90 quiz apps, 8.2 million downloads, ¥20 million+ in revenue — the real picture of mass production, where only 2 in 10 hit

Solo app developers who publish their revenue tend to build the story around the one hit. What makes this case unusual is that the developer himself states plainly, “roughly 2 in 10 apps succeed, the other 8 don’t really generate revenue,” and then names the specific misses along with their download counts. This is a record of an interview with Seiichi Awata (Saaay / Testii), who made nothing but personality-quiz apps for four years, releasing 90-plus of them. All figures date to the interview in December 2017; the developer hasn’t publicly disclosed how things have moved since.

The overall picture at year four

Spreading ¥20 million+ in cumulative revenue over 48 months gives a monthly average of about ¥417,000. The best single month, ¥1.6 million, is roughly 4x that average. Quiz apps tend to spike in the month they go viral and then decay, and this swing itself expresses the nature of the business. As he puts it: “Month to month it goes up and down, but the revenue base has been steadily building bit by bit.”

Cross-checking the numbers surfaces one point worth flagging. Applying “¥5–10 per download” to the cumulative 8.2 million downloads would yield ¥41 million–¥82 million, which doesn’t match the actual cumulative revenue of over ¥20 million. Dividing cumulative revenue by cumulative downloads gives about ¥2.4 per download. The ¥5–10 figure is, as he states, “recent” revenue per download. It doesn’t apply to the cumulative total, which includes an early low-revenue period. Put another way, per-download revenue on the same volume of releases has improved 2–4x over four years.

The gap between hits and misses

The top three titles together total ¥12–16 million, meaning three apps out of ninety account for 60–80% of the ¥20 million+ cumulative revenue. The failure threshold is equally clear: “anything that didn’t exceed 10,000 downloads within a year counts as a real failure.” Romantic Type Quiz’s 5,000 DL and Wild Type Quiz’s 2,800 DL, even converted at the recent ¥5–10 per-download rate, would translate to lifetime revenue of only a few tens of thousands of yen.

No single turning point

There is no single dramatic event marking a turning point in this case. What he cites is the slow start (“at first it was hopeless, and for about six months it was ‘this is bad’ territory”) and how he got out of it.

As I kept making apps, users started circulating between my apps, and gradually I started showing up in the new-release rankings too.

What mattered wasn’t the number of apps by itself, but the cross-promotion network that emerges once the number of apps crosses a threshold. Every new release gets promoted across all his existing apps with a “new app released” notice. His 90th release effectively becomes a simultaneous promotion to the install base built by the previous 89. That generates initial momentum outside the store, which lands him in the new-release ranking, and organic downloads take over from there. Rather than each app being an independent bet, the existing asset base guarantees the initial velocity of the next one.

Breaking down what’s actually working

Video ads placed during the wait. Monetization is entirely ad-based: rectangle ads on the results page, occasional interstitials, and, critically, a video interstitial that plays right before the diagnosis result is shown. That last one alone accounts for about 50% of ad revenue. The placement logic is clear: the moment a user is waiting for their result is when anticipation, and thus the cost of leaving, is highest. It’s a textbook case of the same ad format performing differently depending purely on placement.

How broad a theme is determines install volume. He attributes the split between hits and misses to how broad the topic is. “Mental age” appeals to everyone. “Perfectionism” only resonates with people who already identify as perfectionists. The size of the addressable population determines both search demand and share volume, which in turn determines whether an app sustains installs long-term. He traces the failures of “Easily Swayed” and “Looks-Focused” to the same cause: too narrow a theme.

Building a receiving surface outside the store. He built a separate website where users can create their own quizzes, playable across all the apps. Having a web version also gets it picked up as a quiz page in Google search, and makes it easier to share on Twitter. Adding one channel that isn’t locked inside the app secures traffic beyond store search.

Reframing results drives shares. A key design principle he cites: phrase even unflattering results positively. “Negative” becomes “cautious,” for instance. To get someone to post their result on social media, that result has to feel shareable to them personally. The wording design directly drives the volume of free exposure.

What the five misses reveal

The failures fall into two categories. Wild Type Quiz’s 2,800 DL was a demand-side misread: “quiz apps have way more female users, so there just wasn’t demand for a ‘wild’ theme to begin with,” he analyzes. The outcome was essentially decided at the topic-selection stage.

Romantic Type Quiz’s 5,000 DL was different in kind, not a theme problem but an execution problem. A review said “the art is annoying,” and it never recovered from there. Illustration is outsourced, making style mismatches the hardest part of the pipeline to control.

The cost of the volume strategy is that the labor behind that 8-in-10 failure rate is never recouped. He handles all the programming and quiz-text writing himself, and a miss costs the same effort as a hit. That’s exactly why he built his own tool to speed up writing the quiz text, raising “writing speed” is something he flags as an ongoing challenge. In a mass-production model, the win-loss outcome is decided not by the odds of a hit, but by the cost per title.

How far can this be copied?

The design philosophy is reproducible. If you can read your revenue-per-download figure, you can rough out projected revenue at the planning stage from “downloads × unit rate.” The “how broad is the theme” test also works as an upfront filter. And the division of labor (outsourcing the low-leverage part (illustration) while keeping planning, implementation, and writing in-house) is a form any solo developer can adopt.

What’s hard to reproduce is the 8.2-million-download existing user base itself. Cross-promotion only works once you already have a group of apps to promote from. This trick doesn’t exist for your first or second title, and you’d have to survive the “first six months” he describes. On top of that, the 90%-Android download split reflects an environment where casual apps could spread easily via new-release rankings on the Android market at the time. Given how much time has passed since the interview, whether the same revenue-per-download economics still hold cannot be confirmed from his public disclosures.

The ¥20 million+ cumulative figure is more accurately read not as a record of the talent to pick winners, but as a record of building a cost structure that could tolerate an 80% failure rate.

Related reading

• 24 Kindle books, ¥50,000–70,000/month in royalties: real data from 16 months and ¥750,000 cumulative of “mass-produced” self-publishing — a content business with the same structure: small revenue per unit, accumulated through volume.
• levelsio, running Photo AI solo — a solo developer who kept shipping many products before finally landing a big one. A useful comparison of at-bats versus hits.