How I found a client in a chat and sold a system for $1 900: the whole path, no backstage

How I found a client in a chat and sold a system for $1 900: the whole path, no backstage

One deal in full, with nothing skipped and no backstage: how the client turned up in a job feed, what was in the messages, how the fifteen-minute call went, what its review showed, and what the proposal was assembled from — the one that got paid in full.

One deal in full, with nothing skipped and no backstage: how the client turned up in a job feed, what was in the messages, how the fifteen-minute call went, what its review showed, and what the proposal was assembled from — the one that got paid in full.

---

The gap between "doing" and "monetising"

That is how one person in my survey described in four words what is happening to most of the market. Sixty-two responses, August 2026: 45.2% use AI regularly, 91.9% have earned nothing or earned once, and for nobody has AI become their main income.

The skill is there. The money is not.

The reason is not that people are bad at it. The reason is that the path from "I can do this" to "I got paid" consists of seven or eight steps, and not one of them is about AI. Where to find a client. What to write to them. What to say to "how much". How to run a call. What to send afterwards. Each step separately gets explained by plenty of people. The whole thing, from first message to quote, almost by nobody.

So I walked the path publicly and recorded all of it. Below is the whole route, on one real deal, with the good bits not edited together.

An honest frame first. This is one deal, not market statistics. The client — Zakhar Fomin, owner of a dental clinic — gave permission to name himself and the figures. The sum is the price of a specific project for a specific clinic, not what the market pays. The median for one-off work in my scrape of 848 listings is $30–75, and the two must not be mixed. What is shown here is a method, not a promise of income.

Step 1. Find where the work actually is

Work does not sit in one place. Marketplaces demand a positioning package a beginner does not have. Job boards are for people who can already state their profession in one line. Your immediate circle is finite.

That leaves Telegram, where the flow is continuous — and where you physically cannot sit in a thousand chats tracking everything by hand.

So we built a service for it: chats get checked daily by a scraper, what it collects is filtered by a model and sorted into categories — marketing and social, design, photo and video, automation. Each card shows the source, the task, the type of engagement and the client's contact.

From there I take the 5–10 most recent listings in the category I want and do not choose between them. I will choose later, from whoever replies.

Step 2. Work out who wrote the listing

I copy the listing, give it to the sales assistant: analyse the client type.

The client is looking for an automation specialist: a bot for standard questions and bookings, a website, content, a price list. The requirements mention proactivity, self-sufficiency, finding solutions without constant supervision, the ability to sell.

The verdict: a Controller with a streak of Driver. The main pain, drawn from those same phrasings, is the fear of pouring budget into a theorist who will need chasing and teaching basic processes. The motivations: independence, order, saving time, a measurable result.

None of that is written in the listing directly. But that is what he is buying.

Step 3. Write so they reply

The same request produces a cover letter for that type — in my name, allowing for the fact that I have no relevant case studies in the niche.

It works on one turn: the absence of a classic portfolio is named first and immediately turned into an argument. Not "I'm a beginner but I try hard" but a specific property people with portfolios do not have.

The whole cycle is 5–10 minutes a day for a dozen letters.

Step 4. Do not fall apart on the price question

Zakhar replies and immediately asks the cost.

This is where most people freeze. Name a figure off the top of your head and you lose. Dodge the answer and you lose harder, particularly with a Controller.

The answer rests on three mechanics. Justifying an individual calculation through concrete parameters — which CRM, how many chairs and specialisms, how deeply to configure the assistant. A value anchor: a full build pays for itself in the first month because it replaces the functions of a front-desk administrator while working with no days off, round the clock. And a choice of two rather than an open question: not "when suits you" but "7:30 or 8:15".

No figure named. A call booked.

Step 5. Run the call to a plan

I do not compose the fifteen-minute plan — I ask the assistant to draw one up for that type and keep it open in the next window.

The opening: "We'll keep this to 15–20 minutes and stick to the substance. Three or four questions about the situation, then I'll show you a live example, and at the end we'll discuss the format and a price range." The Controller sees structure, considers the process safe and hands over the lead in the conversation himself.

Diagnosis: open questions about losses. How enquiries get handled at night. How many people leave because nobody answered within five minutes. What is wrong with the front desk. I listen and write markers of pain into a separate document.

A demonstration through the restaurant metaphor: the dining room is the chat interface, the kitchen the AI logic, the store cupboard the knowledge base and CRM. A patient writes at night asking the price of an implant, the live receptionist is asleep, the agent goes to the cupboard, qualifies them, addresses the fear and books a slot.

The economic argument, and closing on a technical funnel audit as the next step.

Step 6. Review your own call

I record the call and upload the audio to the assistant.

It finds two mistakes of mine I had not noticed myself: mid-call I drifted into technical questions and blurred the focus on business metrics; and I did not press the metaphor home when Zakhar raised his fear of "stupid bots" — I should have said right then "this is not a replacement for a doctor, it is a smart concierge".

But something else is more valuable — what it pulled out of the client's words. The fear of losing an expensive patient to a bot. Manual reactivation of the base: once a year he goes through the calendar looking for old patients, or pays a remote administrator. Dead social media — one patient every two months.

And a warning: name a price with no link to payback and the Controller will switch into Analyser mode and take time to think.

The three pains became three blocks of the quote. The warning determined how the figure was presented in it.

