Jatin Hans Kvantum Inc · Aug 2017 – Mar 2021 · Delhi
Keeping customers from leaving
Segmentation, a 7-day propensity model and timed retargeting that moved churn from 40% to 23%, on a base large enough that a point is a lot of people.
Role
Sole technical lead
Base
5M+ customers
Outcome
40% → 23% churn
The stakes
Churn prediction is the most over-solved problem in applied data science and the most under-delivered. Predicting who will leave is straightforward. Changing whether they leave, at a cost per contact that does not wipe out the value of the retention, is the actual job, and the two are not the same objective.
At five million customers, every intervention has a real budget attached. The client was sending retention offers to everyone, or at fixed points in the calendar: some customers got an offer they didn't need, and others got it after they had already gone.
The system, in three stages
1 · Group the customers
clustering on profile and behaviour, into segments the marketing team could name
2 · Predict the next order
a 7-day propensity model, the next-best-action signal
3 · Retarget in the window
a profile rule sets the window; retargeting lands just before the churn point
Three decisions, and what I turned down
01
Time the contact
A propensity model scored each customer's chance of ordering in the next 7 days, and a rule on each customer's profile, including predicted moves between segments, set how long they could still be retargeted and when they counted as churned. Retargeting landed just before that point.
Rejected: offers to everyone, or at fixed points in the calendar. Easy to schedule, and blind to when each customer is actually about to leave.
02
Segments and a score, not one black box
Clustering gave segments the client's marketing team could recognise and name, which made the work explainable to them. The propensity score did the ranking, and the rule turned both into campaign lists that ran in the client's own tools.
Rejected: one black-box churn model. Nothing the marketing team could explain, or run a campaign from.
03
Read it as a before-and-after
The 40%→23% compares the customer base before and after the programme. No group was held back, so anything else that changed over the same period is in that number too. Today I would keep a random group out of the retargeting from day one, and a second group on the old calendar, to prove the timing itself was what mattered.
What came of it
40% → 23%
churn on a five-million-customer base, before and after the programme.
SHAP
used to explain segment drivers to marketing teams, which is how the model came to shape campaigns instead of only scoring them.
This sat inside four years of consulting across CPG, retail, pharma, FMCG and QSR for brands including Coca-Cola, Ross Stores, Abbott, Reynolds and Garden of Life, doing marketing mix modelling, attribution, cross-channel models and market-basket analysis. Different industries, the same discipline: decide what you would have to see to be wrong, before you build.