Promo Abuse: How to Detect and Stop Coupon and Referral Fraud
Promo abuse is one operator posing as many new customers. Four live IP lookups show how the redeeming address changes what your coupon flow should do.

Promo abuse is one operator posing as many new customers to collect coupons, referral bonuses and signup credits. You catch it at the redemption step, because every fake account still has to connect from a network address, and the type of that address decides whether you honour, hold, challenge or decline the incentive. Below: how the schemes work, the responses we would give four real redeeming addresses we looked up on 2026-10-23, the signals behind an IP fraud score and its allow-to-block bands, and where the IP layer goes blind.

What counts as promo abuse
The label covers six distinct schemes, and each calls for a different response:
New-customer coupon farming
- How it works: One operator creates repeat "first" accounts to reuse intro discounts
- Typical target: Ecommerce, delivery, subscriptions
Referral rings (self-referral)
- How it works: The same operator controls referrer and referee, so every fake signup pays a bonus twice
- Typical target: Marketplaces, fintech, gig platforms
Signup and bonus farming
- How it works: Mass account creation to harvest credits, welcome bonuses or airdrops
- Typical target: Fintech, iGaming, crypto
Code enumeration
- How it works: Bots try guessable voucher codes until valid ones respond
- Typical target: Any site with short or sequential codes
Stacking and glitch exploitation
- How it works: Combining codes or exploiting redemption logic beyond intended limits
- Typical target: Retail checkout flows
Loyalty and points farming
- How it works: Fake or automated activity to build up redeemable points
- Typical target: Airlines, hotels, retail loyalty
OWASP catalogs the automated versions in its Automated Threats project. OAT-002 Token Cracking is the "mass enumeration of coupon numbers, voucher codes, discount tokens, etc.", and OAT-019 Account Creation is to "create multiple accounts for subsequent misuse". Five of the six schemes in the table are built on those two primitives. Stacking is the exception, because it abuses checkout logic rather than identity.
Why coupon and referral fraud is scaling now
Coupon and referral fraud is scaling because the losses now run to billions and the tooling is sold as a service.
The money is serious. Juniper Research projects fraud on digital goods will grow from $10.4 billion in 2025 to $27 billion by 2030, a 162% rise, and names promo abuse as one of three key drivers, alongside synthetic identities and friendly fraud (Juniper Research, December 2025). Each lost dollar also costs more than a dollar. The LexisNexis True Cost of Fraud study puts the average cost for US retail and ecommerce merchants at $4.61 for every $1 of fraud once labor and overhead are counted (LexisNexis Risk Solutions, April 2025).
The tooling has industrialized too: account-creation bots, anti-detect browsers, aliased email and rented proxy pools let one operator pose as many customers.
Inside the playbook: how one operator becomes many new customers
Every playbook ends the same way: many accounts, one operator and shared infrastructure underneath. Two public cases show both ends.
The identity-theft end. In May 2021, federal prosecutors in Massachusetts charged 19 people over a scheme that ran from January 2019 to at least April 2021. It used more than 2,000 stolen identities to open driver accounts on rideshare and delivery apps, created fake accounts whose "sole purpose" was referring other fake accounts, and used "automated bots and GPS spoofing technology" to inflate earnings (NBC Boston, May 2021). In 2022 two defendants pleaded guilty to conspiracy to commit wire fraud and aggravated identity theft. A separate San Diego prosecution of a five-person driver-account ring ended with the sentencing of its leader.
The pure multi-accounting end. Token airdrops pay per eligible wallet, so farmers spin up wallets instead of identities. After its June 2024 airdrop, zkSync defended its sybil filtering by arguing that being more aggressive "might falsely flag many organic users" (Cointelegraph, June 2024). That is the same trade every promo team faces. Tighter filters catch more farmers and refuse more real customers.
