The GM told me on the first call: "we're getting the leads, we're just not closing them." I've heard that sentence from about forty dealers now. The words are always identical. The math is always the same. It's not a lead-generation problem — it's a response-time problem, and the agency they were paying wasn't equipped to solve it.
What they were paying for
The previous agency delivered four things per month: a monthly social calendar (12 Facebook posts, mostly stock imagery with a "we care about your family's next car" tagline), a Google Ads account manager who sent a monthly PDF report, boosted-post campaigns on the specials, and a quarterly "strategy call." Total invoice: $2,400/month + ad spend.
That's a completely reasonable service. It's what an agency for a mid-sized dealer group looked like in 2018. It's just not what closes deals in 2026.
Because the actual problem — 1,200 inbound leads a month across three rooftops, salespeople trying to work them between customer walk-ins, most of them cooling off within 12 minutes because the dealer down the road called first — none of that was ever going to be solved by 12 Facebook posts and a PDF.
What we did in 14 days
I flew to Albany on a Tuesday. Sat with the sales manager and pulled his last 30 days of Facebook Marketplace, Cars.com, AutoTrader, and CarGurus leads into a spreadsheet. Marked which ones the team had actually replied to, and how long it took.
Ninety-seven out of every hundred leads were going cold before the salesperson said hello. That's not a training problem. That's a physics problem. There are only so many hours in a workday and only so many salespeople on a floor.
Day 1-3: Setup
We integrated with their DMS (Reynolds & Reynolds) via their standard API, wired in Twilio for the phone numbers, registered the three rooftops for A2P 10DLC, and pulled 60 seconds of their sales manager's voice for the callback voice clone.
Day 4-7: Content build
We built out the specific scripts. This is the part every "AI voice startup" gets wrong — they use a generic template. We don't. The Ford F-150 script is different from the Chevy Silverado script is different from the used Toyota Camry script. Brand voice, feature knowledge, common objections, financing lever points — all mapped per major vehicle class.
Day 8-11: Test conversations
I personally listened to 40 test calls with the sales manager. Adjusted tone, adjusted pacing, adjusted the specific phrases that made Rachel sound too "AI" versus warm. Rachel got a slight upstate NY inflection. Small thing. Made a real difference in the response rate.
Day 12-14: Live
We flipped 10% of inbound to the AI on Monday morning. 50% by Wednesday. 100% by Friday. The rest is Rachel's job.
The three-touch cadence — where the numbers come from
Every lead across every channel gets the same three-touch treatment inside the first 5 minutes:
- Touch 1 — SMS in under 60 seconds. Mentions the specific vehicle they inquired about, offers two appointment windows.
- Touch 2 — Rachel voice call if the SMS is unread after 5 minutes. Warm, honest, discloses AI status, gets to the point.
- Touch 3 — Warm transfer to the assigned salesperson (via MAGS routing to the person who sold to their zip code last), or direct booking to the salesperson's calendar via the DMS.
What happened next
Ninety days in, across all three rooftops:
- Lead-to-appointment rate went from 12.0% to 17.5%. On 1,200 monthly leads, that's +66 additional appointments per rooftop.
- Test drives, specifically, went up by 22 per rooftop per month.
- At their historical 25% close on test drives and $4,200 average gross, that's ~$23,000 in incremental monthly gross per rooftop.
- Across 3 rooftops, that's $69,000/month.
- Glovecars monthly fee for the 3-rooftop deployment: about 15% of that.
The right way to think about our fee is: it's whatever it takes to keep the ratio right. You should be spending 10-20% of what we deliver. When we deliver more, you pay more. When we deliver less, we've earned nothing.
What we did NOT do
We didn't touch their creative. Their existing agency does great creative — we told them so. We just did the response layer.
We didn't run their Google Ads. Their agency does that fine too. We just made sure every click that landed hit a system that answered back in 60 seconds.
We didn't recommend they fire the agency. They did, six months in, because they didn't need what the agency was delivering anymore. That was their call.
What we told them not to do
They asked about deploying our AI to their service department too. We told them not yet — the service department has a different acoustic profile (mechanics walking around, tools) and needs a different voice profile. We're staging that for month six.
They asked if we could clone the previous owner's voice for the callback. We said we could — but recommended against it. The previous owner passed away in 2019. Cloning a deceased person's voice for a sales callback is legal and technically simple, but ethically we don't do it. Their current GM's voice was fine.
The point
Most dealer groups already have a working operation. They already have a marketing agency. They already have Facebook Marketplace, they already have DMS, they already have salespeople who know cars. The gap is the 12-minute window between lead-in and human-out — and that gap is what Glovecars closes.
The math is fixed. Speed-to-lead compounds. It doesn't matter how good the sales team is on the floor if the lead is cold before they see it.
The strategy call is the free version of what I did on my first Tuesday in Albany. I pull your last 30 days of inbound and show you the actual conversations that would have converted with a 60-second response. Then you decide.