Predictive Loyalty Analytics — ProActive Promotions
Client
ProActive Promotions, a Toronto-based loyalty programme operator managing reward programmes for retail and hospitality brands across North America.
Problem
The loyalty programmes ProActive operated had a chronic problem: members stopped redeeming, then stopped shopping, and the operator only noticed when the numbers landed in a quarterly report. Win-back campaigns fired after members had already gone quiet for weeks — late, broad, and expensive. ProActive wanted to know which members were about to lapse, and when, so brands could intervene while it still mattered.
Solution
Gravitas India built a predictive churn model over ProActive's loyalty transaction history. The model scored every member every week on the probability of lapsing within the next 30 days, using transaction recency, frequency, basket value, redemption behaviour, and seasonality. Each brand received a weekly list of at-risk members with the strongest predictive signals — and a recommended intervention window, not just a name.
Implementation / Technical Approach
- Survival-style target definition — a member was defined as lapsed at a brand-specific inactivity threshold, so the model predicted a real business event, not an arbitrary cutoff.
- Weekly scoring pipeline — member scores recomputed nightly, delivered as a weekly at-risk queue per brand, with the reasons driving each score.
- Intervention timing model — the model output not just churn probability but the optimal outreach window, so win-back offers landed while members were still warm.
- Campaign measurement loop — win-back outcomes were tracked back into the model, so the system learned which interventions actually retained members.
- Brand-agnostic design — the same engine served retail, hospitality, and QSR programmes with per-brand calibration.
Technology
Predictive Modeling Churn Analytics Survival Analysis Automated Scoring
Outcomes
Across the brands ProActive operated, member lapse rates fell 18% within two quarters, and win-back campaigns aimed at predicted at-risk members drew 2.4× the response of the old blanket blasts — at 31% lower cost because fewer offers went to members who were never going to lapse anyway. The weekly queue became the default starting point for every brand's retention conversation.