Both share a weak point. Fake identities are cheap and emails are free, but every account still connects from somewhere. Clean, distinct consumer network locations are the one resource a ring cannot mint on demand, so it reuses infrastructure, and reuse shows up in your logs.
What four redeeming IPs looked like on 2026-10-23
The same promotion needs four different responses for four address types, and the fraud score's own recommended_action is not the promo decision in two of them. On 2026-10-23 between 05:03:02 and 05:03:17 UTC we ran keyless lookups on four public addresses, one per type a coupon or referral endpoint sees. None of them sent us real redemptions. They stand in for the origin types:
45.33.32.156 (Linode server, scanme.nmap.org)
fraud_score: 10recommended_action:allow- Methods that fired:
datacenter_ip - What the coupon or referral flow should do: Hold for review. Let the sale through at full price and queue the discount
185.213.155.73 (Mullvad server, Frankfurt)
fraud_score: 100recommended_action:block- Methods that fired:
known_vpn_server,wireguard_open,datacenter_with_vpn_proxy,no_ptr_datacenter - What the coupon or referral flow should do: Step up. Ask for a card or phone check before the incentive applies
104.28.46.99 (iCloud Private Relay egress)
fraud_score: 9recommended_action:allow- Methods that fired:
relay_ip - What the coupon or referral flow should do: Honour, and count it in the per-prefix velocity rule
185.220.101.1 (Tor exit, Hamburg)
fraud_score: 100recommended_action:block- Methods that fired:
tor_exit,datacenter_blocklisted,no_ptr_datacenter,known_spammer - What the coupon or referral flow should do: Decline the incentive. The full-price sale can still go through
The two mismatches are the point. The Linode server scores 10 and allow, because a plain cloud server is not fraud: APIs, integrations and monitoring run from clouds all day. A first-order coupon, though, has no honest reason to be claimed from a rack. On 2026-10-23 the Mullvad exit scored 100 and block, because the API matched it on Mullvad's own server list, which tags it as a WireGuard server. Since our 10 November 2026 scoring change a VPN with no abuse, proxy or Tor evidence stops at 70, so the same server returned 70, step_up, on 11 November, which matches the step-up this flow asks for. But Security.org's 2026 VPN report, a survey of 1,008 US adults in August 2026, found 42% use a VPN, so a VPN redemption is often a real customer. A card or SMS check costs that customer seconds and costs a ring one payment card or phone number per account.
Here is the trimmed lookup for the hosting address (we leave out its stored PTR, which does not match live reverse DNS):
{
"ip": "45.33.32.156",
"asn": {
"asn": 63949,
"org": "AKAMAI-LINODE-AP - Akamai Connected Cloud",
"asn_type": "hosting",
"connection_type": "datacenter"
},
"detection": {
"is_proxy": false,
"is_vpn": false,
"is_tor": false,
"is_hosting": true,
"is_relay": false,
"anonymity_level": "low"
},
"threat": {
"is_cgnat": false
},
"scoring": {
"fraud_score": 10,
"confidence": 0.25,
"recommended_action": "allow",
"detection_methods": ["datacenter_ip"]
}
}
Your rule therefore has to read detection.is_hosting and asn.connection_type directly, because a rule that waits for block would pay this coupon. Earlier versions of this post used a DigitalOcean droplet, 104.248.50.21. On 2026-10-23 it scored 12 with datacenter_ip plus fcrdns_mismatch, a reverse DNS signal unrelated to coupons, so we switched samples.
The relay egress is the opposite case:
{
"ip": "104.28.46.99",
"asn": {
"asn": 13335,
"org": "CLOUDFLARENET - Cloudflare, Inc.",
"user_type": "consumer_privacy_network"
},
"detection": {
"is_proxy": false,
"is_vpn": false,
"is_hosting": false,
"is_relay": true,
"anonymity_level": "medium",
"relay_provider": "iCloud Private Relay"
},
"scoring": {
"fraud_score": 9,
"recommended_action": "allow",
"detection_methods": ["relay_ip"]
}
}
is_relay is true while is_proxy and is_vpn are false. This is an Apple user with iCloud+ Private Relay on, and one relay redemption is a customer. The Tor exit carried anonymity_level: "high" and a Spamhaus ZEN listing, which is why declining the incentive there costs little.
