How We Fixed a One-and-Done Problem on a Daily Puzzle Game

WotsThat.com is a daily close-up picture quiz: a zoomed-in mystery photo every day, guess what it is. Simple concept, the kind of thing that should have obvious built-in habit loops the way Wordle does. It didn't, not at first. This is what the data showed, what we built to fix it, and what's still an open question.

The data said something uncomfortable

A look through Supabase at a month of signups turned up a pattern that's probably familiar to anyone running a habit-based product: most people who signed up in July played exactly once and never came back. A smaller group signed up and never played at all.

Those are two different problems, and they needed two different fixes. Bundling them into one generic "win them back" email would have wasted the send on people who needed a completely different nudge.

Activation vs. retention are not the same email

For people who signed up but never played, the ask is small: come try it once. That became a single automated email, sent once, 48 hours after signup, only to accounts with zero plays.

For people who played once and drifted, the framing is different, they already know what the game is, so the email needed to remind them what they're missing rather than explain the product from scratch. That's a separate win-back send, triggered when someone's last guess was 3 to 30 days ago, sent once per inactive stretch so it doesn't nag.

The bigger lever wasn't email at all

Email gets attention, but it's also easy to ignore, and it requires an address in the first place. The thing that actually seemed likely to change behaviour day to day was a visible streak counter, a small flame badge on the play page and profile showing consecutive days played, hidden entirely at zero so it never shames anyone for not having one yet.

Once the streak existed, it created its own problem: people were losing streaks without realising it until it was too late. So we added a third automated email, streak-at-risk, that only fires for accounts with a 3+ day streak who haven't played yet today, on a separate evening cron job. It's the one email in the system explicitly trying to protect something the player already built, rather than trying to re-recruit them.

Then we questioned the core loop itself

All of that assumed the existing "browse and guess whatever" game was the right on-ramp. It probably wasn't, for a new visitor, an open-ended pool of images with no shared reference point is a much weaker hook than "here's today's puzzle, everyone's playing the same one." So we built a proper Daily mode: one official puzzle a day, same for every player, resets at UTC midnight, playable anonymously with no account required. A signed-up-only Daily Streak sits alongside the older streak rather than replacing it.

That single change surfaced a smaller but very real bug: a new visitor clicking a shared Daily link couldn't figure out how to actually submit a guess, because the guess box stayed disabled until you tapped a separate "reveal" button first, with no explanation. Nobody had noticed because everyone testing it already knew the flow. Typing into the box now auto-reveals the image and starts the timer. Small fix, but it was quietly blocking every first-time player who arrived from a share link, which is exactly the traffic you most want to convert.

Leaderboards had to catch up too

Once Daily became the mode we actually wanted people entering through, the leaderboard pages were still defaulting to the all-time Overall view, requiring an extra click to see today's standings at all. We flipped the default to Daily across the site, and added a compact "today's top 5 plus your own rank" widget directly on the Daily results screen, so players see where they stand the moment they finish, without navigating anywhere.

What's still open

None of this is a finished story. Streaks and daily-default views are well-understood mechanics precisely because they work elsewhere, the honest next step is watching whether the specific combination holds up over a longer window, not just assuming it will because Wordle proved the genre works. Diagnosing the one-and-done pattern in the first place was the actual unlock, most retention advice assumes you already know why people are leaving, and usually you don't until you've actually pulled the data apart.