The first open is the easy one. A QR code, a shared card, an ad — you paid for it, it happened, the number went up. The second open is where almost every mini-program quietly dies, because nothing is paying for it anymore. There's no poster in front of the person this time. There's just a reason, or there isn't.
We used to treat that as a numbers problem: more reminders, more frequently, more channels. It isn't. We're not going to talk about retention metrics here, there's enough written about those. The question is narrower: what actually makes someone open it again, and what does that cost us. And here's the awkward bit — the thing you'd reach for to earn that visit, a notification, is also the thing that can end the relationship for good.
Bought the first open, now what
What's left after the first open? An empty state, a half-finished task, or a vague impression that this might be useful someday. The loop doesn't start with how do we remind them. It starts with did we plant any reason worth coming back to last time.
There are basically three reasons that hold up: something changed (price, stock, status), something is unfinished (a draft, a pending payment, a confirmation waiting), or a time arrived (a booking, a cycle). Everything else — we miss you, it's been a while — shouldn't be sent. Those aren't reasons, they're requests for attention.
Then there's the layer the host decides. Subscription messages usually need a one-off authorisation, one authorisation per message, templates that pass review, and category limits. So a notification isn't a channel you use whenever you feel like it. It's a scarce allowance, and it only pays off if you spend it on the moment something genuinely moved.
A trigger needs a reason, not a schedule
A trigger should be a state the person cares about changed, not the calendar says it's Tuesday. We settled on a filter that works surprisingly well: if this message landed on my own phone, would it annoy me? If yes, don't send it.
The first message after a cold start should be in no hurry at all. Pushing the moment someone leaves is basically telling them we're watching. And there are cases where you don't send, full stop: no new information, late at night, right after the user dismissed one, content unrelated to their last open, or just to fill the weekly send quota. Leave a gap, the dashboard looks slightly worse, but the permission is still there.
| Trigger | Why they'd come back | What the host must allow | Cost of overuse |
|---|---|---|---|
| The thing they watched changed | Because the price, stock or status actually moved | A subscription message tied to that object | They learn the alerts are noise and stop trusting them |
| An unfinished task is still waiting | To finish what they started, not to browse | A saved state plus one allowed reminder | A draft that nags every morning gets abandoned |
| Someone replied or joined | Because a real person is on the other side | A push with the right target screen | Social triggers burn out fastest when faked |
| A booked time is close | Because they asked to be reminded | A scheduled, one-off authorisation | Two reminders for one booking feels like spam |
| A fixed weekly nudge | No reason at all, just a schedule | Nothing; this one shouldn't exist | This is the message that gets permissions switched off |
The permission you can't get back
Now the costs. Notification budget is scarce: one authorisation, one message, gone. Templates, categories and frequency are set by the host, and when those rules shift we have to reshape the product behind them — that part isn't even in our hands. The harshest one is at the end: once permission is off, you don't get a second chance to ask. The user switches it off and that's it. You'll never get another opening to send anything, not even to say fine, I'll send fewer. It's worse than never having sent one at all.
So there's a real tradeoff here: send nothing and people genuinely forget you, or burn a non-renewable authorisation on an empty sentence. It isn't don't send, it's save the allowance for the moment something actually changed. Trust is consumable, easy to spend, impossible to refill.
One more awkward thing: host capabilities don't line up. The same trigger logic lands fine in one host and doesn't even clear template review in another. Don't design for the most permissive one — design for the strictest, then add a little extra only where you have room.
Before you queue the next message, write down in one plain sentence what changed since the last open. If you can't, don't send it. Save that authorisation for the moment the thing they cared about actually moves — and if that moment never comes, that's information too: the product didn't give anyone a reason, and no reminder will invent one.
Sources
- WeChat — mini program developer docs — subscription messages, one-off authorisation and template limits.
- Apple — Human Interface Guidelines — what a notification is for, and when not to send one.
- Android Developers — documentation — notification channels and the permissions users can revoke.
- Nielsen Norman Group — research — the trust cost of interrupting people.