Feature · Lead Scoring Models

Lead scoring models where the model lives on the record and learns from the pipeline.

A lead scoring model is only as useful as the signals it reads and the pipeline it feeds. Classic rule-based models score reliably but calibrate slowly. Predictive models calibrate fast but hide their reasoning. The honest model shape combines both: rules capture the sales team's pattern expertise, predictive signals surface patterns the humans missed, and the model shows its reasoning on every lead. Strkr builds all three model types natively, with the quarterly tuning loop running on the tenant's own closed-won and closed-lost data.

Start free trial See pricing
The three model types every revenue team eventually runs

What each model does well and where each falls short.

Lead scoring models fall into three shapes: rule-based, predictive, and intent-based. Each has real strengths and real limits. Mature revenue teams usually run all three in parallel because each captures a different signal class. The sections below walk the three model types with the honest shape of each, and the capabilities that make them work together on one record.

Rule-based scoring

The team's pattern expertise, codified.

Rule-based scoring is a weighted sum of explicit conditions: industry fit, revenue band, job title, geography, product interest. The weights are configured by the sales ops lead based on the team's pattern expertise. The strength is transparency; the limit is that the rules only capture patterns the humans already know.

Predictive scoring

The patterns the humans missed.

Predictive scoring runs the historical closed-won and closed-lost deals through a model that surfaces the signal combinations that correlate with winning. The strength is finding patterns humans missed; the limit is explaining why a lead was scored high. Strkr AI predictive scoring ships with reasoning on every lead to defuse the black-box problem.

Intent-based scoring

The buyer's research signals.

Intent scoring reads third-party intent data (Bombora, G2, TrustRadius) to detect research activity in the buyer's org on relevant topics. The strength is catching active buying windows; the limit is that intent is noisy at the account level and misleads at the lead level without the right aggregation.

Combined scoring

The three stacked, each with its own weight.

The combined score blends the three model outputs with tenant-configurable weights. The combined view shows the rule-based component, the predictive component, and the intent component as sub-scores. Reps see which component drove the overall score and can argue with the parts they disagree with.

Negative scoring

Signals that reduce the score.

Negative scoring handles the signals that disqualify a lead: competitor employee, outside the ICP, geography out of coverage, open support ticket, prior-lost-recent flag. The negative signals subtract from the overall score and can hard-disqualify a lead with a pattern.

Decay scoring

A lead that goes quiet loses points.

Lead score decays with inactivity. A lead that was hot three months ago and has gone quiet since loses points on each decay interval. The decay reflects the real buying cycle: a quiet lead is less likely to convert than an active one, regardless of the initial fit.

Multi-model per segment

SMB and enterprise models differ.

The scoring model varies by segment. SMB leads are scored on a model tuned to SMB closed deals. Enterprise leads are scored on a model tuned to enterprise closed deals. The multi-model approach prevents the SMB signals from drowning the enterprise signals, which is a common failure of single-model scoring.

Model-tuning loop

Weights adjust on win-loss correlation.

Each quarter, the model-tuning loop correlates closed-won and closed-lost deals against the signals that were present at scoring time. Weights that produced predictive scores stay. Weights that produced noise get re-tuned. The loop is explicit (the weights are visible) and the review is a dashboard.

Reasoning on every lead

The model shows its work.

Every scored lead carries a one-sentence reasoning note explaining the top signal drivers. Reps see "Scored Hot because: 500-employee tech services, SMB sweet spot, high intent on CRM topics." The reasoning defuses the black-box objection and lets reps argue with the model usefully.

Where rule-based scoring is still the right starting point

The patterns a sales ops lead codifies before the predictive layer kicks in.

For most growing teams, rule-based scoring is the right starting point. The team already has pattern expertise. The sales leader can describe the ICP in a sentence. The rules codify that expertise. Over the first two quarters, enough closed data accumulates to layer the predictive model on top. The sections below walk the rule-based patterns that reliably work in the first two quarters of a scoring program.

Firmographic fit

The ICP sweet spot, as a rule.

Industry, revenue, headcount, and geography carry the firmographic weight. A B2B SaaS company selling to mid-market tech might score 50-500 employees heavily, 500-2000 moderately, and under 50 or over 2000 lightly. The rule reflects the ICP sweet spot from the team's own close rate.

Role fit

The right title and seniority.

The role rule weights by title and seniority. A VP of operations at the ICP company carries more weight than a specialist at the same company. The role rule also handles the dual signal of champion and economic buyer: a champion scores middle, the economic buyer scores high.

Behavioral signal

Demo request, pricing page, case study.

Behavioral signals score the lead on specific site actions: demo request, pricing page view, case study download, trial signup. Each action has a weight. Demo request is highest; a trial signup is near the top; a pricing page view is middle; a blog post view is low.

Engagement cadence

Repeat touches signal interest.

