buying · 2 min read
Who owns the AI you pay for?
Code, evaluation sets, prompts, indexes, the data the system learned from: each is owned by someone, and the answer differs between a product, a platform and a build. What to ask before you sign.
When a business pays for an AI system it is paying for several things at once, and each can be owned by a different party: the code, the prompts and instructions, the evaluation set, the indexes built from its documents, the data the system produces, and the model underneath. Products, platforms and tailored builds answer the ownership question differently, and the differences decide what happens when the price changes, the vendor pivots, or the business wants to move.
The layers
| Layer | Product | Platform (configured assistants) | Tailored build |
|---|---|---|---|
| The model | Vendor’s | Vendor’s | Vendor’s, swappable |
| The application code | Vendor’s | Vendor’s | Set in the contract |
| Prompts and instructions | Vendor’s | Yours, inside their platform | Set in the contract |
| Indexes of your documents | In their service | In their service | Can sit in your environment |
| Evaluation set | Usually none | Usually none | Agreed before the build |
| Outputs | Yours, per terms | Yours, per terms | Set in the contract |
| Your data for training | Check the tier | Check the tier | Never, by design |
Why the terms matter
Portability. A build on a model-agnostic framework can change model when a better or cheaper one appears, because the change is configuration. A product cannot; you move when it moves.
Price stability. Know how the price behaves as use grows and at renewal, and what a managed service to run the system costs each month for defined work.
Continuity. Agree what happens if either side walks away: what you keep, what you can go on using, and on what terms. If a product’s vendor disappears, you have a login that stops working.
Evidence. The evaluation set is the record of what the system does and how well. It is what an auditor, a regulator or a customer asks for, so agree who holds it and how you can get it.
Where products are still right
At the base of the Opportunity Pyramid: copilots, assistants, enterprise search, the standard functional tools. Nobody needs bespoke terms for a meeting summariser. The terms matter where the knowledge and the workflow are specific to your business, which is the territory of a tailored build.
Questions to ask before signing
- Who owns the application code, and is it delivered with the system?
- Who owns the prompts and instructions, and can we read them?
- Where do the indexes of our documents live, and can we take them?
- Is there an evaluation set, and is it ours?
- Can the underlying model be changed without a rebuild?
- Under what tier and terms is our data processed, and is it ever used for training?
- What happens on exit: export and deletion?
Our position
Ownership and licensing are agreed in each contract, in writing, before the build starts. Your data is never used to train models, and the frameworks underneath are model-agnostic by design. That is described under AI development, and the honest version of the product answer is in every roadmap we write.