MTS Exolve showed how to automate deal data entry in Bitrix24 with YandexGPT

MTS Exolve demonstrated a practical approach to sales automation: after a call ends, a Python service automatically retrieves the conversation transcript, sends it to YandexGPT, and updates the deal card in Bitrix24. The idea is simple but useful for any team where managers lose details after calls or enter data into the CRM with delays.

How the pipeline works

At the core of the solution is a Flask webhook that receives a call completion event from MTS Exolve. The service then retrieves conversation data, saves technical context in SQLite, and launches a processing chain without manager involvement. This approach transforms a call into a structured object: it has a transcript, an identifier, a connection to the deal, and a set of fields that need to be updated in the CRM.

The next step is transcription. Instead of manual listening, a Call Transcription API is used, which provides conversation text for further analysis. After this, YandexGPT comes into play: the model receives a prompt with the necessary extraction schema and returns not free-form text, but specific lead qualification signals.

The final stage is a call to the Bitrix24 REST API, which updates the existing deal and saves the result directly in the work card. Additionally, this scheme leaves a transparent trace: the team can verify exactly what the model extracted and which fields were changed automatically.

Essentially, MTS Exolve describes not just an integration of three APIs, but a minimal workflow for post-processing sales calls. It doesn't require heavy infrastructure: Python 3.10+, Flask, SQLite, and clear event routing logic are sufficient. For an MVP, this is enough to verify the main effect — whether managers will work faster with qualified deals and lose less context between touchpoints.

Which fields does AI extract

BANT serves as the foundation — one of the most intuitive lead qualification frameworks. But in the example, it's not left in textbook form. The classic Budget, Authority, Need, and Timing are supplemented with practical signals from real conversation: how interested the client is, which competitors they mention, and what objections they raise.

As a result, the CRM receives not a long transcript, but a compact summary by which the next step can be planned.

• client budget or price expectation range
• decision-maker and their role in the process
• articulated need and specific use case
• implementation timeline, pilot, or next contact date
• interest level, competitors, and key objections

This is an important point: the value here is not only in saving time filling out the card. Structured fields allow different work with the funnel — faster filtering of raw incoming flow, prioritizing hot deals, and not starting each next call from scratch. This is especially useful for teams where one manager handles dozens of contacts and can't keep the details of each call in their head.

An additional plus is the moderation of the scheme. The authors don't try to turn the CRM into a form with dozens of mandatory fields. On the contrary, they show a balance between completeness and practicality: take a minimal set that really helps sales and automatically maintain it in current state. For internal AI projects, this is a more viable path than implementing overly complex classification from the start.

What this means

MTS Exolve's material clearly shows where applied AI in sales is moving: from chat assistants to background services that automatically fill the tracking system after an event. This scenario doesn't require replacing the CRM and doesn't look like a research prototype — it's a clear MVP that can be built on a standard Python stack and quickly tested on real calls. If the quality of transcription and prompting is high enough, the company gets a cleaner sales funnel with almost no additional manual work.