Shared addresses get a velocity rule, not a block
Shared addresses can never be blocked by score alone in GeoIPHub, by design. A carrier-grade NAT address is capped at 50, inside review, and an iCloud Private Relay address is capped at 60, inside step_up. Blocking either would block every real customer behind it. So the ring hiding there has to be caught by counting, not scoring:
- IPv4: count redemptions of one promotion per /24 per hour. Our suggested starting point, not a measured industry rate, is three: hold the fourth and later redemptions from one /24 for review, then tune the number on your own outcomes.
- IPv6: count per /64, not per full address. One subscriber usually holds the whole /64, so rotating inside it is free. Keying on full addresses turns one actor into thousands. Homes and businesses often hold a /56 or /48, so widen to those when fresh /64s keep arriving from one allocation (see rate-limiting and blocking IPv6 by /64).
- Relay and CGNAT: keep the same counters, but raise the threshold, because one relay egress or carrier address legitimately serves many customers.
The IP signals that surface promo abuse
Three IP signals surface promo abuse: shared origin across "different" customers, hosting origin and anonymization concentrated at redemption. Each reveals coordination that per-account checks miss.
Shared origin across "different" customers. Many new accounts redeeming one promotion from one address, /24 or /64 is the bluntest signal, and the velocity rules above turn it into a decision.
Hosting ranges where consumers should be. Real new customers redeem from home broadband and mobile lines. Redemptions from datacenter IP ranges, meaning cloud instances, VPS providers and hosting ASNs, are scripts until proven otherwise. GeoIPHub classifies hosting from 853 curated hosting ASNs plus official ranges from 10 cloud providers, which makes this the cheapest first-pass filter on promo eligibility.
Anonymization concentrated at redemption. VPN and Tor flags on ordinary browsing are weak evidence. The same flags spiking on referral submissions and code redemptions are different: the operator is rotating exit points to beat the shared-origin check.
Shared IPs are normal, so treat these as reasons to look closer. Cloudflare found CGNAT IPs are rate-limited three times more often than other IPs, although the median bot rate is nearly identical, 4.8% against 4.7% (Cloudflare, October 2025). A CGNAT flag should dampen your response, not amplify it.
Where the IP layer goes blind: residential proxies and click farms
An IP lookup cannot see a redemption that arrives through a residential proxy or from a click-farm phone on a real home or mobile line. Those connections come from consumer ISP addresses, which is exactly what a real new customer looks like. GeoIPHub does not detect residential proxies. A home ISP address can even pick up the residential ISP signal, which lowers the score by 20.
A ring that rents residential exits has paid to beat network checks, so the evidence has to come from elsewhere:
- Velocity on the account graph. Many referees credited to one referrer within hours, whatever their IPs.
- Device overlap. Accounts that share a device fingerprint despite different addresses.
- Payment and contact overlap. One card, one bank account or one phone number behind several "new" customers.
- Cohort quality. Redeemers who never make a second purchase.
For why rented residential exits are the hardest origin to judge, see residential proxy fraud and why trusted IPs are now risky. For the general version of this problem, multi-accounting detection covers how one user becomes fifty.
A detection playbook you can ship this week
Detection works when you score three events, act on the incentive and review rings rather than accounts.
- Instrument three events, not one. Log the IP and its signals at signup, referral submission and redemption. Most teams only score signup (how to stop fake signups covers that check), while abuse concentrates at redemption.
- Score the referrer and the referee. In a self-referral loop both sides are the same operator. A referrer whose recruits cluster on one /24, one ASN or one VPN provider is the ring's own bookkeeping.
