Lift Conversion
Predictive models that estimate each customer’s likelihood of converting in real time.
When a customer puts a product in their basket or saves a quote, you have a window of a few seconds to understand whether they will buy on their own or whether they need a nudge — and what kind of nudge. We build predictive models that assign to each customer, at that specific moment, a probability of conversion based on their history and behavioural signals. From there, every decision becomes precise: who to offer a discount, who not to, who is comparing products and who is already ready to buy. Not generic funnel optimisation — intelligence on the individual customer, in the moment that counts.
What we build together
We build predictive models that estimate the probability that a specific customer will convert, based on behavioural, historical and contextual signals. Not industry benchmarks — your customer, on your data, with an individual and updatable estimate over time.
Based on the models, we define where and how to intervene — on which segments, at which point of the funnel, with which message or commercial lever. UX, copy, pricing, timing: everything becomes a data-informed decision, not a hypothesis tested blindly.
We design or rewrite the tracking architecture to capture the behavioural signals that feed the models. The quality of the input data determines the quality of the output prediction.
We build the system that evaluates, transaction by transaction, whether to offer a specific customer a discount — and which one. The model cross-references the estimated probability of conversion at that moment with the customer’s historical behaviour: someone comparing products, someone who has already decided, someone who forgets their basket and someone who is looking elsewhere each receives a different proposition — or none at all. The result is a discount policy that protects margin where there is no need to push and converts where the customer was about to leave.
Models are monitored over time and recalibrated when customer behaviour changes. A predictive system that does not update loses accuracy — and value.
How we do it
1) Behavioral Audit
We analyse the existing behavioural data and the current tracking architecture. We identify what is already being captured, what is missing and what is producing noise instead of signal.
2) Data Preparation
We clean, structure and integrate the data sources needed to build the models. This phase determines the solidity of everything that comes after.
3) Model Building
We build the purchase propensity predictive models, calibrated to your specific commercial context and historical data.
4) Activation
The models enter operational processes — marketing, sales, product — with outputs that are readable by those who need to make decisions, not only by those who built them.
5) Handoff and governance
The internal team receives the system with the documentation needed to interpret, monitor and update it. Dependency on us is not a contractual condition.
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 — the first guides immediate conversion decisions, the second guides retention strategies over time.
No. Purchase propensity models apply to any context where a measurable purchase decision exists — e-commerce, SaaS, financial services, utilities. The logic changes; the method does not.
Not perfect, but sufficient. The initial Behavioral Audit assesses the quality of the available data and defines what needs to be integrated before proceeding with the models.
A first complete cycle, from audit to activation, is generally between twelve and sixteen weeks, depending on the complexity of the context and the quality of the starting data.
For B2C organisations and e-commerce platforms with sufficient traffic volumes to train individual-level predictive models (typically above 10,000 transactions per month), with a marketing and CRM team capable of operationalising segmentations and dynamic discount policies. Typically with an existing tracking architecture — even an imperfect one — and a willingness to move from generalised discounts to targeted interventions on the individual customer.
When transaction volumes are below the threshold of statistical significance for predictive models, or when a minimum tracking system is not in place (in that case, the starting point is Ground Data). It is also not the right choice for those expecting measurable results in under 8 weeks, or for those looking for a generic “CRO solution” rather than a structured system.
Let’s talk about your
next project.
Thirty minutes to understand where to start and what makes sense to build.