Prompt
How do I catch promo abuse and refund fraud on an ecommerce site?
Latest observation
To catch promo abuse and refund fraud on an ecommerce site, you want a layered approach: identify suspicious patterns, score risk in real time, and apply controls that are annoying to fraudsters but low-friction for good customers.
1) Define the abuse patterns you care about
Promo abuse
Common forms:
- New-customer coupon farming: same person repeatedly creating new accounts to reuse first-order discounts.
- Multi-account abuse: many accounts sharing the same device, payment instrument, address, IP, or browser fingerprint.
- Synthetic identity / email aliasing:
name+1@gmail.com, disposable emails, or typo variations. - Promo stacking: combining offers contrary to terms.
- Referral abuse: self-referrals or ring referrals.
- Bulk redemptions: same promo code used unusually often from one network/device cluster.
Refund fraud
Common forms:
- Wardrobing / rental abuse: buying an item, using it, then returning it.
- Item not received / chargeback abuse: false claims after delivery.
- Empty box / wrong item claims: customer claims parcel contents were missing or incorrect.
- Serial returners: unusually high return rate or “refund-only” behavior.
- Return policy gaming: repeated returns just before deadline, switching items, or returning worn/damaged goods.
2) Build the right signals
Account and identity signals
- Account age
- Email domain quality / disposable email detection
- Phone verification status
- Address stability
- Name/address/payment mismatches
- Same device across multiple accounts
- Same IP/VPN/proxy/TOR usage
- Browser/device fingerprint similarity
- Number of accounts per household or device cluster
Transaction signals
- First order discount usage
- Order velocity
- Basket composition anomalies
- Shipping/billing mismatch
- High-risk SKUs or categories
- Gift card usage
- Rush shipping on first order
- Repeated use of the same card across “new” accounts
Refund/return signals
- Return rate by customer, SKU, and category
- Time from purchase to return
- Return reason frequency
- Claims against delivery status
- Rate of “not as described” or “damaged” claims
- History of partial refunds, disputes, or chargebacks
- Condition of returned item vs original shipped item
- Return shipping address reuse or mismatch
Network and device signals
- IP reputation
- ASN and hosting provider detection
- Geo velocity
- Multiple accounts from same device/browser fingerprint
- Session anomalies: rapid checkout, form autofill, repeated failed attempts
3) Create abuse rules and risk scoring
Use a mix of:
- Hard rules for clear abuse
- Risk scoring for fuzzy patterns
Examples:
- Block promo redemption if:
- disposable email + new account + same device as prior promo abuser
- Flag refund for manual review if:
- first-time buyer + high-value item + expedited shipping + claim within 24 hours
- Deny or step-up verify if:
- many accounts redeem same promo code from one device cluster
A simple weighted score could include:
- +30 if device linked to >3 promo redemptions
- +20 if IP is proxy/VPN
- +25 if account age < 1 day
- +20 if first order with deep discount
- +30 if return rate > 50% over last 5 orders Then define thresholds for:
- allow
- challenge
- manual review
- block / refund hold
4) Put controls in place
Promo abuse controls
- Limit promo to:
- one per customer
- one per payment method
- one per device cluster
- one per shipping address where appropriate
- Require phone verification or 2FA for risky redemptions
- Block disposable email domains
- Delay code issuance or redemption for high-risk users
- Use unique, single-use codes rather than broad public codes
- Restrict referral rewards until the referred order is fulfilled and not returned
Refund fraud controls
- Tiered return policy by risk:
- trusted customers: easy returns
- risky customers: require photo/video proof, restocking inspection, or label tracking
- Hold refunds until item is scanned at warehouse
- Require serial number / IMEI / lot number tracking for electronics
- Use tamper-evident packaging and weight checks on return
- Compare return weight to shipped weight
- Require return reason selection with evidence for high-risk cases
- Flag customers with repeated “item not received” claims
5) Monitor for ring behavior
Fraud is often coordinated:
- Many accounts tied to one device, address, or IP range
- Same basket patterns across accounts
- Similar timing: all redeem shortly after promo launch
- Shared return claims or chargebacks after purchase
Build cluster analysis:
- customer graph
- device graph
- payment graph
- address graph
Then look for:
- dense clusters
- hub nodes
- repeated cross-linking among accounts and payment instruments
6) Use anomaly detection and segmentation
Compare customers within cohorts:
- New vs returning
- Geography
- Product category
- Promotion type
- Device type
- Acquisition channel
Fraud often stands out when compared to peers:
- unusually high discount usage
- unusually high refund rate
- unusually fast refund requests
- repeated use of first-order offers
7) Add operational review workflows
Don’t rely only on automation. Create queues for:
- high-value promo redemptions
- suspicious first orders
- suspicious refund claims
- repeat claimants
Reviewers should see:
- linked accounts
- order history
- device/IP history
- promo usage history
- delivery scans
- return history
- prior disputes/chargebacks
8) Track metrics
Measure:
- promo redemption abuse rate
- refund rate by cohort
- chargeback rate
- false positive rate
- manual review hit rate
- recovered loss
- customer friction / conversion impact
You’ll want to tune thresholds so you reduce abuse without blocking good customers.
9) Be careful with privacy and fairness
Avoid using protected attributes directly or indirectly inappropriately. Make sure your fraud controls:
- comply with local laws and payment rules
- are documented and explainable
- allow appeals where appropriate
- don’t over-target legitimate users from certain regions or networks
10) A practical starter stack
If you’re starting from scratch:
- Log everything: account, device, payment, promo, refund, shipping, chargeback.
- Implement basic rules:
- one promo per customer/payment/device
- disposable email block
- proxy/VPN detection
- refund hold for high-risk orders
- Build a risk score with 10–20 signals.
- Add cluster detection for linked accounts.
- Create a manual review queue.
- Iterate based on losses and false positives.
If you want, I can also help you design:
- a fraud rules engine
- a risk scoring model
- or a database schema / SQL queries to detect promo abuse and refund fraud.