Most B2B SaaS teams have a lead scoring problem. Not a lack of lead scores - most teams have plenty of those - but a fundamental issue with what those scores are measuring and how well they predict actual revenue.
Standard lead scoring assigns values to behaviours: opened an email (+5), visited pricing page (+10), attended a webinar (+15). Enough points and a lead gets routed to a rep. The logic is that behavioural engagement correlates with buying intent.
The problem is that this correlation is weak. And in a market where your reps' time is your most constrained resource, weak correlation translates directly to wasted effort.
Propensity modelling is a different approach. And when it's applied well - combining three independent propensities rather than a single composite score - it's the closest thing to a reliable buying-readiness signal that exists for B2B SaaS.
What Exactly Is a Propensity Model?
A propensity model is a statistical framework that uses historical outcomes to predict future behaviour. Rather than asking "did this account do something that usually correlates with buying?", it asks "given everything we know about this account and accounts like it, what is the probability that they will buy?"
Unlike lead scoring, which assigns fixed values to behaviours, propensity modelling is adaptive. It examines historical outcomes, identifies genuinely impactful signals, and adjusts dynamically. Lead scoring resembles a checklist; propensity modelling functions as a compass.
In practice, the most effective B2B SaaS propensity models evaluate three distinct propensities simultaneously: the likelihood of ICP Fit (is this the right type of account?), the likelihood of Purchase Intent (is it in-market now?), and Win Probability (can you realistically win this one?). This three-part framework - when built on real subscription event data - is what separates teams that guess from teams that know.
The Anatomy of a Propensity Model
A propensity model has three components:
Signals (Features)
The inputs the model uses to make predictions. In B2B SaaS, high-quality signals include:
- Firmographic: company size, industry, growth rate, geographic market
- Technographic: current tool stack, category usage, adjacent software
- Behavioural: product usage patterns, CRM engagement history, support ticket volume
- Temporal: days since last purchase, renewal proximity, contract age
- Competitive / Intent: competitor research, content consumption, and - most powerfully for SaaS - direct subscription signals: active competitor trials detected, vendor churn, renewal windows, and tech stack changes. MarketSizer provides this layer from 24M+ subscription records, refreshed daily at 90-95% detection accuracy.
Outcome Labels
What the model is predicting. Options include: - Purchase (will this account buy within 90 days?) - Churn (will this account cancel within 90 days?) - Expansion (will this account expand their contract within 180 days?) - Win (will we win this deal if we pursue it?)
Critically, each outcome requires its own model. A model predicting purchase propensity uses different signals than one predicting churn propensity - and combining them into a single score often obscures more than it reveals.
Historical Outcomes
The training data the model learns from. A propensity model is only as good as the historical outcome data it's built on. Teams without sufficient closed won/lost history often need to supplement internal data with external signals - which is where subscription intelligence data at market scale becomes important.
The Three Propensities That Matter Most
1. ICP Fit Propensity
The probability that this account matches the profile of your historically successful customers. Built from firmographic and technographic signals. Tells you: "is this account worth investing in at all?"
ICP fit is the filter. It removes accounts that, based on historical outcomes, are unlikely to convert regardless of timing or competitive context.
2. Purchase Intent Propensity
The probability that this account is actively in a buying window right now. Built primarily from subscription and behavioural signals.
This is where most propensity models fall short - they use content consumption and ad engagement as proxies for purchase intent, when the highest-fidelity signals are direct: is this company trialling a competitor? Is their renewal 30 days out? Did their champion just leave?
The most predictive subscription signals for SaaS buying propensity include: - Competitor trial starts - an account is actively evaluating alternatives - Subscription churn events - a tool has been removed, creating a replacement opportunity - Renewal window timing - a contract is approaching expiry, typically 30-90 days out - Multi-vendor evaluation patterns - simultaneous trials of multiple competitors
These signals are available from subscription intelligence platforms built on real market data - MarketSizer tracks these signals across 154 customer support and live chat vendors using 24M+ subscription records, refreshed daily.
3. Win Probability
The probability that you will win this account if you pursue it, given the competitive context. Built from historical win/loss outcomes in similar competitive scenarios.
