Quick answer: Before signing any dealership AI contract, require written answers to 15 questions across five areas: data ownership and access, model training and intelligence portability, integration and architecture, compliance and accountability, and exit terms. A vendor who answers all 15 clearly is a partner. A vendor who dodges three or more is telling you how the relationship ends. Copy the questions below directly into your procurement process. No attribution needed. We would rather you ask them than credit them.
The demo is theater. The contract is the product.
Every dealership AI purchase eventually comes down to what is in writing, and most dealers discover what is in writing at the worst possible moment: at cancellation, at a data request, or at an audit. This RFP exists so you discover it before signing instead.
Fifteen questions, five sections. Send them to every AI vendor in your evaluation and require written answers. What follows each question is what a good answer looks like and what a bad one is telling you.
Section 1: Data ownership and access
1. What data of ours do you collect, store, or copy, and where does it live? Good answer: a specific map. Fields, systems, storage locations, retention periods. Bad answer: "we take security very seriously." That is not a map. We covered why the map matters in Who Owns Your Dealership Data.
2. Can we export our complete data, including anything you generated from it, at any time, in a standard format, at no charge? Good answer: yes, and here is the format and the process. Anything else means your data has a landlord.
3. Do you sell, share, or aggregate our data with anyone, including anonymized or benchmarked versions? Good answer: a clear yes or no with scope. Aggregation is not automatically bad. Undisclosed aggregation is.
Section 2: Model training and intelligence portability
4. Does our data train or improve your models, and does that improvement benefit our competitors? This is the question vendors least want in writing. If your three years of interactions tune a model your competitor down the street also rents, you are funding their asset. The full economics are in Rented Intelligence Stops When You Stop Paying.
5. If we cancel, what intelligence leaves with you? Scores, profiles, learned behavior, everything. Good answer: an honest inventory. Run the cancellation test on their answer: cancel mentally, list what your store keeps tomorrow.
6. Can the intelligence you build for us run anywhere else, or only inside your product? Portability is the difference between an asset and a subscription.
Section 3: Integration and architecture
7. What systems do you integrate with, and what integration fees are embedded in your pricing? You will not always get a number. Ask anyway. The question changes the negotiation, as we laid out in the integration tax.
8. Do you read from a unified data layer if we have one, or do you require your own copy of our data? Every additional copy of your customer file is another silo, another compliance surface, another version of the truth. Tools built to read from a dealer-owned layer are architecturally ahead of tools that must own a copy.
9. What happens to your product's performance when our data is wrong? Show us, with a duplicated record. The honest vendors will show you degradation. The dangerous ones will claim immunity. AI pilots fail at exactly this seam.
Section 4: Compliance and accountability
10. Are you prepared to operate as a service provider under our GLBA security program, with documentation we can show an auditor? SOC 2 report, contractual safeguards, breach notification terms. A reasonable ask, not a rude one. The full compliance frame is in GLBA and AI.
11. For agents that contact customers: how are consent, recording disclosure, and opt-outs handled, state by state? Good answer: a policy document. Bad answer: "the dealer configures that." Configuration without guidance is liability transfer.
12. When your AI makes an error with a customer, what is your detection, notification, and remediation process? Not whether errors happen. They do. Whether the vendor has a process or a shrug.
Section 5: Exit terms
13. What are the contract term, auto-renewal terms, and full cost of leaving, including data retrieval? Read the renewal clause before the feature list. Always.
14. What is your data destruction process after we leave, and will you certify it in writing? Copies of your customers should not outlive the relationship.
15. Give us two references: one customer who expanded and one who left. The one who left tells you more. A vendor who cannot produce a graceful exit has never had one.
How to score the answers
Simple rubric. Clear written answer: 2 points. Partial or verbal-only: 1. Dodge, redirect, or "that's proprietary": 0. Out of 30, treat 24-plus as a partner, 18 to 23 as negotiate hard with counsel, and below 18 as your answer, regardless of the demo.
One more thing. Answer these 15 for your current stack too, not just new vendors. Most groups find their biggest exposure is already under contract.
FAQ
Can I really send this to vendors as-is?
Yes. Copy it, put it on your letterhead, require written responses. Procurement teams at larger groups should fold it into their standard vendor package.
How does QoreAI score on its own RFP?
Ask us and we will answer all 15 in writing, which is the point. Any vendor unwilling to sit their own exam has told you something.
What if a vendor refuses to answer in writing?
Then the evaluation is complete. Verbal assurances do not survive contact with a dispute. The refusal is data.
Should single stores use this or is it just for groups?
Both. A single store has fewer zeros at stake and the same architecture decisions. The questions do not change with rooftop count.