AI system for predicting customer risks (Risk & Sentiment A

Development of an intelligent system that automatically analyzes the tone of dialogues in the CRM system in real-time, allowing management to detect conflict situations before they become critical.

Challenge:
The client faced the problem of "unexpected" refusals and conflicts. Managers often did not manage to notice changes in the client's mood in correspondence in time, which led to "putting out fires" already at the stage of losing a deal. A tool was needed that could "see" tension in communication 24/7.

My solution:
I designed a system that transforms each text in the dialogue into a risk assessment:

Data integration: Set up a continuous flow of data between the CRM system and the processing service via API (Make).

AI sentiment analysis: Implemented neural network analysis that evaluates each message on a scale of emotional coloring. The system assigns one of three statuses to the dialogue:

Green: Stable loyalty.

Yellow: First signs of dissatisfaction.

Red: High risk of conflict, immediate intervention required.

Preventive notification system: If the dialogue falls into the "Red" zone, the responsible manager or supervisor immediately receives a notification with an excerpt of the risk reason.

Technology stack:

Make (Integromat)

OpenAI / Claude (Sentiment Analysis API)

CRM Integration (API/Webhooks)

Notification Stack (Telegram/CRM Internal alerts)

Result:

Shift to strategy: The client replaced the reactive "firefighting" model with proactive relationship management.

Reduction of churn: Conflict situations are detected at early stages, allowing for the preservation of customer loyalty.

Transparency of communications: The manager sees the "temperature" of the sales department in real-time without having to read thousands of messages manually.