Multi-Agent LLM Simulation for Financial Sales Pitch Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for cross-selling and product recommendations in financial institutions are largely manual, placing a significant operational burden on personnel and lack the ability to simulate real-world interactions effectively, especially due to limited visibility into client data and varying client interests.
Innovation Solution
A multi-agent simulation system using large language models (LLMs) to mimic the sales ecosystem, where agents simulate interactions between clients, researchers, and financial institutions, incorporating external data sources to enhance prompts and iteratively refine sales pitches based on client responses, with a judging agent scoring the effectiveness of these interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual methods are used for cross-selling and product recommendations, then personnel can directly interact with clients, but operational burden increases and efficiency decreases
Solution Approach 1:
The system enables automated self-service through multi-agent simulations that independently conduct sales interactions, evaluate pitches, and generate recommendations without human intervention. LLM-based agents autonomously perform cross-selling tasks, freeing personnel from manual operations while maintaining interaction quality.
Solution Approach 2:
Manual mechanical processes of sales interaction and evaluation are replaced with an automated digital system using large language models. The system substitutes human personnel's mechanical work with AI-based multi-agent simulations that automatically conduct sales pitches, client interactions, and performance evaluations.
2Reliability
If existing manual methods are used, then simple interactions can be handled, but ability to simulate real-world interactions effectively is limited
Solution Approach 1:
The system introduces dynamic adaptability through multi-agent simulations that can adjust to varying client interests and data visibility conditions. The LLM-based agents dynamically modify their behavior based on simulated client responses, making the system adaptable to different real-world scenarios rather than following fixed manual procedures.
Solution Approach 2:
The system performs preliminary actions by pre-populating client data and configuring multiple specialized agents before simulations begin. This preliminary setup includes preparing sales agents, client agents, researcher agents, and judging agents with relevant data and instructions, enabling more accurate and realistic simulations compared to ad-hoc manual interactions.
3Measurement precision
If more client data is incorporated into simulations, then accuracy of recommendations improves, but system complexity increases
Solution Approach 1:
The system segments the complex task of cross-selling into multiple specialized LLM-based agents, each handling specific functions such as sales pitches, client interactions, research, and evaluation. This segmentation manages complexity by distributing data processing and decision-making across independent agents rather than requiring a single complex system.
Solution Approach 2:
The system introduces intermediary judging agents that mediate between sales agents and evaluation processes. These intermediary agents simplify complexity by standardizing the evaluation of sales pitches and client responses, providing a structured method to incorporate diverse client data without overwhelming the system architecture.
Data Source
AI summary
Methods, systems, and techniques for performing a multi-agent simulation. A first large language model (“LLM”) is prompted to act as a sales agent of a financial institution to generate a sales pitch. A second LLM is prompted to act as a client of the financial institution to engage in a conversation with the first LLM in response to the sales pitch. A third LLM is prompted to act as a judge to generate and output a score of the conversation between the first and second LLMs. The score is saved and/or output to a display. The multi-agent simulation is used to create a digital twin of an actual conversation between the sales agent and client.