How much does a chatbot cost: what goes into the price

The short answer

An honest price for a chatbot is given after the task has been reviewed. It comes from the amount of work: channels, CRM integration, material for the knowledge base, how complex the scenarios are, voice and photos, support. We first review your task free of charge, then send a quote for it.

A «from» figure named before any conversation describes someone else's project. Very different things are sold under the word «chatbot»: a button bot with ready-made replies in one messenger, and an AI agent that runs the deal in the CRM across several channels at once. The amount of work differs many times over, so comparing their prices makes no sense.

What the price depends on

On six factors. Each one adds its own share of work to the project, and the same factors are a handy way to compare quotes from different contractors.

Material is usually what moves a quote most. If the price list and terms are already written down, they only need checking and laying out for the bot. If they live in managers' heads, they have to be found out, written down and agreed, and that work goes into the one-off part.

Integrations come next. A notification to a manager in a messenger is quick to set up. Two-way exchange with the CRM, telephony, warehouse and payments is a separate job, and every system beyond the standard set adds both setup and support.

What is paid once and what every month

The one-off part pays for preparation and launch, the monthly part for the bot's work and upkeep. Ask for a quote where these parts are separated: a one-off sum without monthly lines looks more attractive than it turns out to be.

A scripted bot has no «model requests» line: it replies with ready-made phrases. An AI agent always has one. Check with the contractor whether this cost is included in the monthly payment or billed separately as used.

How we work out the price

First a free review of the task, then a quote for it. We do not name a price before the review: until it is clear how many channels you have and what happens to an enquiry after the reply, any sum would be guesswork.

1. Review. We look at where enquiries come from, who answers them and how, what happens to the deal next and where clients are lost.
2. Scope of work. Using the factors from the table above, we define what needs doing: channels, integrations, material, scenarios, voice and photos, support.
3. Quote. We show the one-off and monthly parts separately, with the scope of work for each line.
4. Decision. If the review shows the project will not pay off, we say so straight away. Sometimes it is enough to set up the CRM or change the order in which enquiries are answered.

What makes up the price of an AI agent for sales

A sales robot is at least two different jobs: the agent itself and the CRM it works in. On top of that come channels, conversation volume and, if needed, quality control and rebuilding the sales department. That is why one figure can’t honestly be named without an assessment.

An agent is not always cheaper than a person. It pays off where a task repeats at scale: answering a typical question, qualifying a request, booking a meeting at any hour. In the live GetGate project 27% of requests arrive in the evening and at weekends, and keeping a night shift for them is expensive. On haggling, large contracts and complaints the agent saves little: that is a person’s job.

How to work out whether it pays off

Calculate from your own losses. Take a month's worth of requests, estimate what share goes unanswered for more than half an hour, and multiply by your sales conversion rate and average deal size.

The order of the calculation on a worked example:

Even if half of those sixty «lost» enquiries are recovered, that is 30 enquiries, 6 sales and 240 thousand roubles of revenue a month. Compare that sum with the cost of implementation in the quote: the one-off part plus the monthly lines over the same period.

The opposite case is no rarer. Twenty enquiries a month, a salesperson answering within three minutes, almost no losses — here a bot will never pay for itself, and an honest contractor will say so.

200

requests per month

30% wait over half an hour

60

were left without a quick reply

won back half

30

can be brought back into work

20% conversion

6

sales per month

× average deal 40,000 ₽240 000 ₽revenue per month currently being lost

Plug in your numbers: requests, share without a quick reply, conversion and average deal.

The costs people forget

The setup invoice is not the whole sum. Four items of expenditure surface after launch and spoil the economics of the project.

• Refining after launch. For the first weeks the answers are corrected against real conversations. If that isn't in the price, the bot stays as it was assembled blind.
• Preparing materials on your side. One of your people will spend several days collecting price lists, terms and answers to frequent questions. That is money too.
• Licences for adjacent services. The CRM, telephony, paid access to platform APIs — paid separately and directly.
• Changes in the channels. Messengers and platforms change their rules. Every few months something will need fixing.

Where you can save and where you can't

It is worth saving on coverage — start with one channel and one scenario. You cannot save on CRM integration and on refinement after launch: without them the project doesn't pay off at all.

You can

• Launch in the one channel with the most enquiries and add the rest later.
• Start with a scripted bot if the requests are of the same kind, and move to an AI agent, once you hit the ceiling.
• Collect the materials yourself — that is a noticeable part of the bill.

You can't

• Skip writing to the CRM. Without it the correspondence stays in the messenger and there is nothing to measure the result with — so it stays unclear whether the project paid off.
• Skip support in the first months. A bot without corrections from real conversations is a bot assembled from guesses.
• Take on a contractor with no answer to «who owns the settings». The saving turns into leaving costing more than starting again.

What to ask a contractor to compare quotes

Quotes from different contractors can only be compared if they cover the same scope of work. Ask each of them the same questions and put the answers into one table.

• What is in the one-off part, item by item?
• Which lines are monthly and what makes them grow?
• Are model requests included in the payment or charged separately?
• Which services and licences do we pay for ourselves?
• Who collects the material for the bot, and how much of our staff's time will it take?
• What does post-launch refinement include, and how long does it last?
• How does the bot hand a client over to a manager, and what goes into the CRM?
• Who owns the material, scenarios and settings once the work is finished?

Be wary if the price is named before any talk about your channels and process, if it cannot be broken down into lines, or if support is not discussed. The promise «our AI will learn everything by itself» is also a reason to ask again: without your material it will not.

The good sign is the opposite: the contractor works out your return on investment and, in some cases, tells you that you do not need a bot yet.