How to design referral rewards that are hard to farm
Referral program design abuse can empty a campaign budget overnight. How to set reward timing, caps and qualifying events, and where basic checks break.
Referral rewards that are easy to farm can empty a quarter's budget in one night. This morning your banking app says $71,400. Last night it said $86,400. Nothing was due to leave the account overnight, so you refresh it, and it says $71,400 again.
You spend $14,400 a month. At $86,400 that was six months of runway (86,400 ÷ 14,400), which is the number you gave your board last week. Now it's 4.96 months. So you go looking for the $15,000, working backwards.
It left at 6:00 a.m., in the nightly payout from the referral program you launched yesterday. That's 600 rewards at $25 each, the whole budget you'd set aside for the quarter, spent on the first night. You open the payout file and all 600 went to the same address.
The signups started at 9:40 p.m. and stopped at 5:10 a.m. That's 600 accounts in 450 minutes, one every 45 seconds, all through one referral link. You remember that chart. At 11:40 p.m. it showed 160 new signups in two hours, on a product that gets about 40 a day. Your co-founder posted it in the team channel, you sent a screenshot to your lead investor, and you went to bed happy. The spike you celebrated was one person.
Nobody broke in. The program did what it was built to do, and that's what makes referral program design abuse so expensive: the rules pay out faster than anyone can see who's collecting.
The $25 wasn't all of it. Each new account also got 2,000 welcome credits, about $3 of model usage, spent the moment they landed. Your provider limit was $2,600 for the month and you'd used $1,400, which left room for 400 accounts ($1,200 ÷ $3). Account 400 arrived at 2:40 a.m., five hours in (400 × 45 seconds). From that minute every generation in your product failed, for everyone. The last 200 accounts got errors too and still earned their $25, because the reward paid on signup.
You raise the limit at 8:40 a.m., after six hours of errors for the people paying you $49 a month. By lunch you've refunded 14 of them, and one has cancelled without saying why.
Add the payouts ($15,000), the model usage ($1,200) and the refunds (14 × $49 = $686), and the night cost $16,886. At $14,400 a month, that's 1.17 months, about 35 days of runway. The customers who shared their links yesterday are about to ask where their rewards are, and the budget is empty. By the end of the day you have to choose. You can put in another $15,000 you can't spare, or you can switch off a campaign you announced 24 hours ago and tell your best customers their referrals won't pay. Either way, the money from 6:00 a.m. is in someone else's account, and your investor still has the screenshot.
That morning is a scenario, so swap in your own numbers. The duplicates are real enough:
From our early beta
After a launch went viral on X, one team gained 6,400 new users in a month. 2,700 of them were duplicate accounts. About 32 became paying customers.
That's about 42% duplicates (2,700 ÷ 6,400). At this post's $25, as a what-if, those accounts would be owed $67,500 (2,700 × $25). At $49 a month, 32 customers (32 × $49 = $1,568) would take about 43 months to earn that back.
Why referral programs get farmed
A referral program pays for an event, and you get more of whatever event you pay for. If the event is a signup, producing one takes a few seconds and an email address, and it pays $25. One person can also be both the referrer and the new customer, so a single signup collects twice.
A cash reward takes money you already had, and it's harder to get back than credits are. The first fix most teams think of is a rule against referring yourself. Put it in your terms, because you'll need it to refuse a payout. But a second account doesn't announce itself, so the rule only helps once you've already spotted one.
Set the qualifying event so farming doesn't pay
Reward design does more than any check. Start with one sum:
If that number is positive, your program is paying people to farm it. Here's a $25 reward against different qualifying events, with illustrative prices:
| Qualifying event | Cost to qualify | Profit per reward | What honest referrals lose |
|---|---|---|---|
| Signup | Nothing | $25 | Nothing |
| First $9 invoice | $9 | $16, or $25 after a refund | A wait |
| First $49 invoice, no hold | $49, refundable | $25 after a refund | A wait |
| First $49 invoice, held past refunds | $49 | A loss of $24 | A month's wait for the reward |
On a $9 plan a $25 reward still pays someone to farm it, so the reward has to be smaller than the first payment. And a payment that can be refunded proves nothing until the refund window has closed.
Set the timing, caps and budget
Pay on money received. The first paid invoice is the cleanest qualifying event most SaaS products have. If there's no paid step, a usage milestone beats a signup, but it's cheap to fake.
Hold the reward past your refund window. With 14 days of refunds, a 30-day hold leaves room for one to arrive.
Move the new customer's bonus to the paid side. Money off a first invoice costs nothing until someone pays. Those 2,000 welcome credits cost $3 as soon as they land.
