Referral fraud is the gap between the customers a referral program is meant to attract and the people who actually collect the reward. Its simplest form is self-referral: one person opens a second account, applies their own code, and takes the bonus. No botnet, no stolen cards — a second email address and five spare minutes.
That matters because rewards are usually paid in something valuable: cash, credit, a free month, a discount on a first order. A program that pays at signup rather than on real usage pays for mailbox creation — and the check most teams rely on, “has this email been used before,” is the easiest one to pass.
This guide covers the detection layer: which signals survive a new email address, how to turn them into allow, step-up or block decisions with Prynt Control, and how to run it in mirror mode before you withhold a payout from a real customer.
Why email and IP checks fail
Most programs verify three things: the email, the IP address and a cookie. All three are disposable. Mobile carriers put thousands of subscribers behind one address, offices sit behind a single NAT, and a VPN changes the answer in a click. A cookie vanishes the moment someone opens a private window or a second browser on the same laptop.
So the abuse is not an exploit. It is a person using your program as written, twice, from the same device. Any control that only looks at the account will miss it, because the account is the thing being duplicated.
Signals that survive a new email address
The useful question is not who this account is, but whether you have seen this device before.

Device-level identity
Prynt Identify returns a visitorId computed from device and browser characteristics rather than from a cookie, so the same laptop tends to keep its identifier after a cache clear. You also get VPN, proxy, Tor and datacenter signals plus tampering flags. Our post on device fingerprinting covers what is stable and what is not.
Account links without storing the data
Control keeps phone numbers and cards only as keyed hashes. That is enough to ask whether a new account’s card matches the referrer’s, and not enough to reconstruct either value.
Device trust at login
Control also returns deviceTrust on login — trusted, new or risky — from the same device history. A bonus claimed from a device that has never been seen, seconds before a login from a device that has, is a readable pattern.
Turn the signals into a decision
Control answers a use-case question with allow, step-up or block, a reason code and a suggested step. For referrals the use case is the bonus. Pseudo-code, not the exact API — the real call is in our docs:
const d = decide({
useCase: "bonus",
accountId: newAccount.id,
deviceId: visitorId,
phoneHash,
cardHash,
});
if (d.action === "block") return refuseBonus(d.reasonCode);
if (d.action === "step-up") return challenge(d.suggestedStep); // captcha | verify_email | card_or_passkey
return payBonus();
Note what the code does not do: Prynt never blocks anyone. Control returns a decision and your code enforces it, so you keep the thresholds and the exception list.
From signals to policy
A starting policy for referral payouts:

- New email, new device, no payment history: allow. This is most of your traffic.
- New email on a device already linked to a paying account: refuse the bonus, keep the account, which may still be genuine.
- Same card hash on referrer and referee: step-up with card_or_passkey, or hold the reward until the first paid invoice.
- Several accounts on one device claiming bonuses in a short window: step-up, then review the cluster. Our notes on fraud rings describe how such groups look.
- VPN or datacenter traffic plus a first-day claim: step-up with verify_email. Not a block — plenty of honest people use a VPN.
Step-up makes the cheap version of the abuse expensive while keeping the honest path short. A captcha costs a real customer ten seconds; a second identity costs a farmer another mailbox and another device.
Watch it before you enforce it
Control starts with a 14-day mirror period: it returns decisions and reason codes while nothing is blocked. Use those two weeks to compare its verdicts with what you already know.
- How many rewards would have been blocked?
- How many of those turned into paying customers later?
- Which reason code fires most, and is it the one you trust least?
Then enforce in stages: step-up first, block second, and only for reason codes the mirror period shows are clean. A block that lands on an honest customer costs more than the bonus it saves.
Make farming cost more than it pays
- Pay the reward after the first paid invoice, not at signup.
- Cap bonuses per device, not only per account.
- Meter per device and account, so a private window does not reset the count.
- Route every decision to Slack or your CRM. Prynt Automate does that without glue code.

FAQ
Can a determined person get around device fingerprinting?
Sometimes. Fingerprints are probabilistic, not absolute, and a motivated person can change devices or tamper with the environment. The practical goal is to make each extra identity cost real money and effort, with a step-up path for ambiguous cases.
Do we need to store phone numbers and cards to match them?
No. Control keeps both only as keyed hashes, so you can compare them across accounts without holding the raw values.
Should a suspected self-referral be blocked or stepped up?
Step up unless the evidence is unambiguous. Same device plus the same card hash is strong enough to refuse a bonus; a VPN alone is not. Keeping the account and refusing the reward is the default middle path.
Referral programs pay for introductions, not for inboxes. To see what Control returns on your own signup and bonus flows, start with the 14-day mirror period and read the decisions before you act on them: pryntid.com.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.