Step 7. Assemble the quote in ten minutes

The assistant writes a brief for the deck: the style, the colours, the fonts, the number of slides, which phrases to avoid. Then checks itself in focus-group mode — playing Zakhar and walking through the slides, marking where he would tense up.

Then a ready skill in Claude Code — a chain of agents where a copywriter, a designer, a structurer and a proofreader hand work to each other — assembles an editable deck in 5–7 minutes.

Inside: a trust slide, a losses slide in red, three work modules, an owner's dashboard with metrics, the economics against an in-house team ($720 a month for three people against $1 900 a year for the system), three sprints with staged payment, and a final slide with the CRM as a considered second step at $3 000–3 600 — the thing not being sold now.

Step 8. Payment

The proposal went out as a PDF. The payment arrived in full — $1 900 for a year's system.

Separately about what stayed outside the quote: the second step in the proposal was a CRM at $3 000–3 600, presented as "a considered second step rather than overpaying at the start". That is the point of that final slide — not to try to sell everything at once but to show where somebody goes after they have paid for themselves.

And what the client ended up holding: 11 sections, 157 pages — from audience research and positioning to the content system, the funnel and the sales dialogue scripts.

How long this took

Worth looking at the route in hours rather than steps.

Finding jobs and ten letters — 5–10 minutes. Once a day, not once ever.

Answering the price question — a couple of minutes: screenshots to the assistant, finished text back.

Preparing for the call — about ten minutes for the plan, plus a run in simulation if you are nervous.

The call itself — fifteen minutes to the ground rules.

Reviewing the call — the time to upload the audio and read the review.

The quote and the deck — under ten minutes.

Active work across the whole chain from first application to payment is about an hour, spread over several days. The time is eaten not by the actions but by waiting for replies.

That matters, because the main objection to this approach is "I don't have time for it". In my survey 61.3% are willing to put in up to five hours a week. The route above fits inside that budget with room to spare.

What matters more than the sum

Not one step is about AI. Find, read the person, write, answer on price, run a conversation, review it, assemble a document. The tool is the same throughout — an assistant — but the task is human every time.

Each step eats the previous one's result. The listing gives the type. The type gives the letter. The messaging gives the call. The call gives the pains. The pains give the quote. Remove any link and you break the chain rather than one step.

And the most uncomfortable part. All of this can be done without AI, by hand — just slower and worse. AI here does not replace the skill of selling, it lends it to you until your own is built.

Every step covered separately

This article is the route. Under each step there is a detailed breakdown: where to find work, how to read the type from a listing, the letter with no portfolio, the answer to the price question, the call script, reviewing a call with AI, assembling a commercial proposal.

What you need to walk the route yourself. The job feed for the search. More than 140 AI assistants across 14 categories for everything else, from reading the type to doing the job itself. The ten-lesson sales module — objections, rejections, the four client types, shaping an offer around demand.

And "Freelance 2.0" — a library of 62 prompts across 11 funnel stages: research, positioning, content, lead magnets, websites, funnels, sales with scripts and simulators. The system everything you read above gets assembled with.

Registration is free and opens three days of full Basic access — the sales module, all the assistants, the prompt texts, the skill files and six foundational express courses.

Start with the first step rather than the last. Not with learning Claude Code and not with choosing a model. Open a job feed, take the ten most recent listings in your category and send ten letters. Everything described after that only begins once somebody has replied.

If you know somebody who can do the work and has never once issued an invoice for it, forward this to them. The whole path is here.

---

What exactly was delivered

So that "a system for $1 900" does not sound abstract — here is the composition of the work. Eleven sections, 157 pages.

1. Research — five documents: a segment map, an awareness-stage map, a competitor analysis across 1 850 posts from ten channels, a personality unpacking, a map of meanings and subjects.

2. Positioning — a core and three supports, the voice and the frame, a profile bio, a visual system, seven highlights.

3. The content plan — 18 pillars, 29 subjects across nine weeks, a subject bank and a permanent layer of mechanics.

4. Lead magnets — nine materials by code word, distributed across segments.

5–7. Three platforms by function — 16 carousels and 27 talking-head scripts for the cold top, 27 posts for the warm hub, five playlists and a flagship video script for search.

8. Monetisation — a three-tier product line and three baskets in the existing base with different speeds of money.

9. Websites — a general router landing page and three segment pages, with a running frame of honesty.

10. The funnel — an end-to-end scheme of two tracks and a bot specification with a data model and assembly logic.

11. Sales — four scripts, an objection bank, four reference cases and two AI simulators.

What follows for you. The difference between $200 work and $1 900 work is not the quality of the copy. It is that in the second case the client gets a system they can see whole and that keeps working without you.

---

The series: one deal step by step

Eight pieces about one real deal — from a listing in a chat to the payment received. The client, Zakhar Fomin, owner of a dental clinic, gave permission to use his name and the figures.

---

What to read next

[Where to find AI work](/en/blog/where-to-find-ai-orders) — step one in detail.

[Reading the client type from the job posting](/en/blog/client-type-from-posting) — step two in detail.

[The fifteen-minute client call](/en/blog/fifteen-minute-sales-call) — step five in detail.

[A commercial proposal in ten minutes](/en/blog/proposal-in-ten-minutes) — step seven in detail.

[AI API pricing in 2026](/en/blog/llm-api-pricing) — what a request costs and how to work out your own costs.