What are the risks of AI lead handling at a dealership?

Define the action boundary

Ask precisely which actions the system can take without approval: reply, follow up, quote a payment, commit to a price, book an appointment, reopen a closed lead. Get the list, and get the ability to change it.

Require full logging in the system of record

Every outbound message and every action should be visible in the CRM against the customer record, attributed and timestamped. Conversations that live only in the vendor's console are invisible to the manager reviewing the deal.

Handle disclosure and opt-out

Decide how the store discloses automated communication and how a customer opts out or reaches a human. Consent and messaging rules apply to the dealership regardless of who built the tool.

Test the failure mode, not the demo

Ask what happens when the model is uncertain, when a customer asks something out of scope, and when a message would be factually wrong about inventory or price. The answer to that question is the product.

Setting the boundary in writing

Draft a one-page policy that lists what the system may do alone, what needs approval and what is off limits, for example price commitments, credit discussions and legal questions. Share it with the vendor and require configuration to match.

Update it as you learn from transcripts. A written boundary lets you audit behavior against a standard rather than a feeling.

Reviewing conversations

Assign someone to read a sample of conversations weekly, tagging errors by type: wrong information, missed handoff, tone, duplicate contact or consent problem. Report the counts to management and to the vendor, and track whether they fall.

Keep the transcripts as records. The review routine is the mechanism that keeps a tool safe after the demonstration is over.

Common mistakes with AI lead tools

Launching across all lead types at once, having no named reviewer, failing to disclose automation where required, measuring volume instead of results and having no way to switch the system off quickly are the recurring problems. Begin with one narrow use, monitor closely and expand only when the transcripts and outcomes support it. Keep a manual process ready as a fallback.

Questions to ask before a pilot

Which lead sources and hours will it cover? What can it say about price and availability?

How will it identify itself to customers? Who reviews transcripts and how often?

What happens on a complaint? Can we turn it off immediately?

Answer them in writing, and confirm that the vendor's configuration matches.