What makes up the price of an AI agent for sales

Why a single figure cannot be named

The first question on a call sounds simple: how much does your AI agent cost. The honest answer is “it depends”, and that is not evasion but fact. A sales robot is not sold as a boxed licence with a fixed price list, because the price is assembled from several independent parts: the CRM connection, the number of channels, the volume of calls and whether the sales department has to be built from scratch.

A business owner usually asks not “how much does the bot cost” but “where do those prices come from” — that is, what the price is made of at all. It is the right question: it shows the person already understands that the solution has several parts. Further on we take each part separately, so that you can work out on your own sales department exactly what you are paying for.

We deliberately do not give a range in roubles: it depends on the niche, the number of salespeople and what is already set up in the client's CRM. But the structure of the costs can be shown honestly — it is the same for any business connecting a robot to a working sales department.

The robot licence and the amoCRM integration are different cost items

The first mistake when estimating the price is to think a sales robot is one service. In fact it is at least two different jobs. The first is the agent itself: the dialogue logic, qualifying the client, booking meetings, recognising voice messages and photos, remembering the conversation history. The second is CRM implementation: pipelines built for the client's process, connecting channels and telephony, integrations with payments and documents.

If a business already has amoCRM set up with pipelines for its process, there is less work — all that remains is to connect and train the agent. If there is no CRM at all, or it was put together hastily, you first have to build the pipelines, channels and telephony and only then place a robot on top of them. These are two tasks of very different size, and they are priced separately.

Connecting the channels is a separate item: Telegram, WhatsApp, Instagram, Avito, the website — each channel needs its own setup inside the CRM. The more channels a business has, the more work at the start, but also the fewer enquiries are lost afterwards. On one live project, GetGate, it was exactly the evening and weekend enquiries — 27% of the total — that used to simply wait until morning, until the channels were brought into a single pipeline.

How the number of salespeople, leads and calls affects the price

The second variable is the size of the sales department. Training a robot for a niche with three salespeople and a small flow of leads is one task. Training it for a department with dozens of salespeople, several pipelines and hundreds of calls a day is another, because the volume of conversations the agent has to handle without losing quality goes up.

The amount of analytics work is counted here too: sales analytics builds plan against actual in real time and breaks it down by channel, service and salesperson. The more product groups and business lines a company has, the more breakdowns have to be set up. On the GetGate project five such product groups are counted automatically — and that is a separate setup on top of the basic pipeline, not part of the robot licence.

A working salesperson in this situation usually asks not about money but about whether the robot sets priorities correctly: whether it produces exactly the tasks that are closest to the money, for the right clients. That affects the price too — because tuning priorities for a specific pipeline takes an analyst's time, not just a programmer's.

Call quality control is a separate line, not a bonus

Many expect that since the agent already replies to clients, quality checking is included in the same price. In fact it is a separate service: quality control — that is AI which listens to calls and chats, scores them against a checklist, looks for clients let slip and calculates salespeople's scores. It can work without a sales robot too — simply as a layer on top of a human department.

The reason for keeping it separate is simple: quality control is needed even where only people are on the line, without a single bot. And the other way round — you can put a robot only on the first line of replies and not check live salespeople's conversations at all. These are different amounts of work, and merging them into one price would be wrong.

Checking real correspondence showed that in 80% of cases the assistant's reply was no worse than a human's in a blind comparison. That is the figure from one check of ours, not a guarantee for any business — but it explains why a separate quality control is needed at all: to measure rather than guess where the agent copes and where a human is needed.

What changes in the price if the sales department is built from scratch

A separate case is when the client has no working sales department at all: no scripts, no rules, no clarity about what to do with a deal at each step. Connecting a robot to chaos then makes no sense — first you need to build the sales department: scripts and rules, an analysis of why clients say no, a clear next step for every lead, onboarding for salespeople.

That adds work which would not exist if the client already had a functioning department. On the other hand, this is exactly where the gap we hear in business owners' questions gets closed: “my colleague and I lost money because the salespeople simply sat there without a system” — that is, without rules, without a clear task for today, without any analysis of where the money is being lost right now.

If the department is already built and working, this part of the price simply does not arise — the robot is trained for the existing processes rather than the processes being rebuilt. That is why the first conversation about price always starts with a review: is there a sales department, or does it have to be assembled together with the bot.

When an ordinary salesperson is cheaper, and when an agent is

The honest answer: an agent is not always more profitable than a person, and it is not about the subscription price but about the nature of the task. A live person is needed where the decision requires nuance, bargaining, emotional contact — an agent will not close a complex deal on its own. An agent pays off where the task repeats at scale: answering a standard question, qualifying a client, booking a meeting, doing it at any hour.

That very gap of 27% of enquiries in the evening and at weekends is exactly an example of the second situation. Hiring someone for a night shift for a flow of enquiries that arrives irregularly is expensive and awkward in terms of rotas. Here the agent does not replace the sales department but covers the hours when there are physically no people on the line.

The second benchmark is how much of a salesperson's routine goes on messaging and qualifying instead of calls and meetings. On the GetGate project a salesperson frees up about an hour of working time a day — not because they stopped working, but because the agent took over the initial review of the enquiry. If salespeople in the client's department have little such routine, the saving will be smaller, and the decision may not favour an agent at all.

Why there is no fixed price list on the site

A price list with figures would look more honest to a reader but would be inaccurate in substance: the price of the very same sales robot for a shop with three channels and for a chain with ten salespeople and telephony differs not because of a mark-up but because of the amount of real work. Publishing a single figure would mean either understating it for complex cases or overstating it for simple ones.

So the conversation about price is built not around a price list but around a review: which channels already exist, which CRM is in place now, how many salespeople and leads a month, whether separate call quality control is needed. The answer to each of these points changes the amount of work and therefore the final price — and that can only be understood on a specific sales department, not in a general article.

That is exactly why we describe the logic of the calculation openly: so that the reader comes to the conversation not with the question “how much does this cost in general” but with ready answers about their own channels, salespeople and CRM. Then the conversation about price takes less time and becomes concrete straight away.

Where to start in order to understand your own price

The first step is not a budget calculation but a review of the client's sales department: where money is lost, which enquiries never reach a deal, what is already set up in the CRM. It is the same first step we begin any project with — without it any price figure would be guesswork rather than a calculation.

Next the CRM is assembled for the process, the robot is trained for the niche, the channels are connected — and already at this stage you can see how much work this particular pipeline needs. Only after that does it make sense to price quality control and decide whether the rules and scripts have to be built from scratch or the agent can be fitted into what already works.

If the sales department currently has no answer to what happens to an enquiry outside working hours, or who reviews chats for clients let slip, that is the starting point for the conversation about price. You can read about the review itself and its logic in the blog, and more about each part on the service pages: sales robot and customer acquisition, if besides handling enquiries you also need a flow of them.