An AI agent for Bitrix24: what it costs and what goes into the price

The short answer: an AI agent for Bitrix24 costs from 90,000 ₽ to 540,000 ₽ to implement and from 8,000 to 105,000 ₽ a month to support — depending on what exactly is being called an agent. The six-fold spread isn’t marketing; it’s three different classes of system. Below: which ones, what moves the price within a class, and how not to buy more than you need.

Prices are indicative figures from my pricing for a mid-size company, not a public offer: the exact cost is fixed after discovery.

Three classes of system — three price tags

The “Auditor” — rule-based CRM control. 90,000 – 180,000 ₽ to implement and 8,000 – 20,000 ₽/mo. There is no language model inside: every day the system goes through deals and cards against preset rules and sends the people responsible targeted lists of violations with links. The Basic package covers up to 8 rules and one pipeline; the Extended one has unlimited rules and pulls data from 1C (the ERP/accounting system standard in Russia) into the checks. This is the least expensive class, and for the job of “getting Bitrix24 in order” it is usually enough.

The AI Sales Supervisor with an LLM layer. From 180,000 ₽ to implement and from 25,000 ₽/mo. On top of the rules there is a layer that understands a question in plain words (“how many leads have gone unanswered since yesterday?”) and answers with figures from the live CRM. In one project, a supervisor like this keeps watch over 2,470 leads and 1,815 deals around the clock.

The AI Sales Employee. 240,000 – 540,000 ₽ to implement and 45,000 – 105,000 ₽/mo. This is an autonomous LLM agent (Anthropic API) that sees 1C and Bitrix24 together: every morning it finds debtors, customers who have gone quiet, drops by product group and stalled deals, assigns sales reps tasks with a call brief, and sends the manager a digest. Three packages: “Analytics” (240,000 ₽ — read-only 1C access and reports), “Analytics + CRM” (380,000 ₽ — plus deal monitoring and automatic task creation) and “Full” (540,000 ₽ — plus autonomous call-round planning and free-form analytical questions).

Separately, there is the fleet of eight AI agents for product companies — 45,000 ₽/mo for all of them plus a one-off setup from 60,000 ₽. That is a different job (product metrics and unit economics rather than a sales team), so I won’t go into it here.

What moves the price within a class

For the AI Sales Employee, the package’s base price is multiplied by two coefficients.

• Business scale. A compact company with up to 300 customers and three sales reps pays a quarter less than the base price. A mid-size one (300–800 customers, 4–8 sales reps) pays the base price. A large one (800–2,000 customers) pays 40% more. A multi-branch network with over 2,000 customers is quoted individually.
• Region. Moscow and St. Petersburg — plus 25%; cities over 1 million — plus 10%; cities under 300,000 residents — minus 10%.

Then come the conditions that only become visible during discovery:

• CRM pipelines aren’t set up and need to be built — plus 50,000 ₽, adjusted for scale;
• a customized or modified 1C configuration — plus 30%;
• more than one 1C database — plus 25% for each additional one;
• a rush implementation ahead of the queue — plus 30%;
• the CRM isn’t used and only 1C analytics is needed — minus 20%;
• agreeing to become a public case study — minus 10%; a year’s monthly fee paid in advance — another 10% off.

The calculator on the pricing page applies the first two coefficients straight away; the rest come after discovery.

What the monthly fee includes

A common question: why pay 45,000 ₽ a month if the system is already implemented? Because it keeps running: a dedicated server, monitoring, a weekly reconciliation of changes in 1C and the CRM (new warehouses, fields and pipelines are picked up by the agent itself, with no developer involved), and a response within the working day. Plus the package limits: hours of modifications (4, 8 or 12 a month) and free-form analytical queries (15, 40 or 80). Scheduled reports and quick-access commands aren’t limited.

The minimum support contract term is six months, then month to month with 30 days’ notice.

How to pay

Implementation isn’t paid in full upfront. First comes discovery: 25,000 – 50,000 ₽ depending on scale (from 70,000 ₽ for a multi-branch network), delivering a readiness report, a technical specification and an exact estimate. If implementation isn’t worthwhile, you find that out at the cheapest step; if it is, the amount is credited against the cost. Then 40% at the start of work, 30% when the system’s figures match your report, and 30% after all modules go live. Why it is done this way is covered in a separate insight.

How not to overpay

The rule is simple: pay for initiative, not for the word “AI”. If the initiator is always a human and the questions are predictable, rule-based control for 90,000 ₽ is enough. If the questions change every day, you need an LLM layer. If the problem is that people don’t even have time to ask, you need an agent that asks itself. More on the three levels in the insight on chatbots vs. agents.

What a company actually gets for the money is in the case studies: the AI Sales Employee at a wholesale company (the first run surfaced 18 debtors and 143 missing stock items), the AI Sales Supervisor, the “Auditor”. Results from specific projects, not a guarantee of similar outcomes.