Six AI agents behind one growth strategy

The AI agent began transcribing all inbound and outbound calls of the active-sales managers. Every call was automatically converted to text, after which the agent checked it on two fronts.

First — script compliance: did the manager go through all the required stages of the conversation, ask the right questions, and handle objections to the standard? Second — extracting client data: what matters to them when choosing a provider (price, speed, reliability, level of service), their current connection speed, their monthly spend, and whether they have problems with their current provider.

All of this data flowed automatically into the client record in Bitrix24 — in a structured form, in the right fields. The manager would finish the conversation and the record was already filled in. No manual entry, no delays, no lost details.

The AI agent gathered data on all active deals and built a report for the head of sales every morning — without queries to managers and without manual assembly.

The report showed conversion at every stage of the funnel: reaching the decision maker → agreeing to a meeting → holding the meeting → sending the details → closing the deal and payment. For each stage it was clear how many deals moved forward, how many got stuck, and which managers were performing below the team average.

In addition, the agent analysed call transcripts: at which points in the conversation objections came up most often and how managers handled them. Managing sales stopped being based on gut feeling — now it is data that is updated every day.

Every time a new client moved from an active-sales manager to an account manager, the AI agent began tracking the quality of that handover against a specific checklist.

Did the account manager get to know the client personally? Did they find out what matters to the client in working with a provider — not from the record, but in a live conversation? Did they identify what could be offered as an add-on and when that would be appropriate? Did they set themselves a task with a reminder?

The agent analysed calls and CRM activity over the first month after the handover. If any checklist item was not completed, the manager received an alert the same day, rather than learning about it a month later at a planning meeting.

The fourth agent worked with clients who were already connected — those who had been in the base for a long time and who are easy to stop paying attention to.

For each client the agent checked: was there a planned contact this month, are the problems from past requests closed, and are there signs of hidden dissatisfaction — situations where the client mentioned something in a conversation but never filed a formal request.

At the same time, the agent analysed upsell potential: it looked at the structure of consumed services, compared it with similar clients, and highlighted who could logically be offered what — with a concrete rationale, not just a "potential here" flag.

The fifth agent worked in the contact centre. It analysed all support calls and correspondence: it transcribed conversations, identified the topic of each request and the overall tone — whether the client was satisfied, neutral or frustrated.

On this basis the agent did two things. First, it classified requests by problem type: no connection, low speed, a billing question, a quality complaint. This produced accurate statistics on the real reasons for requests, rather than the categories an operator picked manually. Second, it built a list of clients with a high level of dissatisfaction: those who raised the same problem several times or in whose conversations the agent detected rising frustration.

Working from this list, account managers made a proactive call — without waiting for the client to call themselves intending to cancel the contract.

The sixth agent worked not on the outside but on the inside — helping the operators themselves during the conversation with the client.

From internal documentation, technical regulations and the accumulated history of requests, the agent built a knowledge base. When an operator received a call, the agent identified the topic of the request in real time and suggested a ready answer or a sequence of actions — right in the operator's interface, while the conversation was still going on.

Routine questions — equipment setup, explaining a bill, standard technical issues — the first-line operator began resolving on their own, without handing off to a senior specialist. At the same time, complex cases were escalated faster: the agent helped the operator recognise that a question was non-standard right at the start of the conversation.