- Run velocity per /24 and per /64 per promotion. Count per campaign, not globally, as described above.
- Gate promo eligibility, not purchases. Hosting origin, Tor exits or known-abuse history at redemption should withhold the discount while a full-price sale proceeds. That removes the profit without creating support tickets.
- Send the middle bands to verification. A score from 26 to 75, or a VPN flag at any score, warrants a card, phone or challenge step, the same logic as per-action fraud score thresholds.
- Review clusters, not accounts. Queue groups that share an IP, prefix or device and judge the ring in one pass. Investigating an IP address step by step shows how to read one address's evidence before you act on its cluster.
- Measure cohort quality. Track redemption-to-second-purchase per campaign. A promotion whose redeemers never come back shows where the budget went.
Wire the redemption check in first
Coupon and referral abuse is multi-accounting pointed at your marketing budget. Start with one lookup at redemption: read is_hosting, is_vpn, is_relay, is_tor and scoring.detection_methods, map them to honour, hold, step up or decline, and add the per-prefix counters. The free plan covers 1,500 lookups a day with every field, and the IP fraud score checker shows the same score and reasons for a single address without code. The fraud prevention overview explains how the score is built.
The same signals defend adjacent doors: the anonymization layer in how VPN and proxy detection works, account security in detecting account takeover and credential stuffing, and false-positive discipline in detecting anonymized traffic without blocking real customers. Wire the redemption check in first. It is the cheapest place to turn a bounty program back into a growth budget.
Frequently Asked Questions
What is promo abuse?
Promo abuse (promotion abuse) is the exploitation of marketing incentives such as new-customer coupons, referral bonuses, signup credits, loyalty points and free trials by people who were never the intended recipients. It ranges from one customer creating a second email address to reuse a first-order discount, up to organized rings that run thousands of fake accounts to drain referral and bonus budgets.
Is promo code abuse illegal?
Often it is a terms-of-service violation rather than a crime, which is why platforms handle most of it with detection and account bans. It becomes criminal when it involves identity theft, forged documents or coordinated deception at scale. In 2021 US federal prosecutors in Massachusetts charged 19 people over a rideshare and delivery driver-account scheme that used more than 2,000 stolen identities and created fake accounts whose sole purpose was referring other fake accounts. Two defendants pleaded guilty in 2022 to conspiracy to commit wire fraud and aggravated identity theft.
How do people abuse referral programs?
The core move is self-referral through multi-accounting: the same operator controls both the referrer and the 'new customer', so every fake signup pays out a bonus. Rings scale this with aliased emails, anti-detect browsers, device farms, VPNs and rented residential proxies. An IP lookup cannot tell a residential proxy exit from a real home user, so those accounts have to be caught by velocity, device and payment overlaps.
Which IP signals indicate promo abuse?
The strongest network signals are many 'different' new accounts redeeming the same promotion from one IPv4 /24 or one IPv6 /64 prefix; redemptions from hosting ranges where consumer traffic should come from home or mobile lines; and VPN or Tor flags concentrated at the redemption step. Each needs context. Carrier-grade NAT and iCloud Private Relay put many real users behind one address, so GeoIPHub caps CGNAT scores at 50 and relay scores at 60, and a velocity rule, not a block, handles those addresses.
How do I stop promo abuse without blocking real customers?
Act on promo eligibility instead of the account: when risk signals fire, withhold or hold the discount and bonus rather than rejecting the signup or the sale. Honour low-risk redemptions, hold hosting-origin redemptions for review, step up VPN redemptions with a payment or phone check, and decline the incentive on Tor exits. Score both the referrer and the referee, and treat shared addresses with a per-prefix velocity limit instead of a block.
The team behind the GeoIPHub API
Written by the GeoIPHub team, the people who build the GeoIPHub IP intelligence API and its scanner. Drafts are prepared with AI assistance and edited by the team; API examples are real lookups against the live API unless labelled illustrative.


