Engineering referral loops for regulated money apps

# Engineering referral loops for regulated money apps

Most people who opened Cash App did so because a friend sent them a link with a small cash bonus attached, often cited around $5, payable once the new user sends a qualifying payment. Block is handing real money to strangers for opening a stored-value account. Every part of that offer (the amount, the qualifying action, the deadline, the clawback rule) has been tuned against fraud, unit economics and promotion rules at the same time.

This lesson is about that tuning: picking the trigger, modelling whether the loop compounds, and designing the payout when the reward itself is a regulated financial inducement rather than a free tote bag.

Why the mechanics differ in fintech

In a media app a referral gets you a user. In a money app it gets you a funded account: someone who has cleared identity checks, linked a bank or loaded a balance. The value gap between those two things is usually an order of magnitude, which is why fintech programmes can afford payouts a content app never could.

The price of that is three constraints that run through everything below.

• Paying the bonus is itself a money movement your fraud and AML systems have to clear, not a fulfilment chore for the growth team.
• The invite text is a promotion attributable to you even when a customer writes it. Give referrers pre-written copy rather than a blank message box, and hold it to the standard the compliant messaging lesson sets out.
• The bonus enters your blended acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → on the terms the CACCACCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → lesson describes, and it enters twice, because most programmes pay both sides.

One term to fix before the modelling: viral coefficient (k), the average number of new qualified users each existing user brings in. Above 1 the loop feeds itself. Below 1 it is an amplifier on whatever you buy elsewhere.

Dissecting Cash App and Revolut mechanics

Cash App: reward tied to a payment

Cash App's classic referral pays both the referrer and the new user, but only after the new user sends a qualifying payment (commonly funded by a linked debit card) inside a short window.

Three design choices are doing the work. The reward is two-sided, so the referrer has a reason to send the link and the invitee has a reason to finish. The payout is gated on a behaviour rather than a signup, which filters out dormant accounts. And the deadline is short, which compresses activation.

The trigger action is also the habit the company most wants to build. The incentive and the activation metric are the same event.

Revolut: reward tied to card activity

Revolut's referral offers have typically required the invitee to order a card and make a handful of card payments within roughly a month, with terms varying by market and campaign, and with a cap on how many rewarded referrals one customer can bank in a year.

That is a heavier trigger than sending $5 once. It selects for a customer who actually spends on the card, which is where interchange comes from, and the cap stops a small group of high-volume sharers from consuming the whole budget.

The rule underneath both: choose the trigger that matches your unit economics, not the one that converts most easily. Cash App monetises payment volume, so it triggers on a payment. Revolut monetises card usage and eventual subscription upgrades, so it triggers on card usage.

Modeling the viral coefficientviral coefficientThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition →

Before you fund a single bonus, model whether the loop can compound.

k = i × c

where:
  i = invites sent per existing user
  c = conversion rate of those invites (share who become qualified users)

If each user sends 4 invites and 10% convert to funded accounts, kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → = 4 × 0.10 = 0.4. That is not failure. Every 100 users generate 40 more, who generate 16, and so on: a 1.67x multiplier on paid acquisitionpaid acquisitionVisitors arriving via paid ads or sponsored placements, where you pay a platform to display your message rather than earning visits organically.View full definition → at steady state.

The common self-deception is computing c on signups. A programme can report kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → = 0.9 on signups and kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → = 0.25 on funded, qualified accounts. Only the second number is worth money, and only the second number is the one you are actually paying for.

The cycle time trap

Two loops with the same kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → grow at different speeds if the cycle time (how long one generation takes) differs. Cash App's loop can close in a day. A card-activity loop cannot, because a physical card has to be printed and posted before the qualifying spend can happen, which pushes the generation out to weeks.

Do the arithmetic over one quarter. A kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → of 0.5 on a three-day cycle has effectively reached its full 2x multiplier. A kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → of 0.8 on a 30-day cycle has banked roughly 3x of its eventual 5x, and you have carried the cash cost of pending bonuses on the balance sheet the whole time. The higher coefficient wins in the end; the faster one wins the quarter and ties up less cash.

🎬 [VIDEO: "The Viral CoefficientViral CoefficientThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → Explained" - youtube.com - a short walkthrough of how kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → and cycle time drive compounding growth]

Structuring funded-account triggers

Set the trigger too low and you pay for junk accounts. Set it too high and almost nobody qualifies, referrers stop seeing rewards land, and sharing collapses within two cycles.

