[ MSP.DD ] For investors · Updated
How do you check whether an AI startup's data is really proprietary?
Short answer
Trace each dataset to its source and its contract. Data is proprietary when the company has the right to use it for training and product improvement, competitors can't buy or collect the same thing, and the product generates more of it with every customer.
Why it matters
Data is the most common moat claim in AI companies and the easiest to overstate. Data the company has no clear right to use is a liability, not an asset.
How to check
- 01List every dataset used for training, fine-tuning or retrieval, with its source.
- 02Check customer contracts and terms of service for the right to use customer data to improve the product.
- 03Check whether any data was scraped, or licensed with restrictions.
- 04Ask whether a competitor could buy or collect an equivalent dataset, and how long it would take.
- 05Check personal data against GDPR: lawful basis, retention and transfers outside the EU.
Red flags
- Training data scraped from sites whose terms forbid it.
- Customer contracts that don't grant the right to improve the product with customer data.
- Personal data with no documented lawful basis.
Good signs
- Clear contractual rights to use customer data.
- Data that accumulates with usage and isn't for sale elsewhere.
The numbers
- CRV: "Usage should generate data competitors can't replicate", and the data should improve the product for each new customer. [1]