Catch churn before it happens — win members back automatically.
AI models score every member's churn risk daily and trigger personalised win-back journeys at the moment of maximum recoverability — not months after the customer has gone.
30–60d
earlier churn detection
+24%
lapsed members recovered
daily
risk re-scoring
Section 01
The business problem
Most brands discover churn in a quarterly report — long after the customer has switched. Win-back budgets are then sprayed at everyone who lapsed, discounting people who left for good and ignoring those who could still be saved.
Why existing solutions fall short
- RFM snapshots identify churn after it happened, not before.
- One-size-fits-all win-back offers burn margin on unrecoverable customers.
- No feedback loop measures which win-back actually caused a return.
Section 02
How Fundle solves it
Fundle scores churn probability per member using visit decay, basket shifts, engagement drop-off and category signals — then fires the right intervention (offer, reminder, reward, human call) at the risk threshold where it still changes behaviour, and measures incremental recovery with control groups.
01
Daily churn scoring
Every member re-scored daily on behavioural, transactional and engagement signals.
02
Recoverability tiers
Separates saveable members from lost causes so spend goes where it works.
03
Trigger journeys
Risk-threshold triggers launch WhatsApp/SMS/push journeys automatically.
04
Incrementality proof
Holdout control groups prove the recovery was caused by the campaign, not coincidence.
How it works
From data to outcome — the agentic workflow.
Score
Churn-risk model runs on the unified member profile daily.
Segment
Members bucketed by risk × value × recoverability.
Intervene
Personalised win-back journey fires per bucket.
Measure
Control groups quantify incremental recovery and ROI.
Business impact & ROI
+24%
lapsed-member recovery
−35%
churn among high-value tier
2.8x
win-back campaign ROI
Integrations
FAQ
Churn Prediction & Win-Back — questions buyers ask.
How much history does the model need?
6–12 months of transactions gives strong performance; the model improves continuously as data accrues.
Does it work for low-frequency categories?
Yes — the model adapts inter-purchase-interval baselines per category, so jewellery churn looks different from grocery churn.
Can our team override the AI?
Always. Risk thresholds, offer depth and channel rules are guardrails your team sets; the agent executes within them.
Get started
See Churn Prediction & Win-Back on your data.
A 30-minute working session with the Fundle team — mapped to your brand, mall or programme.