Predict Behaviour
Predictive models that identify customers at risk of churn and those most likely to repurchase.
When a customer starts slowing down their purchases, ignoring communications or reducing their average order value, those signals already exist in the data — but they arrive late, if they arrive at all. We build predictive models that study the behaviour of the individual customer over time and estimate in advance the probability that they will leave, return, grow in value or stop buying. The organisation stops chasing lost customers and starts intercepting those who are about to leave.
What we build together
We build models that estimate the probability that a specific customer will take an action relevant to the business — abandonment, repurchase, upgrade, value reduction. The model works at the individual level, produces actionable outputs and updates over time as the customer’s behaviour evolves.
A specialised layer of the predictive models, focused on identifying customers at risk of abandonment before that risk becomes reality. The system monitors signals of deterioration in the relationship and allows intervention at the right moment — not after the fact.
We identify the patterns and causes that move the relevant KPIs. Data stops describing the past and begins to inform future decisions on where and how to intervene.
How we do it
1. Data Audit
We assess the quality, depth and history of the available data. The feasibility and accuracy of the models depends entirely on this phase — we do not proceed without a clear map of the starting data.
2. Model Design
We define the modelling architecture in function of the objectives: which phenomena to predict, over which time horizon, with what granularity. Individual prediction and aggregate forecasting follow different logics and are designed separately.
3. Build and calibration
We build the models and calibrate them on the available historical data. This phase includes the first accuracy tests and validation of outputs with the team that will use them operationally.
4. Activation
The models enter business processes — with readable outputs, integrated into existing tools, usable by decision-makers without technical intermediation.
5. Handoff and monitoring
The internal team receives the system with the criteria to interpret its accuracy over time, recalibrate it when customer behaviour changes and update it when business objectives evolve.
Answers to your questions
Everything you need to know about how we work and how we can help.
The difference is in the question each model answers. Lift Conversion answers a question about the present: does this customer buy or not buy right now? Predict Behaviour answers a question about the future: will this customer stay or leave in the next 6-12 months? Both work on historical customer data, but with different time horizons and operational objectives.
It depends on the complexity of the model to be built. The initial Data Audit includes an explicit assessment of feasibility — if the available data is not sufficient for an accurate model, we say so before proceeding.
A first complete cycle, from Data Audit to activation, is generally between twelve and sixteen weeks. The models require a historical record of at least twelve to twenty-four months to produce reliable outputs.
For organisations with a customer history of at least 12–24 months and integrable behavioural and transactional data, and with a customer base large enough to train individual-level models. Typically with a CRM, retention or customer analytics owner who has the mandate to move from descriptive to predictive analysis, and the intention to act on customers before the behaviour happens — not after.
When the historical data is less than 12 months or the data cannot be integrated across different sources. It is also not the right choice for those looking for aggregate forecasting on business KPIs (a different type of modelling that we address within Govern KPIs) or descriptive clustering of the customer base (which we address within Earn Retention).
Let’s talk about your
next project.
Thirty minutes to understand what your data might already be telling you about the future behaviour of your customers.