The engagement rule scores on the frequency and recency of touches: multiple page views in a week, multiple email opens, repeat site sessions. A lead that is engaging multiple times in a short window carries a stronger buying signal than one with a single touch.

Source quality

Not all lead sources convert equally.

Lead source weights reflect the historical conversion rate per source. A referral scores higher than a cold inbound from a content download. A paid social lead might score below either. The source weight is tunable and updates based on the quarterly conversion review.

Timing signal

A fresh touch is worth more.

A lead that touched the site in the last 72 hours scores higher than one from 30 days ago. The timing rule reflects the recency bias in B2B buying windows: fresh signals convert at substantially higher rates than stale ones.

Negative firmographic

Outside-ICP signals subtract.

A lead from outside the ICP (wrong geography, wrong industry, wrong size) subtracts from the score. A lead from a competitor subtracts heavily. A lead from a known freemium abuser subtracts. The negative rules keep the queue focused on real buying candidates.

Negative behavioral

Open ticket or recent lost.

A lead with an open support ticket in the parent account subtracts (the account is in a different conversation right now). A lead whose account was lost in the last 90 days subtracts (the buyer's cycle has not reset). The negative behavioral rules prevent misfires.

Decay rule

A quiet lead loses points.

Each week of no activity subtracts a small amount. A lead that was hot three months ago and has gone silent since decays into the warm or cool tier. The decay reflects that a quiet lead is less likely to convert than an active one, regardless of the initial fit.

When predictive scoring earns its weight

The pattern that only shows up with historical data.

Predictive scoring earns its keep once there are at least 200 closed-won deals and 200 closed-lost deals to train on. Below that threshold, the model mostly captures sampling noise. Above it, the model surfaces real patterns the human rules missed. The sections below walk the places predictive adds value on top of rule-based.

Signal combinations

The pairs humans miss.

Humans tend to score single signals. Predictive models find signal pairs: companies with high intent AND recent exec change convert at 4 times the baseline, for example. The pair-wise and triple-wise signal interactions are where predictive adds real value.

Non-obvious correlations

The signal nobody suspected.

A predictive model sometimes finds a strong signal that the humans never would have scored: a technology in the stack, a support-ticket pattern, a role move in the past six months. The non-obvious signals are the compounding value of a predictive layer.

Weight correction

The rule weights the humans got wrong.

Humans guess the weights on firmographic signals. Predictive models actually measure them. When a human-set weight is materially off, the predictive layer surfaces the discrepancy and the quarterly review tunes the rule weights accordingly.

Segment-specific patterns

SMB signals differ from enterprise signals.

Predictive models tuned per segment surface the signal patterns specific to each segment. SMB closes on different signals than enterprise. The multi-model approach prevents the aggregate from washing out the segment-level patterns.

Reasoning preserved

Not a black-box score.

Strkr AI predictive scoring ships with reasoning: "Scored Hot because the top three driving signals are mid-market segment fit, high intent on your category, and a recent VP ops hire." The reasoning defuses the black-box objection. Reps can argue with the parts they disagree with.

Confidence tier

The model admits when it is guessing.

The predictive score carries a confidence tier. A lead with little data gets a Low-confidence tier, which flags that the score is a weak signal and should not drive aggressive outbound alone. A lead with full signal gets a High-confidence tier.

How Strkr runs the three model types together

The architecture that lets rules, predictive, and intent coexist.

The three model types are useful in combination, not isolation. The architecture that makes them work together is: rules run on the lead, predictive runs on the account signal state, intent runs on the account research state, and the combined score blends with tenant-configurable weights. The sections below walk the architectural capabilities that make the combination coherent.

One record

Lead and account share state.

The lead and the account are linked on first write. Signals that live on the account (intent, firmographics, equipment age, OSHA record) feed the lead-level score. Signals that live on the lead (role, behavioral, engagement) feed the account-level score. The two sides inform each other.

Transparent weights

Admin edits the weight, not the code.

All weights are configuration, not script. Admins open the scoring config, see the rule weights, the predictive component weight, the intent component weight, and the segment-specific overrides. Changes apply on the next scoring cycle without a deploy.

Score history

Every score, with the inputs.

Every lead carries a score history: the score, the sub-scores, the signal state, and the reasoning on each scoring pass. The history is queryable. A quarterly review of scoring accuracy runs on the history, not on reconstructed state.

Tiered categories

Hot, Warm, Cool, Cold.

The combined score rolls into a tier: Hot, Warm, Cool, Cold. Tier thresholds are tenant-configurable. The tier is a sortable field on the lead queue. The tier drives routing, cadence enrollment, and SDR prioritization.

Routing integration

The score drives the owner.

The lead router reads the score tier. Hot leads route to the SDR team with a 1-hour SLA. Warm leads route to the nurture track. Cool and Cold leads enter long-form marketing cadences. The score is the primary routing dimension.

Cadence integration

The cadence varies by tier.

