A bot that segments somebody on the fly and delivers what they need
A menu of four buttons makes somebody guess which option they belong to — and half of them leave at that step. Below is the mechanics of a bot that reads a request from the first sentence, delivers the right material and hands a hot enquiry to a live person.
A menu of four buttons makes somebody guess which option they belong to — and half of them leave at that step. Below is the mechanics of a bot that reads a request from the first sentence, delivers the right material and hands a hot enquiry to a live person.
---
A menu of six buttons with no option for you
You have written to a bot like that. "Hello! Choose: 1 — learn about our products, 2 — book a consultation, 3 — ask a question." You press three, write your question — and get "our manager will be in touch."
That is not a funnel. It is an answering machine with extra steps.
The problem is not even the inconvenience. The problem is that a bot like that learns nothing about the person. A beginner who does not understand the basics and a business owner with a budget both see the same menu and get the same answer. It overwhelms the first and insults the second.
And somebody who wrote in their own words falls outside the logic entirely: there is no script for free text.
Meanwhile, in that very first sentence a person usually says everything about themselves that you need.
What a bot has to determine before offering anything
Three things, in order.
First: who is in front of us. Which segment the person belongs to — by intent rather than demographics. A beginner who wants to start earning. A specialist who needs it for their profession. An owner building a system. That determines what to talk to them about at all.
Second: how ready they are. Which awareness stage they are on: only looking around, searching for a solution, choosing whom to buy from, ready to buy. That determines what to offer at the end. Somebody at an early stage must not be offered a purchase; somebody at a late stage must not be offered another free material.
Third: what is in the way right now. The specific objection: too expensive, did not know this existed, did not understand what I get, afraid I will not manage, no computer, access problems.
The material gets delivered at the intersection of the three answers. Not "here is our catalogue" but one specific material that removes that objection for that person at that stage.
How it works this out
Not with questions. A five-question form kills half the people by the second one.
By marker phrases. Every segment has its own language and it is consistent.
Somebody who can do the work but does not sell writes: *"I don't know where to find work"*, *"no case studies yet"*, *"I'm afraid I'll take the job and fail"*.
A specialist bringing the tool into their profession: *"I did a contract review in ten minutes"*, *"they don't know about it at work"*, *"I need it for my profession"*.
Somebody building systems: *"roll it out across the company"*, *"assistants for a department"*, *"build a system in my own business"*.
These are not guesses — they are phrasings collected from the audience's real answers during the research. Which is why a bot cannot work without prior segmentation: the marker phrases come from it.
The stage is determined by behavioural signs: what somebody has already tried, what they lack, what they ask about. "Tried it but without a system" and "I use it every day and have earned nothing" are different stages, and they are audible immediately.
The rule without which a bot breaks
One rule, and it matters more than all the mechanics.
If somebody does not resolve into any segment, the bot does not guess. It asks one clarifying question and only then delivers.
Delivering the wrong thing is worse than delivering nothing. A business owner sent material for beginners leaves and does not come back. A beginner sent an analysis of a corporate rollout feels they have come to the wrong place.
One question is fine. Three questions in a row is already a form, and half the people are lost on it.
What a bot gives that an enquiry form does not
A fair objection: why all this when you could put up a form and call people back.
It works at night. An enquiry outside working hours either gets handled immediately or gets lost — the person finds the next provider within five minutes. That is the most understandable and most underrated line of losses.
It qualifies before the conversation. By the time a live person joins, it is already known who is on the line and what they need. The conversation does not start with "tell me what you are interested in".
It does not tire and does not have moods. The twentieth enquiry of the day gets handled like the first.
And it leaves data. Every conversation is raw material: which phrases recur, where people drop off, which objection comes up most often. An enquiry form gives you nothing but a name and a phone number.
What the bot's base is assembled from
A bot is not a dialogue script. It is a knowledge base plus delivery rules.
What is in the base:
- Segments with their marker phrases — from the audience research
- Readiness stages with their signs — from the same place
- A bank of objections verbatim, as people phrase them
- A library of materials — one for each intersection
- Next-step rules: what to offer at each stage and what not to
- The voice — from the positioning, so the bot speaks like you rather than like technical support
Note where it all comes from. Not one line of the base gets invented at the bot-building stage — it all arrives from the earlier stages. Which is exactly why a bot built first comes out empty: it has nothing to know.
