Idea Autopsy: Tutorbee Had 180,000 Sign-Ups and Almost No Students

• Name: Tutorbee (fictional)
• What it did: An AI homework tutor for 13-to-17-year-olds, sold to parents as a monthly subscription
• Lifespan: 19 months
• Peak: 180,000 sign-ups in its first school term
• Cause of death: Retention collapse, made worse by a product that taught the wrong thing

The pitch

Tutorbee's founders, a former maths teacher and a mobile developer, had a pitch that wrote itself. Private tutoring is expensive, and most families can't afford it. A patient, always-available AI tutor could explain quadratic equations at 11 p.m. on a Sunday for less than the price of one hour with a human.

The early demos were genuinely charming. A student photographed a problem. Tutorbee asked what they'd tried, gave a hint, then another, and revealed the answer only if asked three times. Parents loved it, investors loved the parents, and the company raised a seed round on the strength of a waiting list.

The numbers that lied

September went off like a firework. Back-to-school timing, a few viral clips of students "finally getting it" and a generous free trial pushed sign-ups to 180,000 by December. About 9% of sign-ups became paying customers, and the team celebrated.

Then the curve did something nobody had modelled. Usage clustered in two spikes, the week before mid-term exams and the week before finals, and almost disappeared in between. By March, fewer than one in five paying families had a child who'd opened the app in the past fortnight. Parents, who paid, noticed the charge on their statements. Students, who didn't pay, had stopped caring.

Cancellations arrived in waves after every billing cycle.

Cause of death: three wounds

Wound one: the answer machine. The hint-first design was lovely in demos and easy to defeat in practice. Students learned that asking "just tell me" three times worked every time, and internal data later showed that most sessions ended with the answer revealed. Tutorbee had built a beautifully patient tutor and accidentally shipped a faster answer key.

Wound two: the wrong customer. Parents bought it, but teenagers used it. Product decisions were aimed at reassuring parents with progress emails, weekly reports and a dashboard, while the actual app had to compete with everything else on a teenager's phone. Nobody designed for the moment a 14-year-old decides whether to open the app or text a friend for the answer.

Wound three: the calendar. School is seasonal. A monthly subscription for a product used six weeks a year will always feel like a bad deal by February. Meanwhile, every session cost real money in model usage, and the heaviest use, during the cramming spikes, came exactly when margins were thinnest.

There was a smaller fourth wound. Several schools blocked the app on their networks after teachers noticed homework answers that were suspiciously alike. That wasn't fatal on its own, but it closed a door the company hadn't realised it needed.

Tutorbee didn't fail because the AI was bad. It failed because nobody wrote down what "learning" meant, so the product optimised for what was easy to count.

What the next team should do differently

The underlying need hasn't gone away. Plenty of families still want affordable help with homework. If you're building in this space, here's what the autopsy suggests:

1. Define learning before you define features. Pick one measurable thing, such as a student solving a similar problem unaided a week later, and make it the north star instead of sessions or minutes.
2. Make the answer cost something. Not money, but effort. Require the student to attempt the next step before revealing more. The friction is the product.
3. Design for the teenager at 9 p.m. Interview students, not just parents, and find out what makes them open the app when nobody's watching.
4. Price for the calendar. Term passes, exam bundles or school-licensed plans fit seasonal use far better than a monthly charge.
5. Partner with teachers early. A tutor that teachers recommend gets past the school firewall. A tutor that teachers suspect gets blocked.

The autopsy table

The saddest detail came in the company's last month. A small group of students who had used Tutorbee steadily all year, outside the exam spikes, showed real improvement in their grades, according to the company's own survey. The product worked for students who used it as intended. There just weren't enough of them, and nobody had built the business around finding more.