[GTM glossary](/glossary.md) Propensity model Contextual Intelligence & AI # Propensity model In short A propensity model is a statistical or machine learning model that predicts the probability that a company or person will take a specific action. Common targets are buying, churning, upgrading, or replying to outreach. The output is usually a score or percentage for each record. Teams use it to rank accounts so reps and campaigns focus on the ones most likely to act. ## How a propensity model works The model learns from history. You give it past accounts labeled with the outcome you care about, such as "became a customer within 6 months", along with traits like industry, size, tech stack, and activity. It finds which traits tend to go with the outcome and applies that pattern to new accounts. --- ## Common types * **Propensity to buy:** Ranks prospects by how likely they are to become customers. * **Propensity to churn:** Flags customers at risk of leaving so success teams can step in. * **Propensity to expand:** Finds customers likely to upgrade or add seats. * **Propensity to respond:** Predicts who will engage with a campaign. --- ## A practical example Say a team has two years of closed-won and closed-lost deals. They train a model on firmographic fields and product usage, then score the open account list. Reps start with the top tier and marketing excludes the bottom tier from paid campaigns. After a quarter, the team compares scores with actual outcomes to check the model still works. --- ## Propensity model vs. fit score A fit score rates how closely an account matches your ideal customer profile using rules you set. A propensity model learns the pattern from outcome data and estimates probability of an action. Many teams use both. --- ## Common mistakes * **Too little data:** A model trained on a few dozen deals is unreliable. * **Biased labels:** If reps only worked certain accounts, the model learns their habits rather than real buying patterns. * **Never retraining:** Markets shift, so scores drift. Retrain on fresh data regularly. * **Scores without action:** A score no one uses changes nothing. Tie it to routing and outreach rules. --- ## Propensity model FAQs **What is a propensity model?** A propensity model estimates how likely a person or company is to take a specific action, such as buying, churning, or upgrading. It learns from past outcomes and gives each record a score or probability that teams use to prioritize work. **What is the difference between a propensity model and lead scoring?** Lead scoring often uses manual rules and point values. A propensity model learns from historical outcome data and estimates a probability. Predictive lead scoring is essentially a propensity model applied to leads. **How much data do you need to build a propensity model?** There is no fixed number, but you need enough past examples of both outcomes, such as wins and losses, for patterns to be reliable. Very small datasets produce unstable scores, so start with simpler rules if history is thin. **How do you know if a propensity model is accurate?** Compare its scores with real outcomes on data it has not seen. If top-scored accounts convert noticeably more than low-scored ones, it is useful. Recheck regularly, because accuracy fades as markets and products change. Share term --- ## More related phrases * [Predictive analytics](/glossary/predictive-analytics.md) * [AI lead scoring](/glossary/ai-lead-scoring.md) * [Machine learning](/glossary/machine-learning.md) ## Similar phrases * [Predictive scoring](/glossary/predictive-scoring.md) Predictive scoring uses a model trained on past wins and losses to assign each lead or account a score showing how likely it is to convert, so teams can prioritize effort. * [Machine learning model](/glossary/machine-learning-model.md) A machine learning model is a program that learns patterns from historical data to make predictions or decisions on new data, such as which accounts are likely to convert. * [Predictive lead scoring](/glossary/predictive-lead-scoring.md) Predictive lead scoring uses a model trained on past deals to rank the leads in your database by how likely each one is to become a customer. * [Lead scoring models](/glossary/lead-scoring-models.md) Lead scoring models are the frameworks used to calculate lead scores, including manual point systems, predictive models, and fit and intent grids. * [Fit score](/glossary/fit-score.md) A fit score is a number that rates how closely a company or contact matches your target customer, using static traits such as industry, size, and tech stack.