AI for auto dealerships: leads, service calls, and follow-up

A dealership does not need an AI salesperson. It needs every web lead answered, every service caller given a next step, and every promise written back to the system the staff already use. The useful AI is the layer that stops a busy store from quietly losing conversations.

This guide is for independent dealers and dealer groups in Canada and the US. Automotive AI research is unusually vendor-heavy, so survey results are labeled for what they are. The most valuable benchmark in the article is not an AI claim at all: it is measured lead behaviour across thousands of dealership websites.

Adoption is real; independent proof is thin

CDK Global reports that 39% of dealers were using AI in 2025, with another 27% planning to adopt it within a year. Among users, 69% said it had a positive impact. CDK sells dealership software and AI products, so these are vendor survey findings—not neutral proof that a particular tool increased gross profit. They do show that AI has moved into ordinary dealership operations.

The independent operating question is simpler: where is the store already paying for delay? Foureyes reports sales-tracking data across more than 22,900 US dealership websites. Its 2025 benchmarks put the share of sales leads that eventually buy at about 11.7%; its separate 30-day measure reaches 16.2%, with roughly three-quarters of closes happening in the first three days. Different windows create different rates, which is exactly why a dealer should define “close rate” before showing a dashboard to the GM.

The sales funnel: respond, route, remember

AI works best at the edges of the sales conversation: reading an inquiry, identifying the vehicle and intent, drafting a grounded reply from live inventory, asking one useful qualification question, and assigning a human owner. It can also summarize the thread when the BDC or salesperson takes over. The system must know the difference between “Is this unit available?” and a buyer asking what financing they will qualify for.

• Good automation: after-hours acknowledgement, VIN-specific information pulled from the inventory feed, appointment options, trade-in intake, reminders, and escalation after silence.
• Human-controlled: price negotiation, trade valuation, credit discussion, payment quotes, disclosures, and anything represented as a firm offer.
• Never fabricate: availability, condition, warranty, fees, rates, delivery dates, or features. If the source system is stale, the assistant must say it needs confirmation.

The dealership wins when AI makes a salesperson easier to reach—not when it performs a longer impersonation of one.

The service lane may be the better first project

J.D. Power’s 2025 Customer Service Index found long appointment waits and communication gaps still limiting service satisfaction. Four of its ten most influential service indicators were communication-related, including keeping the customer informed and following up after service. Twelve percent of repairs were not completed correctly on the first visit; only half of those customers said they returned or planned to return to the dealership.

That points to a bounded workflow: book routine service against real capacity, collect symptoms without diagnosing them, send status updates from the repair order, obtain approval through the approved process, and confirm pickup. A voice or text assistant can take the repetitive “Is my car ready?” traffic off the adviser’s queue. It should not invent a completion time, interpret a warning light, approve warranty coverage, or turn a technician’s note into a stronger claim than the note supports.

The money is in recovered funnel steps

Use four numbers for sales: unanswered leads, median first-response time, appointment rate, and sold rate. Use four for service: abandoned calls, appointments booked, repair orders per appointment, and status calls per open order. Compare a four-week baseline with a controlled pilot, broken out by source and hour. A store that answers faster but books low-quality appointments has moved the bottleneck, not fixed it.

Voice-AI vendors publish dramatic recovered-call stories, including Numa’s account of 6,300 calls handled in 30 days at one Honda store. Treat that as a vendor case, not a transferable forecast. Ask for the denominator, the call types, the percentage that reached a completed booking, the human escalation rate, and the number of appointments that actually showed.

The honest bottom line

• Start with response coverage. Inventory questions, booking, routing, and summaries are useful and reversible.
• Connect to live systems. An assistant without inventory, calendar, CRM, and repair-order context is a polished answering machine.
• Keep deals and diagnoses human. Financing, disclosures, trade values, repair advice, and firm promises need an accountable employee.
• Measure downstream. Appointments, shows, sold units, and completed repair orders matter; messages sent do not.

Our AI implementation work begins with the dealership’s actual call and lead flow. The free 7-day audit gives you the map before a platform contract does.

The free lead calculator models automotive search leads at about $44 and uses the measured 11.7% to 16.2% sales range. It is a market benchmark, not a promise for one rooftop.

Sources

• CDK Global: AI at the Dealership Study 2025 (vendor research)
• CDK Global: 2024 survey of nearly 250 North American dealership leaders (vendor research)
• Foureyes: 2026 Automotive Dealer Benchmarks Report (platform data across 22,900+ sites)
• Foureyes: 30-day dealer close-rate analysis
• J.D. Power: 2025 US Customer Service Index Study
• Numa dealership phone case study (vendor claim)
• Office of the Information and Privacy Commissioner for BC: private-sector PIPA guidance