The referred install that converted wasn't won by a banner. It came from a group chat, sent by someone the new user already trusted, with a line that read like a recommendation and not an ad. We've watched referral beat paid acquisition in mini-programs by a margin that surprised even us, and the reason has less to do with the reward than with where the message lands.

Why a share inside a super-app lands differently

On the open web, a referral link competes with everything else in a feed. Inside a super-app it doesn't. The share goes straight into a chat thread the recipient already opens every day, from a name they know, in a context that already has trust built in. The sender's identity does the work that a creative team usually tries to buy.

That changes the math. A referred user arrives pre-warmed: they've already heard the pitch from someone they listen to, and they've already cleared the first hesitation. Conversion from that install tends to be higher, and retention tends to hold longer, because the relationship that brought them in doesn't end at signup. It also helps that super-apps are the default front door in many markets, so the people you reach there are already in the habit of acting inside the app rather than bouncing to a browser.

The mechanics that actually move people

Three things matter more than the size of the prize. First, share-to-chat — one tap that drops a pre-filled, personal message into the right thread, not a generic copy-this-link. Second, frictionless invite: the person inviting should never have to explain how to claim it. Third, a reward that feels earned rather than given.

We've seen teams agonize over the reward amount and ignore the claim flow. A reward nobody can figure out how to collect is just a line item. The win is usually in removing steps, not raising the number.

But the reward is where teams get greedy. Here's what each shape actually buys.

Reward shapeWhat it buysWhat it quietly costs
Cash per installFast volume, dead-simple to explainAttracts arbitrage accounts with no intent
Credits tied to first actionBetter intent, keeps people inside the productMore friction, a lower raw count
Social status or leaderboardCheap and sticky for certain crowdsNarrow appeal, can feel manipulative
Dual-sided rewardSymmetrical reason for both sides to actDouble the payout, needs real margin
Tiered by referral depthRewards your best referrers mostConfusing rules, easy to game the tier

Don't buy empty installs

The trap is rewarding the install instead of the activation. An install is cheap to fake and easy to forget; the new person opens the app once, grabs the credit, and leaves. We'd rather pay out when the referred user does something real — completes a first purchase, invites someone back, returns on day seven. In one program we watched, shifting the reward from install to first purchase roughly halved the share of referred accounts that never came back.

Tie the reward to a behavior that matters to your business, and you filter out the people who were only ever there for the prize. You'll see lower headline numbers and better unit economics, which is the trade most teams refuse to make until the empty installs show up in their retention chart.

Fraud shows up exactly where the money is

We wrote earlier about fraud in mini-program payments, and the same lesson applies here with even less effort: any loop that prints value is a loop someone will automate. Reward mechanics are a magnet for fake accounts, device farms, and the oldest trick in the book — referring yourself through a second profile.

The patterns mirror payment fraud more than marketing. You're looking for velocity spikes from one device, clusters of accounts that share signals, and conversions that convert but never engage. The fix isn't to kill the loop; it's to bind rewards to identity and behavior the way you'd bind a payout — device binding, sane velocity limits, and a human review queue for the edges that look too clean.

One referral layer across many hosts

Here's the part that's easy to underestimate. A mini-program rarely lives in one place. The same product ships into a messaging app, a payments app, a social feed — each with its own sharing API, its own identity model, its own deep-link format. Build referral natively for each host and you're rewriting the same logic three times, and the fraud rules drift apart.

The alternative — forked logic per host — doesn't just cost engineering time. It means the fraud model you tuned in one app is blind in another, so abuse migrates to the weakest port. A runtime that targets several hosts can normalize those differences into one referral layer: one share card, one invite flow, one set of reward and fraud rules that travels with the program. Cross Mini App does this by exposing a single sharing and identity surface across hosts, so the loop you design once keeps behaving the same whether it lands in a chat or a wallet. That's less code and, more importantly, one place to watch for the abuse.

A closing thought

Referral isn't a growth hack you bolt on after launch. It's a feature of trust, and trust is the one thing a super-app already hands you. Design the reward for the user you want to keep, watch the loop like you'd watch a payment, and let the runtime carry the mechanics across every host instead of rebuilding them each time.

Sources