Your Phones Are Being Answered. Only 22% of You Are Seeing the Money

The phone gets answered now. That part isn't in dispute.

Walk into almost any dealership's back office in late 2026 and you'll find some flavor of conversational AI running — answering inbound calls, texting leads back, nudging a lapsed service customer about an open recall. CDK's 2026 Friction Points Study puts adoption at 39% of dealers, up from 28% a year ago. Spyne's own sentiment survey (vendor-commissioned, so season to taste) claims 74% of dealers are now putting budget specifically behind voice agents. Cox Automotive says the number using AI in some form is up to 82%.

Those are the easy numbers. The hard number, also from Cox, is the one that should be taped to every GM's monitor: only 22% of dealers using AI report actual sales or revenue growth from it. Sixty-nine percent expected it would. Less than a third got there. That gap is the entire story right now — not whether AI can answer a phone, but whether anyone's bothering to measure what happens after it does.

Where the money actually showed up

Two deployments this quarter had enough specificity to take seriously. Spyne's Vini AI, now live at 115 US dealerships and reportedly handling 450,000-plus leads and 900,000 conversations, published 30-day numbers from Paragon Honda: $314,000 in recovered revenue and a 48% appointment-to-sale rate. That's not a vague "engagement lift" — it's a dollar figure tied to a defined window, which is rare enough in this space to be worth noting.

On the fixed-ops side, Bob King Mazda — about 700 repair orders a month — ran Vini for after-hours and peak-period service calls and reported 30% higher demand capture and a 12% bump in service appointments, with full after-hours coverage added on top. Service is the right place to point a voice agent first: it's high call volume, low complexity per call, and every missed call after 6pm is a competitor's open slot tomorrow.

Then there's STELLA, which just picked up a $22 million round led by a car dealer rather than a typical VC — itself a signal that people who run stores for a living think this specific wiring (agents tied into the scheduler, the CRM, and outbound follow-up via its Catalyst product) is worth betting on. STELLA's pitch is explicitly aimed at the service drive too: booking directly into the scheduler, chasing lapsed customers and open recalls, not just answering the phone and hoping.

Where it's probably just a subscription

Notice what both of the credible wins have in common: a named store, a defined time window, and a specific metric tied to revenue or appointments — not "engagement" or "sentiment." The Dealer AI Guy's weekly recap lays out the diagnostic dealers should actually run before signing anything: pull your 90-day lead-defection rate, time-stamp your real after-hours response gap (not your BDC's stated SLA), pick one metric, run the agent on overflow calls only for 30 days, and compare show rate — not vibes — against your human team's.

That last one matters most. A voice agent that books an appointment nobody shows up to isn't solving your problem, it's just moving the no-show further down the funnel where it's harder to see. If your vendor's success story is "handled X thousand conversations," ask the follow-up question nobody at the sales demo wants to answer: how many of those became a repair order or a sold unit, and compare it to what your existing team was already closing on the same calls.

The integration layer is quietly consolidating

Worth flagging for anyone shopping right now: Vini is integrated with VinSolutions, Tekion's Automotive Retail Cloud, and — as of this quarter — CDK via the Fortellis Marketplace. STELLA claims the same trifecta: Tekion, CDK, Reynolds & Reynolds. That's not nothing. A year ago, "which DMS do you run" was a disqualifying question for half these vendors. Now it isn't, which means the real competitive fight is shifting from "can you plug in" to "can you prove it pays."

The bottom line: the phone-answering problem is basically solved, by multiple vendors, at scale. The revenue-proving problem is not — 78% of dealers using this stuff are still waiting, or not looking. Before you sign anything, demand the Paragon Honda treatment: a named metric, a 30-day window, and a show-rate comparison against your own team. If your vendor can't produce that, you're not buying AI. You're buying a more articulate hold message.