Our Flutter app had 20 first opens in its first two weeks of logging. One person opened the shop. Nobody bought.
That is too little traffic to A/B test anything, so we fixed what we could see by just using the app.
Here are the numbers from our own event log (Android only, Nov 2-14):
| Step | Devices |
|---|---|
| First open | 20 |
| Finished onboarding | 4 |
| Did the core action | 4 |
| Opened the shop | 1 |
| Started a purchase | 0 |
With numbers this small, a split test would take months to say anything. So instead of testing, we walked the path from "result screen" to "buy button" on a 360dp phone and wrote down every place where a person could stop. We found nine. These are the ones I think apply to most small apps.
1. The buy button didn't look like a buy button
Each paid report in the shop had a row with the price, a "preview" link and, when the store was slow, a "try again" chip. Three small things of equal weight. We changed it to one gold button that carries the price, and made "preview" a quiet text link under it. The retry now lives inside that one button instead of a separate chip.
2. The title broke in the middle of a word
On a 360dp screen, the report title wrapped and split a word across two lines. In Korean this looks especially broken. We gave the title the full row width and moved price and badge to the line below. For Korean we wrap only at word boundaries; Japanese has its own rule for the line tail.
3. The preview hid the button
The report preview scrolled, and the buy button scrolled away with it. Now the button sits in a bar pinned to the bottom of the preview. It is hidden when the report is already owned, when sales have ended, or when the shop is switched off.
4. Every locked chapter said the same thing
The yearly report has one chapter per month. Locked months showed the first line of that month's text. But that first line came from a score band, so most months showed nearly the same sentence. The preview looked like filler. We added a separate teaser field to each chapter and filled it from that month's free reading, which changes month to month. The paid text did not change.
class ReportChapter {
final String title;
final List<String> lines;
/// One line shown on the locked preview. Falls back to lines.first.
final String? teaser;
const ReportChapter(this.title, this.lines, {this.teaser});
}
5. The privacy question came too late
The compatibility check asks for a friend's birthday. People hesitate there. We put the answer on the input screen itself: the friend's birthday is only kept on the device if you save them as your pair. That line matches what the save code actually does.
6. We could not tell "tapped buy" from "store broken"
Our log only knew a purchase started when the store sheet opened. If the store was not ready, the tap was invisible. We added two events at the tap: shop_buy_tap (product, store ready or not) and store_not_ready (product, reason, status). The purchase flow itself is unchanged. Next time the funnel shows "opened shop, no purchase", we will know which side it was.
The other three were smaller: a direct link to today's zodiac reading from the result screen, a one-time prompt to add the home-screen widget, and a "show off my spirit" share button.
What I'd copy
- Walk your own paywall on the smallest screen you support, and count the taps between "I'm curious" and the price.
- One button per decision. Put the price on it.
- Log the tap, not only the store callback.
- If a preview repeats itself, it is selling against you.
We added 21 widget tests for these paths, and fixing three older tests showed that one of them had only passed by luck.
The app is Starlit Alley: Tarot & Saju on Android (free, optional paid reports): Google Play
Written with AI assistance from our own commit and event log; the numbers and code are from the app.

Top comments (0)