A grandmother handed her phone to her grandson so he could buy a skin. The mini-program asked for an ID number. He typed his own. Nothing in that session was fraud, and nothing in it was accurate either: the account belonged to one person, the face in front of the screen to another, and the document to a third. We had technically verified an identity and learned nothing about who we were dealing with.

Lightweight is the whole point of a mini-program, so identity is the place where that promise breaks first. You can't bolt a bank's onboarding flow onto something people open for ninety seconds. But you also can't pretend a host login is a legal identity. The work is deciding, per action, how much proof is actually needed — and then refusing to collect one field more.

How much proof, and from whom

Most teams jump straight to "do we need eKYC or not". That's the wrong first question. Start from the action: what are you actually on the hook for? Age gating, a wallet top-up, a cross-border payout and a shared family tablet are four completely different problems wearing the same label.

SituationHow far you actually need to proveWhat the host already gives youWhat you still have to do
Age-gated contentOver a threshold, not a nameAn age bracket or a verified profile flagRecord the decision, not the birthday
Wallet top-upA regulated identity, market by marketReal-name status, a bound cardRun the checks your own licence requires
Shared family deviceThat this isn't the previous userAccount switching, session isolationRe-auth sensitive actions, clear local state
Cross-border payoutLegal name and jurisdictionNothing reliable — host ID isn't a legal IDCollect, store and justify retaining eKYC data
One-off high-value orderA step-up, once, then leave me aloneHost biometric or re-auth promptKeep a short-lived proof, delete the document

The last column is where the cost lives. Notice that in the easy rows you mostly write down a decision, not a dossier. "This session cleared an age check at 14:02" is a line in a log. A scanned passport is a liability that follows you for years, through every breach disclosure and every deletion request.

The honest tradeoff

Here's the uncomfortable bit: inheriting the host's identity saves you a form and buys you almost no legal cover. A super app account proves someone controls a phone number or a login. It does not prove a legal name, a tax residence, or that the person is allowed to move money in your jurisdiction. When a regulator asks who your customer was, "the host said they were verified" isn't an answer we'd want to give.

Shared devices make it worse in a way no KYC vendor's demo covers. In plenty of markets one phone circulates through a household, a shop counter, or a shift of delivery riders. Session bleed between two people on the same device is a far more common failure than a forged document, and it's invisible if your model assumes one account equals one human. We've had to add re-auth on the three or four actions that actually matter, and accept the small friction that comes with it.

Then there's the cross-border collision. One market wants real-name verification for anything touching stored value; another tells you that collecting a national ID for a low-risk purchase fails data minimisation. Both are reasonable. Running one global onboarding flow means either over-collecting somewhere or under-complying somewhere else, and you don't get to resolve that with a clever UI. We'd rather branch early by jurisdiction and keep the branches explicit, even though it makes the codebase uglier.

And the cliché worth pushing back on: more identity data does not mean better risk. Every field you capture is a field you must encrypt, expire, audit, and hand over correctly when someone asks for deletion. eKYC isn't a checkbox you tick once; it's a decision to operate a small, sensitive database. For a mini-game selling cosmetic items, that's often a bad trade, and saying so out loud will save you a painful audit later.

Prove the action, not the person

Pick the smallest claim that satisfies the obligation, keep the proof of that claim rather than the raw document, and let the host carry only what it genuinely owns. Put re-auth on the handful of actions where a shared device would hurt, and keep everything else frictionless. Store a timestamped decision with a retention date attached, so the data disappears on a schedule instead of when someone remembers.

Then go look at your own signup flow on a borrowed phone, with a sibling logged into the same super app. If you can't tell which human just clicked "confirm", you don't have an identity problem to solve with another form. You have a session problem, and no amount of document capture will fix it.

Sources