When referral farming breaks your growth loop, and how product fixes it

Every referral program starts with a nice idea. Your happiest users bring their friends, you pay a small reward for each one, and growth compounds. In a consumer app that mostly holds up. In a product that moves money, the first people who take the offer seriously are usually not your happiest users. They're the people who read the terms more carefully than you wrote them.

I ran into this at Lume, where I led product across onboarding, funding, trading and withdrawals for a wallet that gave people access to tokenized U.S. equities and perpetuals. We processed more than $4.1M in trading volume across 4,100+ users in the first four months. Getting there meant learning that a growth loop can look healthy on a dashboard while it quietly pays people to leave.

What farming looks like from the inside

Farming rarely looks like fraud at first. It looks like a great week. Sign-ups go up, referral codes get shared, and deposits arrive. Then you look a little closer. A cluster of new accounts funds the minimum, triggers the reward, and withdraws. Some of them never place a trade. Some place exactly one, sized to clear whatever threshold the reward asked for.

None of that is surprising once you see it. If the reward is paid on sign-up, people sign up. If it's paid on a deposit, people deposit and then leave. The program is doing exactly what it was told to do. The mistake is in what we told it.

In crypto-adjacent products the incentive is stronger than usual, because the rewards are often liquid and the audience is used to hunting for them. Airdrop culture trained a lot of people to treat any new product as a set of boxes to tick. That's not a moral failing on their part. It's a design input, and you should plan for it the same way you plan for network latency.

Pay for the behavior you actually want

The first fix is the simplest one to say and the hardest one to agree on internally. Tie referral eligibility to the behavior that makes someone a customer. For us that meant funding and trading, not signing up. A referral only counted once the new user had actually put money in and used the product for what it was for.

This makes your top-line referral numbers smaller overnight. That's the point. A smaller number that tracks real customers is worth more than a big one that tracks reward hunters, and it's a much better input for deciding how much to spend.

Cap rewards on what stays, not on what arrives

The second fix was net deposit caps. A deposit tells you money arrived. It doesn't tell you whether it stayed. Measuring against net deposits, what came in minus what went out, closes the loop where someone funds, collects, and withdraws the same afternoon. Capping rewards keeps any one account or cluster from turning the program into a yield product.

Caps also protect you from your own optimism. When a campaign is working, there's always pressure to raise the reward and pour more fuel on it. A cap forces that conversation to happen on purpose instead of by accident.

Give the reward time to be real

The third fix was delayed-settlement escrow. Instead of paying the reward the moment the conditions were met, we held it and released it after a settlement window. That gives you time to see whether the funding cleared, whether the trading was real, and whether the account looks like a person or like one of fifty accounts behind the same setup.

Users who are there for the product barely notice the delay. Users who are there for the reward notice immediately, and many of them move on. In a referral program, that's the filter working.

Fix money movement before you buy growth

The fourth lesson had nothing to do with rewards. Before a major acquisition campaign, we made funding and withdrawal reliability a prerequisite. If deposits are slow or withdrawals fail, paid growth doesn't just waste money. It brings a wave of new people into the worst part of your product, and they tell their friends about it.

In fintech, the first withdrawal is a trust event. Someone who gets their money out cleanly and quickly is far more likely to put more in. Someone who waits three days and opens a support ticket is gone, and you paid to acquire them. I'd rather delay a campaign by a few weeks than launch it on top of rails that can't take the load.

This connects to the metric I watched most closely at Lume, time to first funded trade, which we brought down from weeks to under 60 seconds. I wrote about that in From weeks to under 60 seconds. A referral program sits on top of that funnel. If the funnel leaks, rewards only make it leak faster.

Read referrals as cohorts, not counts

The last piece is measurement. We used Mixpanel funnels and behavioral cohorts to guide iteration, and referral traffic is exactly where cohorts earn their keep. The questions I care about are simple. Did referred users fund at the same rate as organic users? Did they trade more than once? Did their balances hold after the reward settled? Did they come back the following month?

If referred users look like organic users a few weeks in, the program is bringing you customers and you can afford to spend more on it. If they fall off a cliff right after the reward clears, you're not running a growth loop. You're running a promotion with extra steps.

A short checklist

• Pay for funding and real usage, not sign-ups.
• Measure net deposits, and cap rewards per account and per cluster.
• Hold rewards in escrow until funding settles and activity looks real.
• Make deposits and withdrawals boringly reliable before you spend on acquisition.
• Compare referred cohorts with organic cohorts on funding, repeat use and retained balances.
• Expect farmers, and treat that as a design input rather than a surprise.

Where this leaves me

I still like referral programs. When they work, they're one of the most honest growth channels you have, because a real person is vouching for you. But in a money product the reward has to point at trust, not at activity. Pay for the moment someone believes in the product enough to keep their money in it, and give that belief a little time to prove itself.

Career context: Who is Sunil Vallath? Past proof on the site: Lume case study. Related craft: what first funded trade really measures and building products people use. More on About, Now, sunilvallath.com and the writing hub.