Predicting customer value with 99.91% accuracy
A UK platform guiding first-time buyers through the mortgage journey asked a simple question: of the thousands moving slowly through the funnel, who will actually convert? Using only the platform’s own behavioural data, Barrington built a model that answered it — with precision.
The challenge
In life-and-mortgage journeys the path to conversion is long and uneven. Drop-off is invisible until it is too late, rule-based segmentation lumps very different customers together, and marketing spend is spread evenly across a base where the top few percent are worth vastly more than the rest.
What we did
We applied a decision-intelligence layer on top of the platform’s existing data — learning from real behavioural patterns rather than predefined rules, and designed to feed existing systems rather than replace them.
- Segmented 38,052 customers into three behaviourally distinct groups.
- Found the top segment — 6% of the base — converting at 17.19%, more than 120 times the early-stage rate.
- Trained a predictor reaching 99.91% accuracy and 99.29% recall on high-value customers.
- Structured the engagement to prove value before scaling, with no disruption to existing workflows.
The outcome
Value in the funnel is now measurable and heavily concentrated — and the business knows exactly where. Spend and broker time concentrate on the customers most likely to convert, with a modelled uplift of around five times from targeted, AI-prioritised campaigns.
The top 5% of customers convert roughly 120× more than the lower segments. Knowing exactly who they are changes where every pound and every broker hour should go.