What an AI assistant for a department costs and what makes up the price
A client asks the price and there is nothing to justify it with: it is unclear what it is made of. Below is the composition of the work, in which the building is the smaller part, six parameters determining the sum, and what that sum gets compared against so it stops looking large.
A client asks the price and there is nothing to justify it with: it is unclear what it is made of. Below is the composition of the work, in which the building is the smaller part, six parameters determining the sum, and what that sum gets compared against so it stops looking large.
---
"So how much will this cost?"
The question most freelancers freeze on the first time they sell an assistant to a company.
There is nothing to name a figure from: it is unclear what it is made of. Your own time is easy to count — but the client is not buying your hours.
The client does not understand either. They have seen free chats and cannot see what they would be paying for when "it's just AI".
Let us go through the composition: what the work involves, what determines the price, and what to compare that sum against.
What the work consists of
Six parts. The first two are not technical, and they are the longest.
1. Analysing the process
What sort of department, which tasks repeat, where people waste time, which questions they ask each other ten times a day.
That is an interview rather than technical work. And without it the assistant comes out about what seemed important to you rather than about what hurts.
The typical finding at that step: half of what was requested is not needed, and what is needed nobody mentioned.
2. Gathering the knowledge
The most labour-intensive part of the whole project.
The knowledge sits in people's heads, in scattered files, in message threads. It has to be extracted, consolidated and structured: services, rules, procedures, frequent questions in real phrasings, handling of exceptions.
An assistant with no base answers in generalities and irritates people more than silence. That is the main cause of failures and it has nothing to do with the model.
It gets extracted by interview: the person does not have to write anything, they have to answer questions — what do you do, in what order, what goes wrong.
3. The instruction
Role, base, reasoning algorithm, output format, handover triggers, restrictions.
The last block is the reason assistants invent things that do not exist. A model fills gaps with something plausible unless told not to.
4. Debugging on live conversations
The first version is always full of holes. Only a run on real questions works: where it did not understand, where it invented, where it did not hand over in time.
That is what separates a working assistant from one built in an evening. And what cannot be skipped, however much the client hurries you.
5. Integration
Where it sits: a chat on a website, a messenger, an internal system. How it hands a conversation to a human and what it passes over — a summary, a contact, the history.
6. Handover and training
Instructions for staff, short training, rules for updating the base.
Skipped most often — and that is exactly why systems die within a month. An employee who did not understand the system routes around it inside a week.
What determines the price
Six parameters. Those are what the sum gets justified with in conversation with a client.
The size of the knowledge base. One price list or five business areas with exceptions is a difference of several times. That is the main multiplier.
The number of scenarios. Answering questions is one thing. Answering, qualifying and booking is another.
Integrations. A standalone chat or a link to an existing system of record.
Whether debugging on live traffic is needed or a basic check will do.
Support. A one-off delivery or monthly maintenance with the base updated.
The number of staff to be trained.
A practical technique in conversation: do not name a figure straight away. Name three parameters it depends on, tie the cost to the client's spending, and propose a short call with two specific slots. A figure with no justification turns the conversation into haggling.
What to compare the sum against
Here is the main argument, and it is not about technology.
The comparison runs not against zero but against an existing cost line.
A real example from practice. The client is Zakhar Fomin, owner of a dental clinic. The task: a receptionist answering round the clock, qualifying enquiries and booking appointments.
Sold for $1 900 a year, paid in full. As a second step, a CRM at $3 000–3 600.
Why he agreed: the alternative is an in-house team — a marketer plus a social media specialist plus a remote administrator — at around $720 a month plus taxes, holidays, sick leave and the human factor. The system works round the clock and does not fall ill.
The phrasing that works: "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".
The anchor gets set before the figure, not after. When the sum gets said, it lands in a frame already prepared.
The market price range
From my scrape of 848 job listings over 24 days: a system or pipeline is $490–1 460, against a $30–75 median for one-off work.
