Build beside graduation, but start with a narrow pain
The case describes a founder building an AI voice sales role-play product while completing school and without venture funding. A learner speaks with an AI prospect, then receives a score and feedback. That is a concrete workflow rather than a promise to replace an entire sales team.
The positioning matters because established vendors may sell training seats to large companies while independent sellers and small teams face a much higher per-seat price. A small product can begin with the part of the problem that is expensive, repeatable, and possible to simulate.
The €15K figure is a self-report, not a forecast
A team member wrote on Reddit that the product reached roughly 260 paying users and €15,000 in monthly recurring revenue after seven months. This is grade C evidence: the identity, user count, and MRR come from a team-associated account and have not been independently audited.
The number should therefore stay attached to its time window and evidence label. It is not profit, a typical result for graduates, or proof that an AI wrapper will find customers. The case offers a product-selection clue, not a revenue guarantee.
Serve the customers large products ignore
The reported price gap creates a testable wedge: independent salespeople and small teams may want practice but cannot justify enterprise training software priced at roughly €70–€250 per seat. A builder can ask whether ten such users will repeat a role-play, share a real objection, or pay for a small package.
This is a better first question than asking which model is most advanced. If nobody returns for another scenario or agrees that the result is worth money, adding a stronger voice model will not repair the distribution or pricing problem.
Turn industry experience into a simulation
A graduate does not need a large audience to begin. Internship experience, a sales job, customer-support work, or a familiar professional exam can reveal a costly interaction that people repeat and can safely simulate. Write one scenario, one expected behavior, and one score before building a broad platform.
Then recruit ten people who are normally too small for an enterprise vendor. Watch whether they complete the exercise, ask for another round, and accept the feedback. A return session or a small payment is stronger evidence than a compliment about the demo.
Test voice, scoring, and summaries as separate costs
Voice input, conversation generation, scoring, and a coaching summary can have different quality and cost profiles. Run a small redacted sample through a few candidates instead of funding the whole stack at once. Record usable output, latency, usage, failure type, retries, and the time a human spends correcting the feedback.
Use an isolated, revocable API key for the experiment and define a stopping budget. If the same error repeats, a review takes longer than the training benefit, or no target user returns, stop and revise the scenario before expanding models or features.
Keep the story inside its evidence boundary
The source does not show audited revenue, profit, retention, or a complete acquisition-cost picture. It also does not show that the founder or product used APIToken. The non-replicable conditions include existing sales experience, the team's product knowledge, and the fact that the observation was shared by someone close to the project.
Within the models currently and lawfully offered at https://APIToken.Company, a builder can check public status, create an isolated key, and compare a small real voice, scoring, or summary task while preserving usage records and a budget limit. Today’s smallest action is to list one training scenario and ask ten overlooked customers whether they would repeat it or pay.