Propensity Modeling for Marketing
Propensity models predict a probability, not a value
Where predictive LTV estimates a continuous revenue figure, a propensity model predicts the probability of a specific binary (or categorical) event: propensity to purchase, propensity to churn, propensity to respond to an offer, propensity to upgrade plans. Technically these are standard classification models (logistic regression, gradient boosted trees, or neural nets depending on data volume and complexity) trained on historical examples of customers who did and didn't take the action, scored on features available before the action occurred. The output is a probability score (0 to 1) per customer, used to rank and prioritize marketing treatment rather than to predict a dollar amount.
The most common and most valuable use: propensity-to-purchase scoring for prioritization
A propensity-to-purchase model scores every lead or prospect on likelihood of converting within a defined window, letting sales and marketing prioritize limited outreach capacity (a sales rep's time, a limited retargeting budget, a finite email send) toward the highest-probability segment rather than treating every lead identically. This is structurally the same problem as lead scoring, but propensity models are typically built with real statistical modeling on historical conversion data, rather than the hand-weighted point systems common in simpler lead scoring setups — the distinction matters because a properly trained propensity model can be validated against held-out data and recalibrated as patterns shift, where a hand-tuned point system can't be rigorously validated the same way.
// Propensity-to-convert scoring: train on historical leads with known
// outcomes, score active leads, and route by probability tier.
interface LeadFeatures {
daysSinceFirstTouch: number;
pageviews: number;
emailOpenRate: number;
firmSizeBucket: string;
pricingPageVisited: boolean;
demoRequested: boolean;
}
interface ScoredLead {
leadId: string;
propensityScore: number; // model output, 0-1
}
function routeByPropensity(lead: ScoredLead): string {
if (lead.propensityScore >= 0.7) return "immediate-sales-outreach";
if (lead.propensityScore >= 0.35) return "nurture-sequence-high-intent";
return "nurture-sequence-standard";
}
// In production this score is typically recomputed on a schedule (nightly
// batch) or in near-real-time as new behavioral signals (e.g. pricing
// page visit) arrive, triggering re-routing without manual intervention.Propensity model types in marketing
| Model | Predicts | Primary use case |
|---|---|---|
| Propensity to purchase | P(convert within window) | Prioritizing leads/prospects for sales or retargeting |
| Propensity to churn | P(cancel/not renew within window) | Triggering retention workflows (overlaps with churn risk scoring) |
| Propensity to respond | P(engage with a specific offer/channel) | Suppressing low-propensity recipients from a send to protect deliverability |
| Propensity to upgrade | P(move to a higher plan tier) | Targeting expansion/upsell campaigns |
A propensity score without a calibration check can mislead prioritization
Propensity models often outperform rule-based suppression lists
What's Next?
A propensity score tells you who is likely to respond — the next layer of sophistication decides, for each individual, which specific action or offer to actually present, which is the domain of next-best-action decisioning systems.
I build these systems professionally.
Whether it's a RAG pipeline, analytics migration, or AI workflow — let's talk.