arXiv:2503.23303

SalesRLAgent

Predicts sales conversion probability during live conversations

Deep Dive

Video Explanation

Featuring Preston Rhodes

Watch original on Instagram

What It Does

Real-Time Conversion Prediction

SalesRLAgent analyzes each message in a sales conversation and calculates the probability that the deal will close. As the conversation progresses, it updates predictions in real-time, helping sales reps understand exactly where they stand.

Updates every message turn

See It In Action

Turn-by-Turn Analysis

Customer

"I need to think about it..."

34%
Convert

Agent

"What specific concerns do you have?"

52%
Convert

Customer

"The price is a bit high"

48%
Convert

Agent

"Let me show you the ROI..."

78%
Convert

Customer

"That makes sense, let's proceed"

94%
Convert

How It Works

Reinforcement Learning

01

State Embedding

Each message is converted to a 3072-dimension vector using Azure OpenAI embeddings, capturing semantic meaning.

02

Sequential Learning

A PPO (Proximal Policy Optimization) agent learns from conversation trajectories, treating prediction as a decision problem.

03

Probability Estimation

At each turn, the model outputs a conversion probability based on the full conversation context so far.

Results

Performance

96.7%
Accuracy
85ms
Latency
40×
Faster
+43%
Conversions

Try It

pip install deepmost