[ MSP.DD ] For investors · Updated
How do you tell if a startup is just an AI wrapper?
Short answer
Ask what would be left if the company swapped its model provider tomorrow. A wrapper is mostly prompts and an interface on top of someone else's model, so a competent team could rebuild it on an off-the-shelf model within a quarter. A defensible company owns something the model doesn't: data competitors can't get, a position inside a customer workflow, or integrations that took years to build.
Why it matters
Wrappers can grow fast and still be worth little, because the next model release, or the provider's own feature, can take the product away. The price paid should reflect what the company owns, not what it rents.
How to check
- 01Map the architecture: which parts call external models, and which parts are the company's own code, data and integrations.
- 02Run the rebuild test: how long would a competent team need to reproduce the product on an off-the-shelf model? A quarter or less means wrapper.
- 03Read gross margin by customer cohort. Model costs that grow with usage show up there first.
- 04Ask what happens if the model provider ships the same feature, and look for an answer built on data or workflow, not on speed.
- 05Look for data the product generates that competitors can't copy, and evidence that it improves the product for the next customer.
Red flags
- The demo only works on one provider's newest model.
- Most of the codebase is prompt templates and interface code.
- Gross margin falls as customers use the product more.
- "Proprietary model" turns out to be a lightly tuned open model with no unique training data.
Good signs
- Usage produces data competitors can't replicate.
- The product sits inside a workflow customers can't easily leave.
- The team has shipped production systems in this domain before.
The numbers
- CRV's first test for an AI startup is whether a competent team could rebuild the product in a quarter on top of an off-the-shelf foundation model. [1]
- CRV: "Gross margin usually exposes wrapper economics first." [1]
- CRV expects a strong company to explain how it holds up if a model provider ships the same feature, by owning its data or workflow position. [1]