DEV Community

Youngjun Yoon
Youngjun Yoon

Posted on Originally published at cpcai-studio.netlify.app

The 30 MB budget: why app size decides installs in half the world

I shipped a 47 MB app update once and watched installs drop in three countries the same week. Nothing in the code was broken. The app was simply too big for the phones and the data plans that most of its users actually had.

Since then every build I do starts with a size budget, written down before the first screen: 30 MB download, 60 MB installed, first meaningful screen in under three seconds on a mid-range Android over a slow 4G connection. It is the least glamorous constraint in mobile development and it changes more decisions than most architecture documents.

I build apps alone from Seoul at CPC AI Inc., and I have shipped two of my own products (a mobile strategy game and a crypto market dashboard), plus client apps for markets from Karachi to Ulaanbaatar. The size budget is the one thing that transfers cleanly between all of them.

What actually eats the budget

In an app built with an AI code generator — Lovable, Bolt, Cursor, v0 — the first build is usually 5–15 MB of code and 40–60 MB of assets nobody looked at.

  • Uncompressed images. A hero image exported at 2400×1600 PNG is 2–3 MB. The same image at 1080 wide WebP is 90 KB. Ten of those is the difference between a 12 MB app and a 40 MB app.
  • Two icon sets and three fonts. Design tools hand you packs; every weight of every family ships. Two families, three weights total, subset to the characters you actually use.
  • Video. A 10-second background loop is 6 MB minimum. Almost no business app needs it; the ones that do should stream it, not bundle it.
  • Duplicated dependencies. The AI adds a date library, then a second date library for one formatting function. Ten of these and the bundle doubles.
  • Debug symbols and source maps shipped in the release build. Free wins, routinely forgotten.

What a 30 MB budget forces you to decide

A size budget is really a product decision in disguise. Once you accept it:

  • The offline experience becomes a queue, not a cache of everything. You sync what the user is about to need, not the whole catalogue.
  • Images get a server-side resize pipeline on day one, because "we will optimize later" never happens after launch.
  • The admin panel lives on the web, not inside the app. Half the size complaints disappear when staff features are not shipped to customers.
  • Onboarding cannot be a five-screen carousel of 1 MB illustrations. It becomes one screen and a real first task.

None of that makes the app worse. It usually makes it faster to understand, which is why I now consider the budget a design constraint rather than an optimization task.

The test that matters more than a Lighthouse score

Lighthouse on a laptop lies to you. The test I run before handover:

  1. A mid-range Android phone (the kind that costs a week's average local wage), not my own.
  2. Data throttled to "slow 4G", and then airplane mode in the middle of a form.
  3. A fresh install, no cache, no logged-in Google account.
  4. The core flow timed with a stopwatch: choose → pay → confirm.

If the form loses its content when the connection drops, the app fails — regardless of what the performance panel says. In markets where mobile data is metered and patchy, a lost form is a lost customer, and the fix is a queue plus a local draft, both cheap if designed on day one.

Where this fits in a 3–5 day build

Day 1 — one flow, its screens, the size budget, and the four payment/document answers. Day 2 — the flow built, payment in test mode, a link on your phone. Day 3 — testing on another phone, another account, another card, with throttled data. Day 4 — admin panel, notifications, store-required pages. Day 5 — publish, hand over the accounts (Apple, Google Play, domain, database) in your name, plus the notes for shipping the next version.

A project that takes an agency three months takes three to five days here, at up to a twentieth of the cost, because one person does the whole thing with no sales layer in between. Edits stay free until you are satisfied.

If your AI-built app is sitting at 80% — it runs on a laptop but not on a real phone — send the link. Site: https://cpcai-studio.netlify.app, email: yyhak12345@gmail.com.

Top comments (0)