinations correlated with revenue over the last 12 to 24 months and returns a probability score for every new lead.
How often should a lead scoring model be reviewed?
Rule-based weights need a quarterly review, the same cadence as a comp plan or a territory map. Predictive models retrain on new closed-won data every 30 to 90 days. A scoring model that was accurate a year ago can score noise today if the product shipped new features, the market shifted, or the ideal customer profile changed.
What does score decay mean?
Score decay is a nightly job that reduces the point value of engagement events as they age. A pricing-page visit from 60 days ago is worth less than one from yesterday. Without decay, the top of the queue slowly fills with leads who were interested a quarter ago and have been silent since.
Does lead scoring work for small teams or only enterprise?
Lead scoring works for any team that gets more inbound leads than it can call in a day. A five-rep team with 100 inbound leads a week benefits from scoring the same way a 500-rep team does. The difference is scale, not relevance. The failure mode for small teams is choosing a tool that gates predictive scoring behind an enterprise price tier.
Other questions in this category.
Try it free. Bring your team next week.
No sales call, no migration consultant, no four-month implementation. Enter your card, get 14 days of the full Pro tier, cancel any time before day 14 with zero charge. Spin up a workspace, import your CSV, and have something useful before lunch.