Know who will come back, how much they will spend, and when.
AI models trained on your own purchase and engagement history predict each customer’s next 12 months of value, their risk of lapsing and the week they are most likely to buy again.
Every piece, and what it does.
Predicted 12-month value
A rupee value per customer, with a confidence band, refreshed every week.
Churn risk
Low, medium or high risk of not buying again, based on how far each customer is from their usual rhythm.
Next purchase window
The likely week of the next purchase, including festive and sale-season patterns such as Diwali week.
Category affinity
The category each customer is most likely to buy next, useful for picking the right product in a message.
Explanations
Every prediction lists its top drivers, such as visit frequency or recent return, in plain words.
Accuracy you can check
Back-tests against your last 6 months show how close predictions came, before you rely on them.
Three steps, snapped together.
Learn from your history
Models train on at least 12 months of orders and engagement from your connected sources.
Score every customer
Each profile gets predicted value, churn risk, next purchase window and category affinity.
Use it everywhere
Predictions appear on profiles, as segment rules and as fields synced to your tools.
What is in the box.
The facts your CRM, data and legal teams will ask about.
- Minimum history
- 12 months of orders and at least 5,000 repeat customers
- Refresh
- Weekly retrain, daily scoring
- Outputs
- Predicted value, churn risk, next purchase window, category affinity
- Inputs
- Purchases, visits, returns, engagement and consented channels
- Excluded inputs
- Religion, caste, health, and other sensitive data
- Plan
- Included from 500 Pieces
Responsible AI
- Predictions use purchase and engagement behaviour only. Religion, caste, health, political views and other sensitive data are never used as inputs.
- Names are not used to infer gender, community or any other personal attribute.
- Each prediction shows its main drivers, so your team can see why a customer was scored the way they were.
- Models are trained on one brand’s data only and are never shared with, or improved using, another brand’s customers.
How accurate are the predictions?
It depends on your data. Before predictions go live, a back-test on your own last 6 months shows the error range, so you decide whether to use them.
We are a new brand. Can we still use it?
Predictions need about 12 months of history and 5,000 repeat customers. Until then, Customer Insights gives rule-based value tiers.
Can we turn predictions off for some customers?
Yes. Customers who object to profiling can be excluded from scoring in Consent Center.
The pieces that fit next to it.
The last piece
Your customer base is one picture.
It is already in your POS, your store, your app and your WhatsApp chats. Connect two sources and watch the first pieces join up.
- 2sources free
- 25,000profiles free
- 0card details needed