How do I automate the whole sales process after a call with AI?

After a sales call, I tell my agent to start the automation. That kicks it off. The proposal, the payment link and the onboarding get built without me, and I review them before they go out.

Why automate the follow-up at all?

Because the follow-up is what eats the time. I hit this on an AI training I was selling to a company. I got off the call, he said send me everything, and then I had to go write the proposal, go into Stripe, add him as a customer and create the price.

When an agent does all of that, you hang up and it all comes to you. You review it, hit enter, and it sends everything off. You can do more sales calls because the follow-up doesn't take as much time.

What does the automation do, step by step?

1. Reads the call transcript from my call recorder (I use Fathom).
2. Builds a proposal website specific to that client.
3. Creates a Stripe payment link at the price we discussed on the call.
4. Waits for payment, then sends an onboarding email and an intake form.
5. Turns the intake into deliverables: a styled HTML company overview and a markdown context file the client's employees load into their own AI chats.

The request comes through Slack, and the agent does the rest.

Why does Slack matter so much here?

It's the come-to-you. The value is the agent showing up and saying, I did this for you, go review it. You aren't opening five tools to check on progress. More on that in running AI agents in Slack.

What about the deals you lose?

This is where I think the biggest upside is. My guess is over 95% of businesses either do nothing with lost deals or do the most basic, worthless thing: add them to a list that gets a generic email six months later.

"To do what you're saying can change a business's trajectory from that one agent, especially if it's high ticket or they have lots of potential leads."

Shane Brady of AI for Media Brands walked me through his version on my podcast. His sales agent summarizes the call within minutes, drafts the proposal, creates the Stripe link after he approves, and nudges lost leads using the pain points from their own call. He said it booked two clients while he was at lunch. It also sends him pre-call scripts built from won and lost deals, and he put the cost at about $3 a week to run. Those are his results.

How was mine built?

Claude's dynamic workflows, which spin up parallel sub-agents for each section of a task, made it faster and less error-prone than a single agent handling the whole build. See how multi-agent workflows work.

What else can an agent pull from calls?

Action items. I told my assistant agent to review a call with a teammate, find everything we said that isn't worth my time, and do it. We'd already documented it on the call.

The main blocker is access. Logins and permissions. More in how much access to give an agent.

One idea I saw on Twitter goes further: an agent listens to the live transcript and builds a demo or prototype of the prospect's idea while the call is still happening. Then you say, here's what my system already built for you, is this what you're thinking? How are you not going to hire that person?

Can AI handle outbound prospecting, too?

A peer showed me a version for direct mail. He gave a desktop Claude agent an ideal-customer profile and a Google API key. It pulled every pool-service business in Tampa with phone numbers, website presence and Google review data, then connected the segmented list to a mail API to send postcards. He put the cost at roughly $300 total and says it works for any home-service trade. That's his build and his number.