Shipping Mobile Apps the Fivi Tech Way

Mobile launches go sideways for predictable reasons. Scope creep in week three. App Store rejections nobody planned for. A backend team that finds out about push notifications the day before submission. We've shipped 90+ apps and the calm ones all follow the same four-stage rhythm.

Stage 1, Discovery (week 1 to 2)

One workshop, one technical spike, one written scope. Discovery is not "endless workshops until the client is comfortable." It is a forced function that ends with a signed list of what is in version one and what is explicitly out.

We always run a one-day technical spike during this stage, usually around the gnarliest integration: payments, biometrics, a legacy backend. Spikes catch about 80% of the risk you'd otherwise discover in week six, when it's expensive.

If discovery does not feel slightly uncomfortable, you are deferring the wrong conversations to later.

Stage 2, Prototype (week 3 to 5)

By the end of week five, the client is tapping through a real device build with mock data and three real screens of polish. The rest is gray boxes. The point is to verify the architecture, not the pixels.

We use Expo EAS for over-the-air updates, so internal stakeholders are testing the latest build automatically. No more "which TestFlight build are you on?" threads in client Slack.

Stage 3, Polish (week 6 to 9)

Polish is where most teams quietly lose two weeks. We schedule it on purpose. Polish covers empty states, error states, offline behaviour, loading skeletons, accessibility passes, and the 47 micro-interactions nobody briefed but everyone notices.

pre_release:
  - lighthouse_mobile_pwa: ">= 95"
  - bundle_size_kb_max: 4500
  - cold_start_p95_ms_max: 1800
  - crash_free_sessions_min: "99.5%"
  - a11y_audit: passing
  - store_metadata_review: signed_off

Stage 4, Scale (post-launch)

Day one is not the finish line. The first 30 days are when feature flags earn their cost. We launch with everything risky behind a flag, monitor crash-free sessions hourly, and roll out region by region. If a number wobbles, we pull a flag instead of pulling a release.

The real reason calm launches feel boring is that every stage ends with a written artifact: a signed scope, a measured benchmark, a checklist that someone signed off on. Each of those artifacts saves you a week somewhere later. That's the whole trick.