The cadence engine enrolls the lead based on the score tier. Hot leads enter a short high-touch cadence. Warm leads enter a medium cadence. Cool leads enter a long nurture. The cadence shape reflects the real buying rhythm per tier.

Dashboard reporting

Score distribution and win rate.

A dashboard shows the lead score distribution, the win rate by tier, the conversion funnel by tier, and the quarterly delta. The reporting surfaces whether the scoring is predictive and tunes the review process accordingly.

Model-tuning loop

Quarterly review, data-driven.

Each quarter, the tuning loop surfaces weight proposals: raise the weight on signal A, lower the weight on signal B, add signal C as a new rule. Admin reviews and approves. The weights update. The next quarter runs on the tuned model.

Pricing

Every model type, every tier.

Rule-based, predictive, and intent-based scoring all ship on every paid Strkr tier. Multi-model per segment and the quarterly tuning loop are on Pro and above. No per-score meter, no separate scoring SKU. The seat price is the full price.

Lead scoring models where rule-based, predictive, and intent all run on one record.

Open a free trial, codify your rule-based model in the first week, layer in predictive on the historical data, and watch the combined score drive routing and cadence on every inbound lead. The seat price is the full price. No separate scoring SKU, no credit meter.

Start free trial See lead scoring
Common questions

What buyers ask about this feature.

What are the main types of lead scoring models?

Three model types cover almost every revenue team's needs: rule-based (a weighted sum of explicit conditions configured by the sales ops lead), predictive (a model trained on historical closed-won and closed-lost data that surfaces non-obvious signal patterns), and intent-based (reads third-party intent data from Bombora, G2, or TrustRadius to detect research activity). Mature revenue teams usually run all three in parallel because each captures a different signal class. The combined score blends the three with tenant-configurable weights, and the sub-scores are visible on every lead so reps can see which component drove the overall score.

When should we move from rule-based to predictive scoring?

Predictive scoring earns its weight once there are at least 200 closed-won deals and 200 closed-lost deals to train on. Below that threshold, the model mostly captures sampling noise. Above it, the model surfaces real patterns the human rules missed: pair-wise and triple-wise signal interactions, non-obvious correlations, and weight corrections where the human-set weight was materially off. For most teams, rule-based is the right starting point for the first two to four quarters. Predictive layers on top once the historical data is sufficient. Strkr runs both simultaneously, so the transition is a gradual re-weighting rather than a replatform.

How does the model explain its scores to the reps?

Every scored lead carries a one-sentence reasoning note explaining the top signal drivers. Reps see "Scored Hot because: mid-market tech services, SMB sweet spot, high intent on CRM topics, recent VP ops hire." The reasoning note shows the top three to five contributing signals for the score. For the predictive component, Strkr AI provides a plain-English explanation of why the model weighted the lead the way it did. The reasoning defuses the black-box objection common to predictive scoring and lets reps argue with the parts they disagree with. Managers auditing scoring decisions see the same reasoning on every lead.

Can we run different models for different segments?

Yes. Multi-model per segment is a core capability. SMB leads are scored on a model tuned to SMB closed deals. Mid-market leads are scored on a model tuned to mid-market. Enterprise leads are scored on a model tuned to enterprise. The multi-model approach prevents the SMB signals (which dominate volume) from drowning the enterprise signals (which dominate revenue). The segmentation is a configuration, not a code change. Each model carries its own rule weights, its own predictive weights, and its own tier thresholds, so the Hot tier in SMB is calibrated differently than the Hot tier in enterprise.

How often do the weights get re-tuned?

The model-tuning loop runs quarterly by default and can be configured to run more frequently for high-volume teams. Each tuning pass correlates closed-won and closed-lost deals against the signals that were present at scoring time. Weights that produced predictive scores stay. Weights that produced noise get re-tuned. Strkr AI proposes weight changes based on the correlation analysis; the admin reviews, approves, and the updated weights apply on the next scoring cycle. The tuning history is preserved so a quarterly review can see which signals have strengthened or weakened over time.

Is lead scoring included on every tier?

Core lead scoring (rule-based model, basic predictive layer, model-called category, event triggers, audit trail) ships on every paid Strkr tier. Advanced capabilities (multi-model per segment, intent data integration via Bombora or G2, the quarterly weight-tuning loop with Strkr AI proposals, decay scoring, negative scoring) are included on Pro and above. Enterprise-grade features (custom predictive model training on bespoke features, intent integration with multiple providers, confidence-tiered scoring with per-tier policies) are included on Scale and Enterprise. There is no separate scoring SKU, no per-score charge, and no credit meter. The seat price listed on the pricing page is the full price.

Related features

Pair this with...

Lead Scoring Lead Scoring Criteria Examples B2B Lead Scoring Criteria Examples Lead Scoring Engine Criteria Examples Lead Management Lead Routing Software AI CRM

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.

Start free trial See pricing