What to Ask an AI Vendor in Automotive Before You Sign
Every AI pitch in automotive sounds the same right now, which is a problem for you and, honestly, for us. The decks use identical words. The demos are rehearsed on data that always cooperates. And the differences that matter — what happens to your data, what the thing does when nobody's watching, who's liable when it's wrong — don't come up unless you raise them.
So here are the questions worth asking, and what a good answer versus a bad answer actually sounds like. Use them on us too.
1. "Show me something your system did last month that nobody asked it to do."
The single most useful question in the category, because it can't be answered with positioning.
Good answer: a specific example with a date. "On the 14th it flagged that this source stopped delivering and opened a ticket."
Bad answer: a demo of someone typing a question into a chat box. That's a search interface with better manners. Genuinely useful, but not autonomous, and it shouldn't be priced as though it were.
2. "What data sources are you connected to — and what can't you see?"
An AI can only reason about what it's connected to. One scoped to your ad platforms will confidently explain a lead drop in marketing terms when the real cause was an inventory gap.
Good answer: a specific list, plus an honest statement of what's out of scope.
Bad answer: "we integrate with everything." Nobody integrates with everything. Ask which DMS, which CRM, and whether the connection is live or a nightly file.
3. "Who owns the data, and what happens to it if I cancel?"
Good answer: you own it, you can export it in a usable format, and here's the specific process and timeline.
Bad answer: anything vague. Follow up with: can I export the historical data, or just the current snapshot? Two years of unified history is worth real money, and losing it on cancellation is a switching cost dressed up as a technicality.
Also ask whether your data is used to train models other dealers benefit from. There isn't a single right answer — but you should know which one you're agreeing to.
4. "Does this talk to my customers?"
If yes, it's a different product with a different risk profile, and it needs its own evaluation: compliance exposure, brand voice, disclosure requirements, and a failure mode that becomes a public review.
Good answer: a clear yes or no, and if yes, specifics on disclosure, escalation to a human, and what happens when the AI is wrong.
Bad answer: blurring analytical AI and conversational AI into one pitch so the capabilities of one imply the other.
5. "What happens when it's wrong?"
It will be. The question is what that costs.
Good answer: recommendations carry the underlying data so your team can check the reasoning; actions that move money require approval; here's the audit trail.
Bad answer: confidence. Anyone claiming their system doesn't make mistakes is telling you they haven't measured.
6. "What does implementation actually require from my team?"
Good answer: a specific list of credentials and access, a realistic timeline, and who at your store has to do what.
Bad answer: "it's plug and play." Ask how long until it's producing something you'd act on. If the honest answer is six weeks, a vendor who says two days is either wrong or about to hand you a lot of homework.
7. "What are you not good at?"
A vendor who can't name a limitation either doesn't know their product or isn't being straight with you. Both are disqualifying.
We'll go first: this category doesn't fix broken processes. If leads aren't being worked, better lead scoring produces a better-sorted list of people nobody calls.
8. "How is this priced, and what's gated?"
Ask specifically: is there a separate AI license? Per-seat charges? Usage metering that scales with how much you actually use it? What's held behind the next tier?
Tiered pricing in this industry is usually about access rather than work — the next tier unlocks features rather than adding effort. Worth knowing which levers are being held.
9. "What's the contract term, and what's the out?"
Good answer: clear term, clear notice period, no auto-renewal trap.
Bad answer: a long term justified by "it takes time to show value." Sometimes true. Also the standard argument for locking in a customer before they can evaluate results.
10. "Can I talk to a dealer using it who isn't on your reference list?"
References are curated. Asking for one that isn't tells you how confident they are in the median customer rather than the best one.
The meta-question
After all ten, ask yourself one thing: did the vendor's answers get more specific as the questions got harder, or less?
Good products get more concrete under pressure, because there's something underneath. Weak ones retreat upward into abstraction — "holistic intelligence," "transformative," "purpose-built." Watch the direction of travel. It's more reliable than any individual answer.
Where we'd tell you not to buy
If your data is fragmented across systems that don't talk, no AI layer fixes that — the intelligence is bounded by the plumbing, and the plumbing is the unglamorous part that has to come first.
If you have someone in-house who's good and has the hours, keep them.
And if what you actually need is enterprise reporting standardized across a large group for your 20 group, buy that. It's a legitimate product and it isn't what we are.
Astra is the agentic layer; Control Center is the data layer it reasons across. We'd rather you run these ten questions at us than take the deck at face value — and if the answers don't hold up, that's useful information too.