A caveat: an explicit budget appears in 6% of listings, with between two and fifteen observations per category. The median is reliable; the breakdown is a guide.
And the $1 900 in the example is the price of one specific project by our methodology rather than market statistics. The two numbers must not be mixed.
What separates the lower bound from the upper: not the difficulty of building but the size of the knowledge base and whether there is support. An assistant on one price list with no maintenance and a system with five business areas and monthly updating are different products.
What gets delivered to the client
A list worth showing in a proposal in full — it justifies the price better than any words:
- An analysis of the process with a map of where the department loses time
- The knowledge base in structured form
- A system instruction with the algorithm and the restrictions
- Rules for handing a hot enquiry to a human and the format of the summary
- Integration into the working environment
- Instructions for staff and training
- A procedure for updating the base
A client seeing seven lines instead of one sum argues with the items rather than the total. That is already a normal working conversation.
Where the main money is
Not in the first sale.
In support. Prices change, services change, new questions appear. Without updating, the base ages within six months and the assistant starts lying to customers. That is a recurring payment and it is an honest one.
In expansion. An assistant for one department works — a request arrives for a second. Then for an integration, then for analytics.
In my dataset 21.9% of clients are explicitly looking for somebody long-term and 20.3% mention a flow. Almost half the market does not fit into one-off work.
What is available on the platform
More than 140 ready assistants across 14 categories — before building your own, look at the ready ones. A substantial share of tasks is covered, and sometimes a client needs a curated set rather than development. That is a service too, only cheaper and faster.
The bot architect from the engineering category — writes system instructions with an algorithm, handover triggers and restrictions. Access from the Basic plan.
Building your own assistants for a company, with its knowledge base — the Creator+ plan.
The "personal AI bureau" express course — designing instructions and selling assistants to clients. $9, the Creator plan. Not in the three free days of Basic.
The ten-lesson sales module — four client types, handling price objections, shaping an offer around demand. Included in Basic.
Registration is free and opens three days of full Basic access.
Where to start today
One action. Take a company you know — your own or a client's — and count one figure: how many enquiries reach them outside working hours and what happens to those enquiries.
That figure is your quote. Not "what does an assistant cost" but "what does not having one cost".
That is where a conversation starts in which price stops being the main question.
---
What to compare your price against
A client asks why so much. The answer is easier with the prices of ready solutions of the same class beside you.
| Solution | Market price |
|---|---|
| Ad campaign analyst | from $240 |
| 24/7 AI manager with CRM integration | $480 |
| Phone conversation analyst | $600 |
| Quote assistant | from $600 |
| Messenger conversation analyst | $360–1 200 |
| AI integration into a CRM | from $1 200 |
| Sales department monitoring system | from $9 600 |
What that gives you in conversation. You are naming not your own price out of thin air but showing what solutions of this class cost. From there the discussion is about scope rather than the sum.
And an honest fork worth raising with a client. Some of those solutions exist ready-made: they open configured, you enter credentials and they run. If the task is standard, they do not need development from scratch, and saying so plainly is more profitable than selling them something extra. A bespoke build is for where the processes are non-standard.
What the work consists of
Five lessons of module 9 show the full cycle: system instructions, copywriting, conversation design, marketing and building a sales bot that walks a client through a staged funnel.
The key part is not the building but the knowledge base. Almost anybody can assemble a bot; extracting from a company what only two people know and reworking it for retrieval is what people pay for.
Module 9 — Creator; the ready solutions — Make, N8N and Full.
---
What to read next
[A company knowledge base as a service](/en/blog/company-knowledge-base) — the most labour-intensive part of the project.
[Anatomy of a system prompt](/en/blog/system-prompt-anatomy) — the technical part of the work.
[AI agents: how they work and what they cost](/en/blog/ai-agents-guide) — a real $1 900 deal.
[An itemised quote instead of buttons](/en/blog/quote-not-buttons) — how to break a price into items.