An expansion propensity model scores installed accounts on their likelihood of buying more inside a defined window. It replaces the quarterly ritual where CSMs nominate accounts they feel good about, and it produces a number you can grade after the fact.
What the Score Predicts
A usable score names three things. The probability of expansion, the window it applies to, and the expected amount. Drop the window and the score cannot be validated, because every account eventually expands or leaves and any model looks correct on a long enough horizon.
The prediction target should match how the business books revenue. ORM separates expansion into new customer ARR, new product ARR, and increased product ARR inside its monthly retention waterfall. A model that predicts seat growth and a model that predicts cross-sell of a second product are answering different questions and need to be scored apart.
Inputs That Carry Signal
Product behavior leads. Seat utilization against entitlement, depth of feature use beyond the original purchase, and the direction of usage over recent months tell you whether an account has outgrown its contract.
Support behavior is the input most teams read backwards. ORM's finding is that support volume follows a curve rather than a line. Accounts with no support cases are at churn risk, since silence means nobody is using the product enough to have questions. Accounts with seven or more cases in a year are also at risk. The healthier population sits at three to five non-severe tickets, engaged and getting help.
Contract history matters too. Time since the last amendment, prior expansion behavior, and whether the account has ever added a product all shift the base rate before any usage data is considered.
Why Rep Sentiment Underperforms
Sentiment carries real information and inconsistent calibration. One CSM marks every friendly call as an opportunity, another only flags accounts with budget confirmed, and the roll-up averages two incompatible scales.
Sentiment also lags behavior. Usage decays before a relationship cools, and a model watching product signals sees the change while the relationship still feels fine on calls.
Where the Score Gets Used
The score should drive two decisions. Which accounts get worked this quarter, and what expansion number goes into the plan. Feeding the second is where it pays, since expansion is usually forecast by asking CSMs and then discounted by leadership on instinct.
Grade the model the same way you grade new business. Compare predicted expansion to booked expansion by cohort, and track how the misses land against net revenue retention. If your expansion motion runs through a formal pipeline, hold it to the same win rate scrutiny as new logo deals.
Frequently Asked Questions
What does an expansion propensity model predict?
The probability that an existing account expands inside a stated window, usually the next one or two quarters, along with an expected amount. A score without a window cannot be validated against outcomes.
What inputs work best for expansion scoring?
Behavioral inputs from the account itself. Seat utilization against entitlement, adoption of modules outside the original purchase, support engagement patterns, and time since the last contract change all carry more signal than firmographics.
How do support cases relate to account risk?
ORM finds that accounts with no support cases are at churn risk and accounts with seven or more in a year are also at risk, while accounts logging three to five non-severe tickets are less likely to churn because they are engaged and getting help.
Is an expansion propensity score the same as a health score?
No. A health score summarizes current condition. A propensity model predicts a specific future purchase event and gets graded against whether that purchase happened.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like expansion propensity model into prescriptive action for your team.
Schedule a Demo