If you follow a good salesperson around for a week, you’ll notice how little of it is selling. Most of the day goes to answering new leads before they go cold, chasing the ones who went quiet, booking and rebooking appointments, and writing it all down. By the time a buyer is actually standing on the lot, there isn’t much of the day left for them. We started RevDesk to give that time back.

The work around the sale

Buying a car, a repair, or a new roof still comes down to trusting the person across the desk. Nobody we’ve met, buyer or salesperson, wants a model in that seat. The rest of the job is a different kind of work: the first reply to a 10pm web form, the third follow-up to a buyer who went quiet, the reminder the night before, and the reschedule when something comes up. Those jobs need speed and persistence more than they need a person. They’re also exactly the ones that slip when a team is busy with the customers already in the store.

Our agents take that work for dealerships and service businesses every day. They reach new leads, usually within seconds of a web form. Then they keep the conversation going across texts, iMessage, WhatsApp, email, and calls until the buyer books, replies, or asks them to stop. When it’s time for a person, the agent hands the buyer to a salesperson with the whole conversation in front of them.

What the internet left out

Language models learned from what people wrote down. That covers a lot: product pages, reviews, sales books, and years of forum threads about haggling over a trade-in. It’s more than enough to explain how a lease works or draft a polite follow-up.

What it leaves out is whether any of it worked. Nobody publishes their lead follow-up. The few threads that end up online never include what the buyer did afterward. A model could read every dealership website in the country and still have no idea why one shopper booked a test drive while the next one stopped replying. Without the ending, all that text can teach is style, and style was never the hard part.

Logistics still takes judgment

Following up sounds mechanical until you watch someone who’s good at it. They can tell that "what’s my trade worth?" sent at 10pm means the buyer is close, and that "just looking" on a Saturday morning usually means the opposite. They know when a call will land better than a text. They also know whether a buyer will welcome a third follow-up or it’s time to stop.

Those are judgments about timing and about people. In a transcript they show up only as a choice someone made, with no record of whether it was the right one. People pick them up by trial and error. We think a model has to learn them the same way.

Where the missing data lives

The data that teaches this only exists in production, in conversations whose ending is known. When something comes of a RevDesk conversation, whether that’s an appointment set, an appointment kept, or a unit sold on the lot, the outcome is recorded against it, along with whether the agent got there on its own or with a salesperson’s help.

The internet never had that record. It belongs to the businesses we work for, which is why the way we use it matters as much as the data itself.

Two models, one promise

We’re building two things that work together. The first is a single global model, the judgment layer that decides when to reach out, what to say, when to stop, and when to bring in a person. It’s built on foundation models, licensed and public material, and conversations from businesses that choose to contribute them, which they can only do once an admin turns contribution on.

The second is an agent for each business, which learns that business inside its own workspace from its knowledge, its playbooks, and how its own conversations ended. It learns by remembering rather than training. It only ever works for the business that runs it.

Your data and conversations never train anyone else’s models, and they train ours only if you choose to help improve them.

The trap in rewarding sales

Anything rewarded for sales will eventually find shortcuts. Pressure works once, and so does a deadline that isn’t real. Each of them spends the trust that brings a customer back for service, for their next car, and one day for their kid’s first car. The people on the sales floor build that trust. An agent’s job is to protect it.

Some of the guardrails don’t depend on the model at all. Consent, quiet hours, and attempt limits are checked by rules before a message is ever written, so there’s no way for the model to talk its way past them. When someone replies STOP by text or iMessage, the calls, texts, and iMessages to that person stop. Every decision an agent makes is written down with its reason, so a manager can always see why it reached out.

The harder part is the score itself. We want a model that helps win the second sale as well as the first. That means a sale that costs the customer’s trust can’t count as a win. We’ll keep writing here about how we measure that.

The bet

Language models learned to write by reading what people had written. We think the next skill they pick up will come the way people learn it, by trying something, seeing what happened, and doing a little better the next time. The work around a sale is where that loop is easiest to see. The ending gets written down: the test drive either happened or it didn’t.

Our agents hold these conversations every day for real businesses. Every ending is recorded. The work ahead of us is turning that record into a model that handles everything around the sale well enough that salespeople can spend their day with the buyers in front of them.