Toma picks up.

It is 7:42 a.m. in Boulder, Colorado. A service writer at the Nissan store on 28th Street is unlocking the parts cage. The phone rings. Nobody runs for it. The phone gets picked up anyway - politely, on the second ring, by something that is not a person.

The voice on the line is calm. It knows the dealership's hours. It knows the loaner pool is empty until Thursday. It knows the difference between a 2018 Rogue and a 2018 Rogue Sport, which, if you have ever called a Nissan dealer, is more than most humans on the other end of the line know. It books the appointment. It hangs up. The service writer never had to look up from the cage. The voice belongs to Toma.

Toma is a two-year-old company with a stubbornly old-fashioned idea: that the most boring infrastructure in American commerce - the dealership phone tree - is, in fact, where the money lives. Roughly half of a franchised dealer's gross profit is generated by the service drive. The service drive is almost entirely operated by phone. The phone, in most dealerships, is answered by whoever is closest to it. This is not a system. It is an accident that has been happening for sixty years.

01 / The ProblemHalf the revenue is sitting on hold.

Walk into any dealership in America at 9 a.m. on a Tuesday and you will find a Business Development Center, or BDC, which is industry jargon for a room of people who are trying very hard to call you back. The BDC was supposed to fix the phone. It did not. According to internal benchmarks Toma now cites freely, somewhere between 30 and 60 percent of inbound service calls at the average store go to voicemail, are dropped, or are answered by someone who cannot actually schedule the appointment. Every one of those calls is a customer with a wallet open.

The industry has tried to solve this with software before. There are call-tracking platforms, CRM overlays, chatbots that live in the corner of dealer websites and ask "May I help you?" with the enthusiasm of a hostage. None of them, notably, pick up the phone. That, until recently, was a thing only humans could do, and only badly.

02 / The BetTwo engineers move to Mississippi.

Monik Pamecha and Anthony Krivonos met as the kind of restless engineers who had already done the prestigious thing. Pamecha had been at Braze. Krivonos had been at Scale AI. They started Toma in early 2024 with a vague conviction that voice AI was finally good enough for real work, and almost no opinion about which real work. They went through Y Combinator's Winter 2024 batch hawking a more generic voice agent. It went the way most generic things go.

Then they did something the deck-makers in their batch did not do. They flew to Oklahoma. Then Mississippi. They stayed in the kind of motels that share a parking lot with the dealership, and they sat in service lanes for days at a time, listening. They watched a parts manager answer a call mid-bite of lunch. They watched a BDC agent get yelled at by a customer whose name she had no way of knowing. They learned the word "ROs" without anyone having to explain it. By the time they came home, they had a product nobody else had bothered to design.

03 / The ProductAn operating system disguised as a receptionist.

What Toma sells, on the surface, is an AI voice agent. Customers call the dealership; the agent answers, identifies the caller, pulls their history out of the dealer management system, and books a service appointment, orders a part, or routes the call to the one human who can actually help. It does this in about ninety seconds. It does it twenty-four hours a day. It does it in English and Spanish and, increasingly, in the kind of regional small talk that makes a Tulsa customer believe they are talking to a Tulsa person.

Underneath, it is something larger. Toma IQ is the company's learning layer - the part that watches every conversation a dealership has and gets quietly better at being that dealership specifically. The Inbox consolidates calls, texts, voicemails, and missed connections into a single queue, so the humans see the work that actually needs them. Automated recall alerts surface open safety bulletins. Dropped-call SMS catches the calls the AI did not. The whole thing plugs into the existing dealer software stack, which is roughly as elegant as a 1998 Oldsmobile, but Toma does not ask the dealer to rip anything out. That, more than the AI, is the trick.

04 / The ProofThe receipts, in hours and dollars.

The interesting thing about selling software to dealers is that they do not particularly want to be impressed. They want to be paid. Toma's customer case studies, which are unusually specific for an AI company, read like service-bay shift reports. Boulder Nissan: 180 additional appointments booked, 43 hours of staff time returned to the BDC every week. Harvey Auto Group: roughly $3 million in incremental service revenue, 56 hours saved weekly. Martin Management Group: 9,000 appointments routed through the agent, with the BDC workload cut by 40 percent. None of these stores fired their staff. They redeployed them - to the customers in the lane, who, having actually shown up, tend to spend more money.

a16z noticed. The firm's Seema Amble, Olivia Moore, and Eric Zhou wrote in their May 2025 investment memo that they had vetted essentially every voice AI company chasing the automotive vertical, and that Toma was the one whose customers refused to give them up. The Series A came together quickly after that, with checks alongside from Y Combinator, Scale Angels, and the auto-industry commentator Yossi Levi, whose endorsement, in the world of franchise dealers, is worth several term sheets.

05 / The MissionUnderserved industries, served at last.

Pamecha is fond of describing Toma's mission as "AI for underserved industries," which is the kind of sentence that sounds bland until you remember that nearly every venture-backed AI company is currently building for the same six software-savvy buyers in San Francisco. Toma is building for the parts counter in Hattiesburg. The wager is that the dealership is the wedge - that once an AI operating layer has been trusted to run the most labor-intensive service business on Main Street, the same pattern will work for HVAC, for trucking, for clinics, for the long tail of the real economy.

There is a cultural cost to this approach that the founders seem to enjoy. Toma's engineers are expected to ride along with sales, sit with dealers, and answer support pages at hours that San Francisco does not normally honor. The company is roughly sixty-one people. It is, by all accounts, not a place where you talk about the elegance of your agent architecture for very long without someone asking when the next ROs go out.

06 / The TomorrowWhat happens when the phone is, finally, a feature.

If Toma is right, the dealership phone will quietly stop being a problem and start being a product surface. Customers will not notice the AI on the other end. They will notice that the appointment got booked, that the part was in stock when promised, that nobody asked them to repeat their VIN twice. The dealer will not notice the AI either, in the way that nobody notices plumbing that works. Both groups will notice the money.

The broader question - whether the operating layer Toma is sneaking into dealerships will travel to every other phone-bound trade - is unanswered. The skeptical reading is that automotive retail is a uniquely lucrative special case. The more interesting reading is that almost every legacy service business in America looks, when you sit in its lobby long enough, like a slightly different dealership. Pamecha and Krivonos are, at minimum, willing to find out by sitting in the lobby.

Back in Boulder, the parts cage is open. The phone, somehow, has stopped being the loudest object in the room. The service writer takes a sip of coffee for what is, statistically, the first time this week. Somewhere on a server in San Francisco, Toma has just answered another call. It will, in about ninety seconds, be done.