Cap each referrer. Ten rewards per referrer per month limits one account to $250. Raise it by hand for anyone real who writes in.
Make the payout job enforce the budget. A budget in a spreadsheet stops nothing. Over a 13-week quarter, $15,000 is about $1,154 a week, so a weekly limit in the job turns the worst night into $1,154.
Run the same night through these rules. 600 accounts sign up, none pays an invoice, and $0 leaves at 6:00 a.m. Collecting all 600 rewards would now mean paying 600 × $49 = $29,400 to receive $15,000. The price is that your real advocates wait a month, and some will lose interest.
It doesn't finish the job, though. If your plan costs less than the reward, or you reward free signups because that's the growth you want, the sum stops protecting you. A cap counts referrer accounts, so 60 accounts at 10 rewards each is the same 600. And a first-invoice discount for new customers can be claimed again by an existing customer on a new account. All of that comes down to telling whether an account is someone new.
Basic self-referral checks and how they break
Each of these catches something.
Email checks
Email verification proves someone can open an inbox. That catches typos and dead addresses, and says nothing about who's behind them. A unique-email rule stops the same address twice and costs almost nothing. A disposable-domain block catches known throwaway services, misses new domains and every ordinary mailbox, turns away people using privacy forwarding, and needs someone to update the list forever.
Then there's how email itself works. Gmail ignores dots, so jane.doe@gmail.com and janedoe@gmail.com reach one inbox, and mail to jane+anything@gmail.com lands there too. Treating those as one person catches one mailbox written several ways. That's one more rule to keep up to date. None of it touches a second real mailbox.
CAPTCHA, phone and card requirements
A CAPTCHA stops simple scripts and has limits of its own. It doesn't slow someone opening accounts by hand, and at one every three minutes, $25 a signup is $500 an hour. It also gets in the way of people using screen readers. A phone requirement stops casual second accounts, misses anyone with another number, shuts out people without a reliable mobile, and bills you for every text. A card requirement misses anyone with another card and loses people who want to try before paying. That last one costs a referral program more than most, because a referred friend hasn't decided anything yet.
One-per-IP limits and VPN blocks
A one-per-IP limit catches repeat signups on one connection, and it's the worst fit on this list. Referrals travel between flatmates and people in the same office, so you'd be blocking the exact signups you're paying for, and anyone on another network isn't affected. VPN blocks catch the ranges on your list, miss the rest, and land on privacy-minded people and staff on company VPNs.
Rate limits, usage caps and smaller rewards
A rate limit stops bursts from one source, and one signup every 45 seconds isn't much of a burst. A usage cap limits one account, and every new account arrives with a fresh one. A shorter campaign or a smaller reward lowers what you lose per account and what your honest advocates earn, and more accounts make up the difference.
Manual review
Reading the payout list catches 600 rewards to one address. It misses 60 accounts taking 10 rewards each to 60 addresses. At two minutes a referral, 600 referrals is 20 hours, again next campaign. And in a spreadsheet, the customer who brought in their whole team looks just like someone farming you.
Add them up and you've got a signup form with five hurdles for a friend of a friend, a list to maintain, an exception queue and a reviewer. Someone with a second real mailbox still gets paid.
Ask whether this claim is a repeat
One question is left, and no cap can answer it: does this claim look like a repeat?
AbuseZero was built for exactly that, and you don't have to build your own abuse prevention system to get it answered. Your server makes one API call before a reward is released and gets a clear allow or deny, under rules you control. With multi-layer AI risk analysis, AbuseZero examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. There's also a proprietary algorithm, with AI that checks the claim again. Where email rules ran out, AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing, and it checks mailbox aliases, disposable domains and mail routing. Where an IP limit gave up, it sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence, and signals are read together, never one at a time. It blocks spammy bots, browser automation and malicious scripts while allowing genuine AI agents and assistants. Creating a fresh account isn't enough to erase previous activity.
Keep your hold, your cap and your budget, because those are yours to set. Each reward becomes a benefit with its own eligibility, and Observe mode shows you what would be denied before anyone is blocked. The quickstart walks you through the first one.
Frequently asked questions
What's a good qualifying event for a referral reward?
The first paid invoice, held past your refund window.
Should referral rewards be cash or credits?
Credits can't be withdrawn, and they only cost you what it takes to serve them.
Is every referral from the same household abuse?
No. Families and colleagues are who referrals are meant to reach.
Protect your next referral campaign
Start with Builder
Pay as you go. Evaluate your first claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
Large campaigns and custom terms.