What will enterprise customers ask before buying my AI product?
The AI questions in enterprise security reviews and contracts, including model training, data retention, AI subprocessors, data processing terms, output risks, and certifications, and how to have the answers ready.
Short answer
Enterprise buyers will ask where their data goes, whether it trains any model, how long it's kept, which AI providers touch it, and who is responsible when the AI gets something wrong. The questions arrive as a security questionnaire, a data processing agreement, and AI-specific contract terms. Most of the answers already exist in your code, provider settings, and policies. Writing them down once, accurately, shortens sales cycles and keeps your team from promising something the product doesn't do.
What they'll ask
Training and data use
- Does our data train your models, or anyone's? This is usually the first AI question. Expect a contract term prohibiting training on customer data, often including de-identified or aggregated data. Know your own practice and your AI providers' settings for the plan you're on. See Can I train or fine-tune a model on my users' data?
- Retention. How long prompts, outputs, files, and logs are kept by you and by your AI providers, and whether zero or shortened retention is available.
- Isolation. Whether one customer's data can show up in another customer's results, for example through shared retrieval indexes, caches, or memory features.
AI providers as subprocessors
- A subprocessor list. Customers expect a list of every vendor that processes their data, including model providers, hosting for open-weight models, vector databases, and observability tools, with notice before you add new ones.
- Flow-down terms. Your commitments have to be backed by your providers' terms. If you promise no training and short retention, your provider agreements need to say the same. See Can I send user data to an LLM provider?
- Location. Where data is processed and stored, and how you handle transfers out of the EU or UK if you have customers there.
Data processing terms
- California. When you process personal information for a business customer as its service provider, the CCPA regulations require a written contract that, among other things, names the specific business purposes, prohibits selling or sharing the data, prohibits using it for other purposes or combining it with data from other sources except as the rules allow, and gives the customer the right to take reasonable steps to check your compliance. The rules let a service provider use the data to improve its own services, but not to perform services for another business. Enterprise contracts often go further and prohibit training entirely.
- The EU and UK. Under the GDPR, a processor contract has to cover processing only on the customer's instructions, confidentiality, security, subprocessor approval, help with individuals' rights requests, deletion or return of data at the end, and audits.
- Your own DPA. Having a standard data processing agreement ready, one that matches what your product actually does, avoids negotiating a different one with every customer.
Security and certifications
- Questionnaires. Expect questions about access controls, encryption, logging, incident response, and vulnerability management, plus AI-specific questions about prompt injection, output filtering, and who can see customer prompts.
- Certifications. SOC 2 reports are the common ask in the US, and ISO/IEC 27001 internationally. Some buyers now ask about ISO/IEC 42001, the standard for AI management systems. Early on, a clear written security overview and honest answers can carry a deal while a certification is in progress.
- Incidents. How quickly you'll notify the customer of a security incident affecting its data, often 24 to 72 hours in enterprise contracts.
Outputs and liability
- Accuracy. Buyers will ask how you test outputs and what you say about errors. Your contract should be clear that outputs can be wrong and need review where it matters, and your marketing should match. See What can I claim about my AI product in marketing?
- Ownership. Who owns inputs and outputs. See Who owns AI-generated output?
- Indemnities. Customers increasingly ask for indemnities against claims that outputs infringe intellectual property. Know what your AI providers offer you before agreeing to offer it onward.
- Caps. Limits of liability, and whether data breaches or confidentiality get a higher or uncapped amount. These are business decisions worth making deliberately, not clause by clause under deadline.
AI-specific laws
Buyers in hiring, lending, insurance, health, and education will ask whether your product makes or supports consequential decisions, and how you help them meet laws like Colorado's automated decision law and California's automated decision rules. See Does Colorado's automated decision-making law apply to my product?
How to get ready
- Write an AI data use statement. One page on training, retention, isolation, and providers, kept accurate as the product changes.
- Publish a subprocessor list, with a way for customers to get notice of changes.
- Prepare a standard DPA that matches your product and your providers' terms.
- Keep a security overview and a completed standard questionnaire you can share.
- Decide your positions in advance on training, indemnities, liability caps, and breach notice timing.
- Track promises. Keep a list of non-standard commitments made to each customer, so the product keeps them.
Checklist an agent can run against the codebase
- List every AI provider and vendor that receives customer data, for comparison with the subprocessor list.
- Check the training and retention settings configured for each AI provider account.
- Find where customer data could reach training, fine-tuning, or evaluation sets.
- Check that retrieval indexes, caches, and memory features are separated by customer.
- Find where prompts and outputs are logged, and the retention for each log store.
- Check where data is stored and processed, by region.
- Check who can access customer prompts and outputs internally, and whether access is logged.
- Compare the code with the AI data use statement, DPA, and security overview.
Sources
- California Code of Regulations, title 11, section 7051 (service provider contracts)
- California Code of Regulations, title 11, section 7050 (permitted service provider uses)
- GDPR Article 28 (processor contracts)
- AICPA: SOC 2
If this applies to you
An AI feature review prepares the AI answers enterprise buyers ask for: how customer data flows to AI providers, training and retention commitments your providers' terms support, and privacy documents that match the product. For one question, such as whether you can promise a customer zero retention with your current provider, a pre-build feature check covers that decision before you commit to it.