Win probability is the dimension most teams omit entirely. They know if an account is in-market. They don't know if it's a fight they historically win. A company in an active Zendesk trial is a different proposition from one in an active Freshdesk trial - your win rate against these competitors may be very different, and that difference should determine how much resource you invest.
How These Three Propensities Work Together
The power of this framework comes from evaluating all three simultaneously rather than collapsing them into a single composite score.
A single composite score hides important information: - High ICP fit + low purchase intent = great future customer, not a current opportunity - High purchase intent + low ICP fit = in-market, but not for you - High ICP fit + high purchase intent + low win probability = a live opportunity you're unlikely to win
Separating the three propensities makes these distinctions visible. Reps can see not just that an account is high-priority, but why - and can calibrate their approach accordingly.
An account that's ICP-fit and in-market but low win probability might still be worth a low-touch touch to maintain relationship. An account that's all three - ICP-fit, in-market, and high win probability - is a Qualified Opportunity worth immediate focus.
Building a Propensity Model in Practice
For teams with limited internal data
If you don't have sufficient closed won/lost history (typically 200+ outcomes minimum), supplement internal data with external subscription intelligence. A market-level dataset covering competitor trials, renewals, and churn events across your category provides the volume and diversity of signals needed to build reliable purchase intent propensity scores.
For teams with strong internal data
Use your CRM history as the foundation for ICP fit and win probability models. Layer external subscription intelligence on top for the purchase intent dimension, which internal data typically can't capture (you don't know what your prospects are doing with competitors).
For CS and expansion teams
Run the same three-propensity framework in reverse: ICP fit for the customer (are they still in the profile of a successful long-term customer?), expansion propensity (are they showing signals of needing more?), and churn propensity (are they showing signals of looking elsewhere?). External subscription signals - competitor trials running in a customer account - are often the first and most reliable indicator of churn risk.
Conclusion: From Scores to Qualified Opportunities
Propensity modelling transcends a simple lead scoring upgrade - it represents movement toward truly predictive GTM strategy. For SaaS companies facing growth pressure, it's transitioning from optional enhancement to essential capability. The end goal isn't a score - it's a Qualified Opportunity: an account where propensity to fit your ICP, propensity to be in-market now, and win probability all align. That's when GTM effort converts most reliably.
Frequently Asked Questions
What is a propensity model in B2B SaaS? A propensity model is a predictive framework that uses historical outcome data to estimate the probability that an account will take a specific action - buy, churn, expand, or respond. Unlike lead scoring, which assigns fixed values to arbitrary behaviours, propensity modelling is built on actual outcomes and adjusted dynamically as new data arrives.
What is the Opportunity Score in B2B propensity modelling? The Opportunity Score is a three-part propensity framework that evaluates accounts across ICP Fit (how well the account matches your best customers), Purchase Intent (signals that it is actively in-market right now), and Win Probability (likelihood of winning based on historical competitive outcomes in similar scenarios). Rather than a single composite score, this approach treats each dimension independently - an account can be high intent but low win probability, or perfect ICP fit but not yet in-market. All three must align to qualify as a Qualified Opportunity.
What subscription signals are most predictive in B2B SaaS propensity models? The most predictive subscription signals for SaaS buying propensity include: competitor trial starts (an account is actively evaluating alternatives), subscription churn events (a tool has been removed, creating a replacement opportunity), renewal window timing (a contract is approaching expiry, typically 30-90 days out), and multi-vendor evaluation patterns (simultaneous trials of multiple competitors). MarketSizer tracks these signals across 154 customer support and live chat vendors using 24M+ subscription records, refreshed daily.
How is propensity modelling different from lead scoring? Lead scoring assigns fixed point values to behaviours and sums them to a score. Propensity modelling estimates the probability of a specific outcome using machine learning on historical outcomes. Lead scoring is a rule-based system; propensity modelling is an evidence-based one. The practical difference: propensity models adjust to new data automatically, whereas lead scores require manual recalibration.
Do you need a large dataset to build a propensity model? For ICP fit and win probability, you typically need 200+ historical closed won/lost outcomes to build a reliable model. For purchase intent, internal data is often insufficient - you can't see what your prospects are doing with competitors. Subscription intelligence data at market scale (24M+ events across 154 vendors) fills this gap and is available without building a large internal training set.