Procurement and AI in SaaS

If you are selling SaaS to large companies, you should expect to see the buyer proposing clauses which regulate your use of AI. Here are the main ones, with some pushbacks which usually work.

📈Data Use for Model Improvement. The buyer will say: “Supplier can’t use Buyer Data for improving its AI model”.

You should propose: “Supplier can use aggregated and anonymised data for improving its AI model”.

Most buyers can live with this. You may need to add some additional wording to reinforce the fact that the data is aggregated and anonymised and can’t be reversed engineered, and some wording specifying more closely what counts as model improvement, but this compromise generally works.

📱Adding AI Features. The buyer will say: “Supplier can’t add new AI features without buyer’s approval”.

You can propose: “Supplier can add new AI features but will consult the buyer before doing so”.

It’s not a great outcome - who wants to have to consult loads of buyers - but it’s workable and better than giving the buyer a veto right.

For those buyers that insist a veto right, you should have the right to terminate the contract if the buyer exercises its veto. You can’t have a situation where one buyer holds up the whole product.

⚖️Bias. The buyer will say: “Supplier will ensure that its AI is free from bias”.

You can propose: “Supplier will ensure that its AI conforms to applicable AI legislation and Good Practice In The Industry, and will monitor and regularly review its AI for bias”.

This issue only arises if your product is making decisions or recommendations (directly or indirectly) about people, but it’s the trickiest one to handle because a) you don’t control how the LLM was built and b) it’s never that clear how LLMs reach a decision.

There is no simple fix but, if the buyer wants the upside of AI (speed and/or cost) it needs to understand that it can’t be totally insulated from the downside.

23rd September 2025