Propensity Modelling

Target the Prospects Most Likely to Act

Illustration of propensity modelling turning an undifferentiated customer base into groups classified by low, medium, and high likelihood to act.

Propensity modelling estimates how likely each customers is to take a specific action based on their past behaviour, giving you a priority list of who to focus your sales and marketing effort on.

From first data review to scored output, the analysis is shaped around identifying the high-value opportunities hiding in your data.


Book a 30-minute call



What you’ll gain

A propensity model brings order to an otherwise undifferentiated list of customers and prospects, so your team stops treating a cold lead the same as a warm one.


Know who to focus your efforts on

Illustration of an undifferentiated customer base sorted into high, medium, and low propensity groups, each with a recommended action: contact first, contact if budget allows, or exclude.

Scoring each contact against patterns in historical conversion behaviour pinpoints the prospects most worth pursuing.

Concentrate time and budget on the prospects with the highest score and maximise your conversion rate.


When propensity modelling is useful

Propensity modelling works best when you need to prioritise limited sales or marketing resources across a large and varied customer or prospect base. Common situations include:

  • Campaign Efficiency - You are running a direct campaign and want to concentrate spend on the contacts most likely to respond

  • Inbound/Outbound Wastage - Your outbound sales team is spending equal time on prospects with very different chances of converting

  • Proactive Retention - You want to spot customers at risk of disengaging so you can intervene before they churn


How I work

Data Audit & Alignment

  • Map available data sources and confirm linkability via persistent ID

  • Agree target behaviour and prediction window

  • Assess data volume, quality, and coverage

Build & Compare

  • Engineer features & estimate multiple models

  • Evaluate and compare models performance

  • Select best-fitting model for data and objective

Validation & Sign-Off

  • Stress-test outputs & validate with business team

  • Agree final model & commercial threshold

Scoring & Deployment

  • Score full customer or prospect base

  • Deploy to CRM/CDP if required

  • Agree refresh cadence

Solution are tailored to business context, data maturity, and specific behaviour you need to predict.


FAQ

First-party data, such as transactions, CRM records, and website activity provide the most accurate and commercially valuable foundation for analysis.

It’s important that data sources are linkable across platforms through a persistent customer identifier.

Sometimes not all of these data feeds are available or usable. We work with what exists, flag meaningful gaps, and scope the model accordingly.

Platform signals are proxies derived from behaviour within a single ecosystem and are limited to the platform itself.

A propensity model built on your own first-party data provides a more holistic view that is not tied to any platform’s attribution logic.

The model ranks every contact by their likelihood to respond, but where you draw the line depends on your situation. A business managing a tight cost per acquisition sets a different boundary to one prioritising volume or protecting an existing customer relationship. We agree that threshold based on your commercial objective, data, and appetite for risk, so the output is calibrated to your specific situation rather than a generic default.

Yes. Scoring the customer base and deploying those scores into your existing infrastructure is available as part of the engagement. The approach will vary depending on your tech stack.

Typically six to ten weeks, depending on data readiness and the complexity of the behaviour being predicted. Well-structured, accessible data moves things along quickly; gaps in quality or access extend the timeline.

Customer behaviour shifts, and a static model loses accuracy over time. A regular refresh, typically quarterly or bi-annual, keeps scores relevant and predictions reliable.

No. Analysis is conducted using open-source frameworks whenever feasible, keeping work portable, transparent and easy to maintain & scale. This means no vendor lock-in, transparent code, and extensibility as your internal capability grows.


Previous projects