Most companies use AI to try to replace humans with mediocre chatbots. Preply did something different: they use it to measure the human. The result is brilliant, but their economic incentive design is, from an ethical standpoint, a disaster.
The Win: Observability with AI Review #
I’ve been testing the AI Review system (in beta), and as a data engineering use case, it’s one of the best out there right now. It doesn’t try to replace the teacher-student connection; it acts as an observability layer that extracts value from a synchronous session:
- Interaction Metrics: Bar charts exposing the balance of talk time. Who dominated the class?
- Level Analysis (NER): Automatic vocabulary classification using Named Entity Recognition (B1, B2, C1).
- Precision Feedback: Grammar and pronunciation reports with the exact snippet of the original audio to compare the error.
It’s a tool that empowers the expert: the teacher focuses on flow and pedagogy, while the AI manages the technical logging that a human would miss in real-time.
The Systemic Failure: Extractive Incentives #
Where the platform fails is in the distribution of the value generated. It’s the “Uber” model applied to education: the platform provides the infrastructure, but the expert absorbs all the risk and operating costs.
- Acquisition at the expert’s expense: In the trial class, the student pays with a 50% discount, but the teacher receives 0 USD. Preply retains 100% of the margin while the professional gives away their time to close a sale that primarily benefits the platform.
- Commissions based on volume, not loyalty: The system forces you to put in hundreds of hours to drop from a 33% commission to 18%. It’s a model that rewards raw quantity, ignoring the teacher’s ability to retain and build loyalty with students over the long term.
Redesign Proposal #
If we analyze the data through the lens of technical efficiency and mutual aid, a healthier ecosystem model would look like this:
| Metric | Current Model (Extractive) | Proposed Redesign |
|---|---|---|
| Trial Class | 100% for Preply | 50% Platform / 50% Teacher |
| Base Commission | 33% | 25% |
| Discount Lever | Total Hours (Volume) | Months of Recurrence (Quality) |
| Floor Commission | 18% | 10% (After 6 months with the student) |
This change would transform the teacher from a “rented resource” into an ecosystem partner. If the teacher manages to retain the student, the platform should charge less, as the maintenance cost of that recurring revenue is minimal compared to acquiring a new one.
Conclusion #
Preply is an example of how to implement AI to improve a product, but it’s also a warning of how poor incentive design can suffocate the talent that sustains the business. In the AI era, the value remains with the human; technology should only be the means to liberate them, not to milk them.
Tech Stack #
To avoid depending on someone else’s rules, these are the tools I use to manage my projects outside the corporate ecosystem:
- Infrastructure: Infomaniak (Swiss-based cloud and email, no tracking).
- Learning: DataCamp (To understand what happens behind the interface and learn to build data solutions).
Content Support: This blog does not have ads and never will. If you find this content useful and want to support its existence, you can do so via Liberapay.