From lightest to heaviest:

1. Signup only. Cheap, worst quality, and an open invitation to fraud. Avoid in fintech.

2. Account funded. The user moves money in. Better, still gameable by transferring in and straight back out.

3. First qualifying transaction. Cash App's model. Proves real usage.

4. Sustained behaviour. Repeated card spend, a recurring deposit, a balance held past day 30. Highest quality, slowest, and the hardest to explain in one line.

5. Split payout. A small amount at funding, the larger amount at day 30 retention. This buys you both a fast-feeling loop and a quality gate, at the cost of a more complex offer to describe.

Match the trigger to lifetime valuelifetime valueLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition →

Your combined two-sided payout should sit comfortably below the expected value of a user who reaches that trigger, discounted by the share who reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → it. Build the table before launch: trigger type, expected conversion, expected value per qualified user, maximum affordable payout. If the payout exceeds the value, scale makes the hole deeper.

One finance detail that surprises marketers: cash paid to your own customers is often netted against revenue rather than booked as marketing expense, so the loop can quietly depress reported revenue while your channel dashboard shows a falling acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition →. Agree the treatment with the CFO before you launch, not in the quarterly review.

Designing payouts when the reward is regulated

When the reward is a security

Robinhood's referral gives a randomly selected share of stock. The company has disclosed the probability distribution behind it, with the large majority of rewards worth a few dollars and a small chance of something much larger. That disclosure is not decoration: a stock reward is a promotional inducement to open a brokerage account, its value moves after you grant it, the recipient normally cannot withdraw the proceeds for a set period, and promotional stock is commonly reported to the recipient as miscellaneous income.

Compare that with what Robinhood did before it had a product at all: a waitlist where referrals moved you up the queue, which reportedly reached about a million people pre-launch. Queue position costs nothing, carries no tax reporting and creates no regulated inducement. If you are pre-launch, that mechanic is available to you and a cash bonus is not.

The mechanic can be banned outright in a market

A referral offer is not portable. Under the FCA's financial promotion rules, refer-a-friend bonuses are banned for high-risk investments, and that regime was extended to cryptoassets from 8 October 2023. A crypto referral bonus running happily in the US cannot simply be switched on for UK users. Check the mechanic per market before you build one loop and assume it ships everywhere.

Clawbacks, fraud and the support bill

Every programme needs published terms covering eligibility (age, geography, account status), the qualifying action, the caps, and the anti-fraud rules on self-referral and duplicate identities.

Fraud here is not only a budget problem. Paying bonuses into synthetic accounts creates AML and KYC exposure, so the identity check belongs inside the trigger: no verified customer, no payout, ever.

The failure mode nobody budgets for is the clawback storm. A referrer shares a link into a large group chat, dozens of signups arrive from similar devices, the fraud engine reverses the bonuses, and you get hundreds of support tickets and public complaints from people who did nothing wrong. Reversing a reward in a way the customer could not have anticipated is exactly the kind of conduct the CFPB's UDAAP guidance treats as unfair, and the Consumer Financial Protection Bureau reads complaint volume. Write the caps into the offer up front so the ceiling is visible before anyone shares.

Putting it together: a launch checklist

1. Pick the trigger that matches your economics: payment volume, card spend, deposits held.

2. Estimate k and cycle time on qualified accounts only, and assume most invites never convert.

3. Size the payout below the value of a user who reaches the trigger, and agree with finance how it hits the P&L.

4. Check the mechanic per market before building, because some rewards are prohibited outright.

5. Put identity verification and caps inside the trigger so you never pay on a synthetic account and never claw back a surprise.

6. Give referrers the words. Pre-written share copy protects you and converts better than whatever they improvise.

The best fintech referral loops are boring to explain and hard to abuse. Clarity is both a growth tactic (people share what they understand) and a defence.

Key Takeaways

• Trigger on funded behaviour, not signups. Cash App pays on a first payment, Revolut on card activity. Each rewards the action that pays for the bonus.
• Model k and cycle time on qualified accounts. A programme can show kkThe average number of new users each existing user generates through referrals. Above 1.0, growth compounds on itself and becomes exponential.View full definition → = 0.9 on signups and 0.25 on funded users; only the second is real, and a fast loop banks its multiplier inside the quarter.
• The reward is a regulated product. Stock rewards carry probability disclosure, holding periods and tax reporting; UK rules ban refer-a-friend bonuses for crypto and other high-risk investments.
• Caps and identity checks belong inside the trigger, because a clawback wave costs more in complaints and support time than the fraud it recovers.
• Know how finance books it. Bonuses paid to customers are often netted against revenue, so a loop that looks cheap on the channel dashboard can still shrink the top line.