Projects / Bitrix24 implementation / Bagration

Implementation case study

Bagration: an AI‑robot calls 400+ stores a day and collects orders in Bitrix24

For a dairy plant in East Kazakhstan we replaced the daily manual ring-round of retail outlets with on-premise Bitrix24, the client’s own Asterisk PBX and an AI robot. The robot calls stores, recognises the answer and moves the deal through the pipeline. Operators only deal with those who are ready to order.

On-premise Bitrix24Asterisk1C integrationAI callingSpeech recognition

I want the same How it works

An AI robot calls stores and sorts their answers into a Bitrix24 pipeline; orders go to 1C

900+robot calls a day with no operators involved

15simultaneous calling lines

90 minto call 430 stores in the morning wave

2people per shift instead of a whole calling team

About the client

A dairy plant with its own distribution network

Bagration Ulan LLP has been operating since 2004. Its milk processing plant is in the village of Proletarka, Ulan district, East Kazakhstan Region, and its office is in Ust-Kamenogorsk. The company makes cheeses, cultured dairy products, cottage cheese, sour cream, butter, yoghurts and traditional kurt, as well as products under its own Nanolakt brand with bifidobacteria.

The products are perishable, so sales run as a daily cycle: sales reps are assigned to stores, stores order often and in small quantities, and orders go to 1C and ship the next day.

BAGRATION

Bagration Ulan LLP · bagration.kz

Industry: Dairy production
Region: East Kazakhstan, Ust-Kamenogorsk
In business: since 2004
Sales channels: retail stores, sales reps, online store
Project timeline: August — October 2026

1,000+counterparties in 1C, about 650 of them retail outlets with a calling schedule

6sales reps, each with their own list of stores

~300product items with prices and stock changing every hour

2calling waves a day: the main one in the morning and an evening one for “daily” stores

The problem

Calls ate up the working day, and orders got lost

Before the project, order collection relied on people and the phone. Every morning staff manually rang hundreds of stores: “Do you need anything today?”. Most calls ended in “no” or “no answer”, but each one cost operator time.

Routine instead of selling

Hundreds of identical calls a day. Little time was left for the store that actually wanted to order.

No control

Who was called and who wasn’t, who promised to call back, what they said — it was all in people’s heads and notebooks. There was no way to check the work.

A gap between 1C and the phone

Calling schedules, sales reps and recent orders lived in 1C, but calls were made from paper lists. The order was then keyed into 1C by hand.

Project goals

Four goals on one platform

The client did not need a “calling robot” but a working cycle: from the daily calling plan to an invoice in 1C.

Build the daily calling plan automatically

Bitrix24 itself works out whom to call today: by the store’s schedule, the “deactivated” and “do not call” flags and whether there is a phone number. A deal is created for each store with the products from its last order, so the operator sees what this outlet usually buys.

Call straight from Bitrix24, without connectors or third-party software

Drop cloud telephony and paid integrations. The client’s own Asterisk PBX, inbound and outbound calls right in the Bitrix24 interface, with the company card and the call recording in the deal.

Hand the first call over to an AI robot

The robot calls the store, asks the question, recognises the answer in Russian and Kazakh, understands “yes”, “no”, “later”, silence and a hang-up, and moves the deal to the right stage. A person steps in only for those ready to order.

Connect everything to 1C both ways

Counterparties, products, prices and stock flow from 1C into the CRM; calling schedules, sales reps and purchase history come in a separate export; the finished order goes back to 1C as an invoice.

The solution

How it works

The platform is on-premise Bitrix24 on the client’s own server. All logic is written as separate modules and services and the Bitrix24 core was not modified, so portal updates do not break the project.

1C:Enterprisecounterparties, prices, stock, calling schedules, sales reps, order intake

standard connector and morning export⇄invoice for the order

Bitrix24 (on-premise)

Auto-deal moduledaily plan, assignment by shift

CRM: calling pipelinedeals, products, invoices

Browser telephonycalls from the record card

Call monitoringrecordings and transcripts

calls and events⇄

Asterisk PBXcarrier line, inbound and outbound calls, call recording

⇅deal stages

⇅call, recording

AI calling dispatchertakes “To call” deals, runs up to 15 calls at once, waits for the recording, recognises and parses the answer, moves the deal

⇄

AI servicesspeech recognition (RU/KZ), understanding the meaning of the answer

Call recordings, CRM data and speech recognition stay inside the client’s infrastructure.

Auto-deal module

Bitrix24 builds the calling plan itself and hands it out to staff on shift

Our own Bitrix24 module goes through every CRM company each morning and picks those to call today: by the day-of-week tick in the store’s schedule, without the “deactivated” and “do not call” flags, and with a phone number. For each one a deal is created in the calling pipeline with the products from the store’s last order: the operator immediately sees what this outlet usually buys and does not re-enter the items.

The evening wave covers the “daily” stores that prefer an evening call. At the end of the working day the module closes unfinished deals itself so the pipeline is clean in the morning.

Assignment is built around the real shift: an employee opens their working day in Bitrix24 and picks today’s role — “operator” or “dispatcher”. Deals of their sales reps go to them; if nobody is on shift, the deals wait with a backup owner and are handed out as soon as someone comes online. An operator sees only their own deals and cannot edit company data.

Telephony

Asterisk connected directly to Bitrix24

Instead of cloud telephony and off-the-shelf connectors we set up the client’s own Asterisk PBX with a carrier line and connected it to Bitrix24 with our own integration. Staff make and take calls in the browser, in the standard Bitrix24 interface — no desktop softphones, integration licences or middlemen.

