Build an AI Agent That Turns a Sales Call Into a Finished Quote

Every quote you send says the same thing twice. Once in the sales call. Once in the price you type up afterward, from memory.

Quoting Is the Job Nobody Automates First

Marisol runs a four-person marketing agency in Austin. Every week she takes two or three discovery calls with new clients. Every week she does the same thing after each one.

She opens last month's proposal. She copies the format. She scrolls back through her notes to remember what the prospect actually asked for. She checks the rate card in her head. She writes it up in her own words, in her own template, hoping she did not forget the add-on the client mentioned at minute forty of the call.

An hour later, she has a quote. Sometimes two.

Nobody starts a business because they love writing quotes. And yet quoting is often the first real thing a prospect sees from you: the moment they decide whether you are organized enough to trust with real money.

Here is how quoting usually works, on repeat, at agencies, contractors, consultancies, and service shops everywhere:

• A sales call happens. Nobody writes it down cleanly.
• Whoever owns the relationship remembers the details, or they do not.
• The quote gets built from memory, an old file, and pricing rules that live in someone's head.
• Every quote reads slightly differently, because every quote was handwritten under time pressure.

You are not slow. Your quoting process has no memory. It resets to zero after every call, and you rebuild it from scratch, every single time.

What an AI Agent Actually Needs to Write a Quote

A chatbot answers a question. An agent finishes a job.

Ask ChatGPT or Claude, cold, to write you a quote for an IT support contract, and you get a generic template. It has no idea what you charge, what you offer, or how you phrase things. That is the chatbot experience, and it is why most owners try AI for this once and give up.

An agent is the same model, with your business sitting next to it in a folder.

That is the whole trick, and it is simpler than it sounds. IT consultant Curtis Branum demonstrated it on his channel in June: he opened a plain folder on his computer, named after the client project, and inside it were four things.

1. A short instructions file (in his case, a CLAUDE.md) stating the agent's one job: turn a discovery call into a clean, accurate, client-ready quote.
2. A company file with his pricing rules, in his own words.
3. A services file listing his packages and what is included in each.
4. A folder of past quotes, so the agent matches his tone and formatting instead of inventing its own.

Drop the call transcript into a "calls" folder, type one sentence asking for a proposal, and the agent reads all of it: the transcript, the pricing rules, the old examples, and writes a finished quote that already matches how the business prices and talks.

Why This Works Now, and Not Two Years Ago

Two years ago, giving an AI model your pricing meant pasting it into a chat window every single time and hoping it remembered by the next message. It did not. Every conversation started over.

What changed is memory that persists in files instead of a chat window. Anthropic's Claude Code reads a project folder the way a new hire reads an onboarding binder, every time it opens, without you retyping anything. Feed it your business once. It reuses that context for every quote after.

This is not limited to construction or IT services. Agencies, consultants, contractors, studios, anyone who turns a conversation into a priced proposal, can run the same setup. One founder posted on X that he was using a similar agent, built on GPT-5.6, to handle quoting, scheduling, insurance, and compliance for an entire company he was launching. Take that as one person's early account, not a proven system, but it points at where this is heading.

Isn't This Just a Fancy Proposal Template?

No, and the difference is the part that matters.

A template is static. You fill in blanks, and it has no idea whether what you typed makes sense against your margins or your past deals.

An agent reads the actual conversation. It notices the prospect has 25 users, not 15. It notices the compliance requirement mentioned once, at the seven-minute mark, that changes which package applies. It prices against your real rate card and explains why it chose that tier, the way a sharp account manager would, not the way a mail-merge would.

The output still needs your eyes on it before it goes out. This is a draft that is 90 percent done in two minutes, not a system that replaces your judgment on price.

How Do I Automate Quotes for My Small Business With AI?

There are two honest paths here, depending on how comfortable you are with a folder full of text files.

Path one: the agent-folder method, no CRM required.

1. Create one folder per client project, or one shared folder if you quote similar work repeatedly.
2. Write a short instructions file that states the agent's one job: turn a call transcript into a client-ready quote in your voice.
3. Add a company file: your pricing logic, margins, and any rules about discounts or minimums.
4. Add a services file: your packages, what's included, and what costs extra.
5. Drop two or three of your best past quotes into an examples folder, so tone and formatting carry over automatically.
6. After each sales call, save the transcript, or a recording's auto-transcript, into a calls folder.
7. Ask the agent, in one sentence, to build the quote from the last call.
8. Read it, adjust the number if it needs adjusting, and send it.

Path two: the no-code stack. If you would rather not touch a folder of text files, the same idea runs through connected tools instead of a coding assistant: a form or a shared inbox captures the request, a workflow tool like Make.com or Relay.app routes it to an AI model, the model matches it against your price list, and a template in Canva or your CRM turns it into a branded document. Quote-specific platforms such as DealHub and Jinba package this whole flow for teams already running on HubSpot or Salesforce. Either path lands you in the same place: a business that remembers its own pricing so a person does not have to, every single time.

Three Ways to Quote, Compared

What This Looks Like at Marisol's Agency

Marisol takes a call from a local bakery chain wanting social media management and a website refresh. She used to block ninety minutes the next morning to build the proposal.

Instead, the call recording's transcript lands in her calls folder the moment the meeting ends. Marisol types one line: build the quote from today's call with the bakery.

The agent reads the transcript. It sees three locations, a request for weekly content plus one paid campaign a month, and a mention that the client's last agency was slow to reply. It checks her services file, prices the retainer tier that matches three locations, adds the one-time website line item from her rate card, and writes a cover note promising a same-day response window, because that objection came up on the call.

Two minutes later, Marisol has a draft that already sounds like her. She tightens one sentence, checks the total against her margin floor, and sends it before lunch.

What You Have After This

A week in, you have one working example: a real quote that came out of the agent instead of out of your memory.

A month in, you have a folder that knows your business better than a new hire would on day one, and every quote reads like the same person wrote it, because in a sense, one system did.

Six months in, quoting stops being the thing you dread after every good call. It becomes the fastest step in your sales process instead of the slowest one.

Same pricing. Same services. A completely different quoting machine.