SalesGen

Sales Conversation Conversion Classifier

Overview

This model is designed to analyze a sales conversation between a customer and a sales representative and predict whether the conversation is associated with a potential conversion.The model is intended to serve as an initial classification layer for a broader lead-intelligence system. Its purpose is to help identify conversations that may require further attention from a sales team.

Intended Use

The model can be used for:

• Classifying sales conversations into conversion and non-conversion outcomes.
• Prioritizing conversations for further review.
• Supporting automated lead-identification and follow-up systems.
• Providing an initial prediction before structured customer information is extracted.

Model Variants

Two model versions were evaluated during development. The primary difference observed between the evaluated versions was the number of false-positive predictions:

• Model 1: 881 false positives
• Model 2: 854 false positives

The second evaluated model produced fewer false positives in the reported evaluation.

Evaluation

The models were evaluated using standard classification metrics and confusion-matrix results. The evaluation included:

• Precision
• Recall
• F1-score
• Accuracy
• ROC-AUC
• PR-AUC
• Confusion matrix

Classification Report

<!-- IMAGE PLACEHOLDER: Insert classification report screenshot for Model 1 here -->

Image: Classification report of both the models are almost same (minute difference).

Limitations

The available evaluation is based on the selected sales-conversation dataset and its associated conversion outcome labels. The model's current prediction target represents conversion outcome and should not automatically be interpreted as a definitive business-level lead classification. Real-world performance may vary depending on:

• Industry
• Product or service
• Customer behaviour
• Conversation style
• Quality of transcripts
• Business-specific definitions of a qualified lead

Intended Role in the Larger System

The model is intended to act as a prediction component within a larger customer-intelligence system. Its output can be combined with structured information extraction and business rules to support sales operations and lead management.

Status

Development / Prototype The model is currently being evaluated as part of a hackathon prototype and may undergo further improvement with domain-specific and real-world conversation data.