Outbound. Clicking a number in a deal or company dials out through Asterisk. The call card immediately shows the store name and a link to today’s deal, and the call recording lands in the timeline.
Two-step inbound. When a store calls back, the PBX asks Bitrix24 who is responsible for that outlet today and rings them first if they are on shift. No answer — the call goes to everyone on shift. An unknown number goes to everyone straight away.
Robot and people on one PBX. One line serves both AI calling and operators’ manual calls. Live calls appear in the same monitoring as the robot’s.

03 · AI calling

The robot makes the first call to every store

This is the heart of the project. The question “will you be ordering today?” is asked by a robot, not a person. It works through the whole day’s pipeline, and operators get an already filtered list of stores that said “yes”.

Step 1DispatcherConstantly picks up “To call” deals and launches up to 15 calls in parallel. Deal locking is atomic: a store is never called twice at the same time.

Step 2CallAsterisk dials the store, plays the greeting and records the other party’s voice from the first second. If the store starts answering before the greeting ends, the answer is not lost.

Step 3RecognitionThe recording goes to a speech recognition model deployed on the client’s side. It understands Russian and Kazakh and works without sending audio to external services or paying per minute.

Step 4UnderstandingA language model parses the answer text: it tells “yes” from “no”, “call back later”, unclear speech and a polite “hello” with no answer.

Step 5Action in the CRMThe deal moves to the stage that matches the outcome, and the transcript and a link to the recording are written to the timeline. A store that said “yes” goes to an operator; everything else closes on its own.

Rules agreed with the client

What the robot heard	What happens	Deal stage
“Yes, we need some”	The deal goes to an operator, who confirms the order and the products	Order
“No, nothing today”	Closed without anyone’s involvement	Declined
“Call back later”	Passed to an operator, who calls back	Problem call
No answer	All the store’s numbers are tried; one retry later the same day, a second failed attempt closes the deal	Call back later → Declined
Silence or hang-up	Closed immediately, no retries	Declined
Unclear	Passed to an operator together with the recording	Problem call

The chain is protected against typical failures: the attempt counter is kept in Bitrix24 rather than in the handler’s memory, so a dispatcher crash does not cause repeat calls. If the language model is temporarily unavailable, the recording is parsed later rather than lost.

1C integration

1C remains the source of truth; the CRM gets everything needed for the call

The exchange is built from two independent flows so as not to break the standard mechanism while still getting the data it cannot transfer.

Standard Bitrix24 — 1C connectorRegularly syncs counterparties into CRM companies, and products, prices and stock into the catalogue. More than 1,000 counterparties and about 300 items are always up to date.

Morning export of extra dataThe connector does not transfer additional counterparty properties, so every morning 1C provides a separate secure export: the calling schedule, the sales rep, the “deactivated” and “daily” flags, phone numbers and purchase history. The importer updates only the changed fields and normalises phone numbers.

The order back to 1CWhen the operator saves the product list, the deal moves to the “Sent to 1C” stage, an invoice is generated from it, and 1C picks it up for processing.

Exchange monitoringA separate admin page shows both flows: what arrived from 1C, what did not match, where a file turned out empty. Problems are visible before they reach the operators.

Control

Call monitoring and fast order intake

The “AI call monitoring” page inside the portal shows every call of the day: the store, the direction (robot, inbound, manual outbound), the outcome, the transcript and the recording. Filters match the deal stages, so the manager sees the calling pipeline exactly as the operators do.

The “Products” tab in the deal was adapted for taking orders by phone: search by name from the first letters, “+” and “−” buttons for quantity, totals recalculated on the fly. A typical order of 10–15 items is entered while the store is still on the line.

Results

Calling became a background process, not a working day

Go-live was on 21 October 2026. The figures come from the system logs for the first working days.

800–930robot calls a day. People would otherwise have to make the same number by hand.

430+stores in the morning wave called in an hour and a half.

~85%of calls close without an operator: “no”, no answer, silence.

2 wavesa day, morning and evening for “daily” stores, without compiling lists by hand.

100%of calls with a recording, transcript and outcome in the deal. Any conversation can be checked.

0subscriptions to cloud telephony or connectors: the PBX and the integration belong to the client.

The main change is in people’s roles. Two people per shift, an operator and a dispatcher, no longer spend the morning asking “do you need anything today?”. They take orders from those who have already said “yes”, call back on problem calls and answer inbound calls from their stores. The rest is done by Bitrix24, Asterisk and AI working together.

The project keeps evolving. The robot’s rules are refined using real call recordings, and since all the logic is in the client’s hands, changes take hours, not weeks.

Technology

What it is built on

On-premise Bitrix24

Our own auto-deal module, CRM roles, deal card improvements, monitoring pages. The core is untouched.

Asterisk

The client’s own PBX, carrier line, calls from the browser, call recording, connection to Bitrix24.

Automation and AI

Call dispatcher and handler, speech recognition on the client’s side, a language model to understand the answer.

1C:Enterprise

Standard connector for counterparties and the catalogue, morning export of extra data, invoices back to 1C.

A similar task? Regular customer calling, taking orders by phone or your own PBX instead of a cloud one — we will look at your process and show what part can be handed to a robot. More about Bitrix24 implementation and AI agents for business.

Discuss a project

Tell us how you collect orders today

How many customers you call, who makes the calls and where the data lives: in 1C, spreadsheets or a CRM. That is enough for us to see at the first meeting which part of the calling can be handed to a robot.

We reply within a business day and meet online or at our office in Almaty or Astana.
We will show how the calling in this case study works.
We will advise whether you need your own PBX or cloud telephony is enough.

Have a specification or a process description? Send it to info@oneit.kz (PDF, DOCX, XLSX up to 20 MB) — and mention it in your message.