Free guide · AI buying

25 Questions to Ask Every Dealership AI Vendor

A buyer-side checklist for the next AI pitch in your conference room. 25 questions across data ownership, integration, security, pricing, and measurable outcomes.

Updated 2026-07-08·10 min read·No email required·

The new rules of AI vendor evaluation

Every dealership AI pitch in 2026 sounds the same. 'Powered by AI.' 'Trained on millions of records.' 'Increases gross by 17 percent.' The deck is gorgeous and the demo is rehearsed.

The questions below cut through it. Bring this list to the next AI vendor meeting and you will know in 30 minutes whether the product is real, who owns the data, and what it actually costs at scale.

If a vendor cannot answer the first five questions in plain language, the meeting is over. There is no AI product without a real answer to those five.

The 5 must-ask questions (do not skip)

These are the threshold questions. Real vendors answer them in one sentence each. Hand-wavers cannot.

1. Which model are you using, and is it ours or shared with other dealers?

You want to hear a specific foundation model (GPT-4 class, Claude, Gemini, Llama, etc.), an honest answer on fine-tuning, and clarity on whether your data trains a shared model or stays isolated. 'It is proprietary' with no detail is a red flag.

2. Where is our data stored, and who can see it?

Get the cloud provider, the region, the encryption standard, and a clear answer on whether vendor employees can read your data. SOC 2 Type II is the floor, not the ceiling.

3. If we cancel, what happens to our data?

You want a contractual right to a full export in a standard format within 30 days at no cost, plus written confirmation that vendor copies are deleted within 60 days. If they hesitate, walk.

4. What does the price look like at 10x our current volume?

Per-record, per-API-call, and per-user pricing all behave very differently as you scale. Get the math for 10x volume in writing before signing.

5. Can I talk to three dealers using this in production for more than 6 months?

Not a name from a slide. Three customers, your size or larger, willing to take a 15-minute call. If they cannot produce them, the product is younger than the pitch suggests.

Data and integration

Which of our systems do you integrate with, and how?

Get the specific integration method for each system. Direct API, file drop, screen scrape, or third-party connector. Screen scraping is fragile and frequently violates DMS terms. File drops are stable but always lag. Direct API is the gold standard.

How fresh is the data your AI sees?

Real-time, hourly, daily, weekly? For inventory and pricing, anything slower than hourly is a non-starter. For lifecycle marketing, daily is fine. Match the freshness to the use case, do not pay for real-time you do not need.

Do you reconcile to source-of-truth reports?

If their dashboard says 47 deals last month and the DMS DOC says 51, which one is right? Force them to commit to a reconciliation process and a tolerance threshold. 'It should match' is not an answer.

Security and governance

What certifications do you hold, and can we see the report?

SOC 2 Type II is the floor. Ask for the actual report under NDA, not just the badge on the website. Read the exceptions section. That is where the truth lives.

How do you handle PII and PCI data?

Field-level encryption, tokenization, and a documented data-minimization policy. If the AI does not need a customer's SSN to answer your question, it should not have access to it. Period.

What is your breach notification SLA?

72 hours is the GDPR standard and a reasonable baseline. State-level dealer data laws are tightening rapidly. Get it in the contract.

Do you train on our data, and can we opt out?

The honest vendors say yes, they train on aggregated and anonymized data, and yes you can opt out. The dishonest ones say no and bury it in section 14.3.b of the terms. Read the actual contract.

Pricing and contract

What is the all-in monthly cost at our current volume?

Base platform fee plus per-user, per-record, per-API-call, per-integration, and per-store fees. Get a single number for your specific scenario, in writing, before the next meeting.

What is the minimum commitment?

Avoid multi-year lock-ins for unproven AI. Month-to-month after a 90-day evaluation should be the default. Vendors confident in their product offer this. Vendors who are not, do not.

What is included in implementation, and what is extra?

Data mapping, integration build-out, training, change management. Each one can be a five-figure surprise. Get a fixed-fee SOW or a not-to-exceed cap.

Outcomes and proof

What specific business metric will this improve, and how do we measure it?

Closed deals, gross per copy, RO close rate, F&I product penetration. Pick one or two. Agree on baseline and measurement method before you sign. 'Productivity' and 'efficiency' are not metrics.

How long until we see results?

AI vendors often need 60 to 120 days to deliver measurable lift. Anyone promising 30-day ROI on a new AI deployment is selling a screenshot, not software.

What does success look like at month 12?

Force a specific answer. Number of users actively in the tool weekly, percentage of decisions touched by the AI, measurable lift on the chosen KPI. If they cannot describe success, neither can you.

Frequently asked

How long should an AI vendor evaluation take?+
From first demo to signed contract, 30 to 60 days is healthy. Faster than 30 days means you skipped due diligence. Slower than 90 days usually means the vendor or the buyer cannot describe success clearly.
Should we run a pilot first?+
Yes, almost always. A 60-day paid pilot with one rooftop, one use case, and one success metric agreed in writing. Pilots that lack a success metric become permanent and expensive.
What if the vendor will not answer the security questions?+
That is your answer. A vendor that cannot describe its security posture in 15 minutes either does not have one or is hoping you do not ask. Either way, do not buy.
Is it OK if our AI vendor also runs our marketing or DMS?+
Be careful. Vendors that monetize your data in a second business have an incentive to keep it locked inside their walls. The safest pattern is to keep the data infrastructure layer (where the AI reads from) separate from the application vendors (who consume it).

Want to put this into practice across your store?

QoreCloud is the dealership intelligence platform. We unify the systems, govern the data, and put answers in your team's hands in seconds. Live in 60 days.

Book a Demo