Trust & Risk · Fraud Defence
Every referral rupee is a target
Growth incentives attract abuse faster than they attract members. The defence agent scores at the moment of the payout, not in a monthly report, and a genuine member is stepped up rather than locked out.
Payouts blocked this month
1,840
Value protected
₹18.4L
False-positive rate
1.8%
reviewed weekly, appeals honoured
Genuine members locked out
0
step-up auth, not lockout
Signals and catches
What was stopped, and how
Shared sensor fingerprints across fresh installs during a referral push.
Graph cycles between invitee and inviter blocked before payout, ₹18.4L saved.
Velocity + card-BIN pattern, refunded automatically with a plain explanation.
Impossible-travel plus behavioural drift; step-up auth, no lockout.
Vendor score plus layout forensics; sent to a human, never auto-rejected.
Catch volume
Accounts or events stopped this month
Where a human always decides
- Account restriction or closure — the agent recommends, a named human signs.
- Synthetic-document rejections — reviewed by a person, because a real member with a bad scan looks identical.
- Appeals — a human reads them, and an upheld appeal becomes a labelled case for the model.