What the bot delivers at each stage
The matrix becomes clearer with an example. The same person at different stages gets different things.
Only looking around. A small everyday result, one tool, something simple and usable today. Nothing about money and jobs — too early.
Tried without a system. A route, a choice of direction, a case with figures. Not a catalogue of hundreds of options — at this stage that makes things worse.
Can do it but does not sell. A price list, a place to find work, a script, a portfolio. Here you must not show "what a neural network can do" — the person already knows.
Choosing who to learn from. The composition of the product, what specifically they get, plans by task. A discount works worse than clarity here.
Ready to pay. Terms and the first three actions. Another free material at this stage is a way to lose the person.
What the business ends up with
- A funnel diagram: which branches exist and where each leads
- A bot script with the logic for determining segment and stage
- A bank of marker phrases — which words identify whom
- A delivery matrix: segment × stage × objection → a specific material
- Escalation rules: when the bot hands over to a live salesperson
The last is often forgotten, and it is the most lucrative part: a bot has to know when to stop being a bot. The sign of a hot client is a question about price, timescales or specific terms. At that point the conversation goes to a human.
By the methodology I use, the funnel and sales stage takes three days — with the research, positioning and content system already done.
That is the sixth of seven stages of a full positioning package: 3–4 weeks, 25–35 hours, around $1 900. Honestly: that is the price of the service by our methodology, not market statistics.
Do it yourself or commission it
Do it yourself. The mechanics above are the "Funnels" and "Sales" blocks from "Freelance 2.0": 62 prompts across 11 stages, including the funnel structure, the bot script, call and message scripts, and simulators where AI plays a realistic client.
The course costs $9 and is included in Basic. Registration is free and opens three days of full Basic access — plus 140+ AI assistants across 14 categories, the ten-lesson sales module and the automation templates for the technical build.
Commission it. The service exists. And now you know the main thing about it: a bot built in isolation from the audience research always comes out as a menu of six buttons. The order matters more than the execution here.
Do one thing today: open the last twenty incoming messages to your business and sort them by three questions — who is writing, how ready they are, what is in the way. Usually it turns out there are not twenty variants but three or four, and they repeat. That is your delivery matrix.
---
How this looks on a real project
What follows is the same work on a live client. Fomin Studio, "the Fomin Method": two doctors, Zakhar and Alexandra, husband and wife. The work was delivered as eleven sections, the tenth being the funnel and the bot as its hub.
Why this example is here. It shows that a positioning package is not an abstraction but a service with a clear composition. And it sells to any business that has an audience and a product: a clinic, a school, an agency, a shop. The mechanics are the same; the contents change.
An analysis of that deal from the first message to the payment is included in the sales module on the Basic plan.
Two tracks, one hub
The funnel runs in two modes in parallel, and both converge in one place.
The cold track: content → bot → booking. Medium-term, dominant after 6–12 months. It provides scale.
The warm track: the existing base → bot → booking. Fast, money in the first month or two. It provides speed.
The bot is the central hub of both.
Five stages of the cold track
Traffic across five channels and 29 subjects → nine lead magnets by code → warming by awareness stage → the converter with the location split → closing without pressure.
The key principle: warming lasts as long as the awareness stage requires. The bot does not sell until it has understood who is writing.
Each channel for its own stage
Short-form video — stage 1→2, virality on fear and aspiration · YouTube — 2→3, search and depth · the Telegram hub — 3→5, warming and conversion · the Instagram feed — 4→5, cases, testimonials, the offer · a local community channel.
How the bot itself works
Four steps in its logic:
Entry — a code word or a free-form message.
Diagnosis along three axes — segment, awareness stage, location.
Value and material — the contact is saved to the base.
The offer — only after qualification.
It hands out value, qualifies and remembers the person. It does not sell until it has understood who arrived — and that is not politeness, it is what holds the conversion rate up.
---
What to read next
[Lead magnets by code word](/en/blog/lead-magnets-code-words) — what the bot hands over at the entrance.
[Anatomy of a system prompt](/en/blog/system-prompt-anatomy) — how this is built inside.
[AI for developers](/en/blog/ai-for-developer) — a real system prompt taken apart.
[Four scripts and two simulators](/en/blog/sales-scripts-and-simulators) — what happens after the handover.
[AI agents: what they are and what people pay](/en/blog/ai-agents-guide) — the full guide to agents, with market rates.