¥2.5 Million a Month From a Chat-Analysis App: 85% of Traffic From TikTok, 37.5% CVR — an Indie Design "Reverse-Engineered From the Distribution Channel"

Indie app revenue reports are common, but records that disclose CVR trends, cost ratios, and even failed support operations with screenshots are rare. The account of the chat-analysis app “IsTalk,” published on note by Kei (36, an engineer with 15 years of experience), is a first-class document that traces the entire indie-development pipeline, concept → viral moment → improvement → price increase, in real data. Let’s confirm the numbers, then break down why this design worked.

The Published Numbers

Rearranged as a timeline, the growth is anything but a straight line.

What the App Is, and Why It Grew on TikTok

IsTalk reads LINE chat histories and “analyzes” conversation patterns (reply speed, imbalance in message volume, emoji usage) and displays the results. Its function in one sentence: “diagnose your chats with your crush.” It lands precisely on the desires of women in their teens and twenties who want to check the answer key on their love lives.

The essence of this app is that its analysis results are screens you immediately want to post to TikTok. The diagnosis becomes conversation material, and each post recruits the next user. The figure of 85% of traffic from TikTok and friend referrals shows this loop running with almost zero ad spend (TikTok ads amount to only about ¥20,000/month). It is the subscription-monetized version of the structure by which Peing exploded through “shareable output”.

The “Post-Date Emotional Vibe” That Moved CVR 13.3 Points

The substance of the 24.2% → 37.5% CVR improvement was almost entirely presentation changes, not feature additions. He redesigned the App Store screenshots in a style evoking the “emotional afterglow of a date,” refreshed the subscription pitch screen with designs aimed at young women, added theme-color selection (special colors are premium-only), and inserted tutorials at first launch and on the analysis-results screen. That CVR moved 13.3 points without touching the core features shows that once your understanding of the user base sharpens (90% female, 90% aged 10–29), unifying “who each screen is designed for” moves revenue more than development does.

The revenue structure is engineered too. Analysis runs entirely offline with no servers, keeping the cost ratio under 20% (about ¥40,000/month). He rewrote the app entirely from SwiftUI to Flutter, cutting analysis time from over an hour to under 10 seconds. ¥2.5 million in monthly revenue against ¥40,000 in costs — this margin is the strength of on-device processing, in sharp contrast to AI apps whose cloud costs scale with revenue.

The Record of Stumbles — Going Viral Is Not Pure Blessing

The value of this account also lies in its disclosure of failures. Early on, code quality was poor, bugs were frequent, and the iOS and Android implementations had drifted apart. He admits he dreaded the Flutter rewrite and got through it by adopting an AI coding tool (Cursor). The most painful episode was operations during the viral spike: he couldn’t keep up with support-email replies during the traffic flood, which fed negative reviews. As countermeasures he added an update-announcements page in the settings screen and a message shown at contact time explaining the developer’s situation, after which bug-report inquiries “dropped dramatically.” The traffic peak arrives at the exact moment your support capacity is weakest, this time bomb is structurally embedded in every indie developer’s viral moment.

He also offers a rule of thumb for when to start monetizing: around DAU 100 (MAU 3,000) is the line for introducing payments. Put another way, wiring in payments below that scale yields no valid test, a piece of hard-won operational knowledge.

Reading Between the Numbers

The pricing design bakes in the “low” 18.8% weekly retention. Diagnosis apps get stale fast, and most usage is in fact one-off. IsTalk’s revenue rests not on long-term retention but on converting users to monthly or annual plans during that high-enthusiasm first week (a 37.5% CVR is an extraordinary level). If a tool’s lifespan is short, build a payment curve that recovers value within that lifespan, low retention is not necessarily a defect.

The price increase (¥300 → ¥500) was executed at the user-count peak. In August 2024, amid nearly 200K new users a month, he raised prices while grandfathering existing users. Raising prices after demand is proven is the most reliable revenue lever in indie development, confirming the same lesson as Pallyy’s $15 → $18.

“15 years as an engineer” is not the point. Technically this is chat-history parsing plus statistics display, not hard. What made the difference was a concept reverse-engineered from the distribution channel (TikTok) and the operational discipline of CVR improvement, the indie version of Jenni AI’s “distribution over product”.

Conditions and Limits of Reproduction

What you can take away is less any single technique than the sequence: reverse-engineer the concept from the distribution channel (output that shines on TikTok), unify presentation around the user segment once the viral spike reveals it, and raise prices only after demand is proven. The under-20% cost ratio via offline processing is likewise a structure you can deliberately choose for diagnosis or utility apps. On the other hand, the Japan-specific raw material of LINE chats and the fit with the TikTok culture of women in their teens and twenties sharply limit this case’s reproducibility. Follow the same steps in another genre and, without “output people want to share,” the loop’s starting point simply doesn’t exist. The timeline also matters: of the three and a half years, the first two and a half were run-up.

Related Reading

• Peing — the precedent for “shareable output” design
• Jenni AI — growth design reverse-engineered from TikTok distribution