We delivered an AI Sales Operations Manager for a North Charleston architectural millwork company — one that takes the transcript from a salesman's face-to-face calls, updates the CRM and the knowledge base, and writes the proposal, addendums and contract with no human involved until a finished draft lands in his inbox for one decision: send, or send back.

He has sold custom millwork in Charleston for 31 years. Bars, reception desks, hotel lobbies, the paneling in a dozen restaurants on King Street, the judge's bench in a courthouse in Summerville. The company is a 38-person shop off Rivers Avenue in North Charleston, and for most of its history he has been the entire sales department. He closes in person.

He sits in the client's unfinished space with a tape measure and a legal pad, and two hours later they have agreed on the wood, the finish, the edge profile, the hardware, the delivery date, what happens if the drawings change, and who pays for the crane. He is very good at this part. What he is not good at, and has never been good at, is the part that comes after. Every one of those two-hour conversations used to produce four to six hours of desk work.

Typing up the notes. Entering the deal into the CRM, which he did late and thin. Writing a proposal that captured 40 decisions made verbally, of which he would remember 32. Drafting the contract, the addendums for the items that were still open, the schedule letter, the deposit invoice.

On a good month he did 18 of these calls. The paperwork ran a week behind, and a week is long enough for a general contractor to get a second quote. The AI Sales Operations Manager starts with a Plaud Pro on the table. He sets it down at the beginning of the meeting and forgets it.

It records and transcribes the whole conversation, the tape-measure asides included. That is the only thing that changes about the meeting. Everything after that happens without him. When the transcript arrives, the Sales Operations Manager reads it and produces a structured summary: every decision, every open item, every number, every name, every promise made in either direction.

It writes the deal into the CRM with the contacts, the stage, the value, the next step and the date. It writes the specifications and the client's stated preferences into the company's internal knowledge base, so that the project manager and the shop foreman see them the same afternoon rather than hearing about them three weeks later on the floor. Then it writes the documents. The proposal, usually 11 to 14 pages, with the scope, the materials, the finish schedule, the exclusions, and the pricing built from the company's own price book and the last four comparable jobs in the knowledge base.

The addendums for whatever was left open in the room. The contract, on the company's own template, with the payment terms and the change-order language he agreed to out loud. The schedule letter, which it checks against the shop's current backlog. Where a public fact matters, it goes and gets it: the permit status of the client's building from the county, the flood zone if the job is on Sullivan's Island, the client's own website when the proposal should describe their brand back to them accurately.

It reads the drawings the architect sent and reconciles them against what was said. No person touches any of this. About 40 minutes after he walks to his truck, an email is waiting. The full package, attached, with a short note on top: what was decided, what is still open, and anything in the transcript that looked like a risk.

He reads it. If it is right, and it usually is, he hits send and it goes to the client from his own address. That is his entire job now, after the handshake. If it is not right, he does not fix it.

He forwards the email to a second AI, the Editor, with one line: "Take out the brass pulls, they went with black. Push delivery two weeks. Cut the crane." The Editor takes the Sales Operations Manager's package, makes exactly those changes across every document where they appear, and returns a fresh draft to his inbox.

He has never opened a proposal in a word processor since June. The numbers are what you would expect and still surprising. Paperwork per call went from about five hours to about eight minutes of reading. Proposals now go out the same day in 94 percent of cases, against a prior average of six days.

He is doing 27 calls a month instead of 18, because the afternoons came back. The close rate on same-day proposals is 61 percent, against 44 percent for the ones that used to take a week, which the company had suspected and could never prove. And the shop foreman says the specs are right when they reach him for the first time in 31 years. "I sell.

It does the rest. I read my email and press one button. That's the job I thought I had when I started."