A propensity to buy model assigns each account or open opportunity a probability of purchasing inside a defined window, learned from what comparable records actually did. The model reads resolved history rather than opinion. It compares an active record against thousands of closed ones and returns a number you can rank, filter, and forecast against.
What Goes Into the Score
Propensity models draw on three families of input. Firmographic fit covers size, industry, geography, and installed tech. Behavioral signals cover email replies, meeting frequency, stakeholder count, product usage, and buyer-initiated contact. Deal shape covers amount, stage, age, source, and how many times the close date has moved.
Behavioral inputs outweigh firmographic ones once an opportunity is open. Fit tells you whether an account should buy. Behavior tells you whether it is buying. A perfect-fit account with no buyer-initiated activity scores below an imperfect-fit account whose champion is chasing legal review.
Propensity Score Versus Stage Probability
Most CRMs ship a static probability bolted to each pipeline stage. Stage 4 equals 60 percent for every deal in Stage 4. That figure is an average describing no individual deal. A propensity model replaces it with a record-level estimate that changes when the record changes.
The difference shows up in forecast accuracy. Stage weighting only moves the number when a rep advances a deal, and reps advance deals late. A propensity model moves when buyer behavior moves.
Where Propensity Models Break
First failure: a field whose meaning shifted mid-year. ORM's position is that every company has messy data and it rarely matters, because a model can learn from garbage as long as the garbage stays consistent. Redefining a stage or a source picklist halfway through the training window does real damage.
Second failure: training on a market that no longer exists. Pricing pressure from a new entrant, a rate move that freezes buyer budgets, a territory reshuffle that distracts the field, and general buying hesitation all change the relationship between signal and outcome. A model trained before the shift keeps scoring by the old rules.
Third failure: scoring accounts nobody works. Propensity is conditional on being contacted. A high score on an untouched account predicts nothing.
Putting the Score to Work
Rank, then act. Route the top band to your strongest reps and your most expensive plays. Move the bottom band to nurture and stop spending seller hours there. In pipeline review, flag every deal where the model and the rep disagree by a wide margin, and make the rep produce evidence. That disagreement is where deal slippage gets caught before it reaches the number.
Recalibrate quarterly. A propensity model is a claim about how your market behaved, and markets stop cooperating. Pair the score with a disciplined sales forecasting process so the output gets inspected instead of trusted blindly.
Frequently Asked Questions
What is a propensity to buy model?
It is a statistical or machine learning model that scores an account or opportunity on its likelihood to purchase within a defined window. The model learns the relationship between signals and outcomes from resolved historical records, then applies that relationship to open ones.
How is a propensity score different from a CRM stage probability?
A stage probability is one fixed number applied to every deal sitting in that stage, so it describes the average and no individual deal. A propensity score is computed per record and updates when the record's signals change, which is usually earlier than a stage advance.
Does messy CRM data make a propensity model useless?
No. ORM's position is that every company has messy data and it rarely blocks accurate prediction, because a model can learn from inconsistent-looking data as long as the inconsistency is consistent. What actually breaks a model is a field whose definition changed partway through the training period.
How do you know a propensity model is still working?
Check calibration by score band each quarter. Take every deal that scored 70 to 80 percent and measure what share actually closed. If the band scores 75 and closes at 50, the model has drifted and needs retraining on more recent outcomes.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like propensity to buy model into prescriptive action for your team.
Schedule a Demo