Back to notes

June 11, 2026 · 5 min read

We Overhauled a Push Notification Strategy, Opt-Outs Dropped 40%

The push strategy wasn't badly built. It was built around what the platform made easy to send, not around what users actually wanted to receive.

Case StudyMobile

The client's push notification opt-out rate had been climbing for a year, and the team's response had been to A/B test copy and emoji usage, which moved the needle by basically nothing. The actual problem wasn't what the notifications said. It was how many of them there were and how little they had to do with what any given user actually cared about.

What the old strategy looked like

Every user received the same broadcast schedule — a daily engagement nudge, a weekly feature highlight, promotional pushes whenever marketing had something to announce — regardless of what that user had actually done in the app. A user who'd never touched the feature being highlighted got the highlight push anyway. Someone who'd just made a purchase got a 'come back and try this' nudge the next morning.

A notification that's irrelevant to the person receiving it doesn't just get ignored. It teaches them that your notifications in general are worth ignoring, including the ones that would have mattered.

What we changed

  • Segmented the broadcast schedule into behavior-triggered sends — the feature highlight only goes to users who haven't used that feature, not everyone.
  • Capped total push frequency per user per week regardless of how many campaigns were technically eligible to send to them, with a priority system to pick the single most relevant one.
  • Added a lightweight preference center letting users opt into categories rather than all-or-nothing, which itself reduced full opt-outs because partial control satisfied people who'd have otherwise opted out entirely.
  • Suppressed promotional pushes for 48 hours after any purchase, which was a one-line rule that fixed one of the most obviously tone-deaf sends in the old system.

Opt-outs dropped 40% within two months, and engagement with the notifications that did go out actually went up, because the volume reduction made each individual send more noticeable rather than one more item in a stream of noise. Nobody tested a single new word of copy to get there.