Data-driven predictive model based on customer historical behavior
1. Teranalytics developed an algorithm to predict customer churn based on internet traffic data, feature usage, and payment history. Customers with highest churn likelihood are flagged with a lead time sufficient to implement marketing action in line with the drivers identified and selected by the custom algorithm. Efficiency has been demonstrated by comparing a control group (no action) vs a group where marketing outreach has been implemented as per the algorithm recommendation.
2. Teranalytics centralized thousands of B2B time series reflective of customer behavior across AMER, LATAM, MEA, and APJ regions, merged these data with publicly available financial reports, and derived a series of financial metrics to quantify behaviors along multiple dimensions.
3. The resulting data were used to build an AI predictive engine that provides:
(1) A probability of churn with a certain lag parameter (e.g. predicting churn at 3 months, at 6 months, etc).
(2) A driver analysis highlighting the most likely reason for the customer churn.
4. The findings, based on a Random Forest classification coupled with financial analysis, can be used to tailor marketing strategies to specific customer needs.