How to Ensure Accurate Representation of Travel Packages in Sales Conversations?

I’m working on a travel booking project where we need to analyze sales conversations to ensure that an agent accurately represents various travel packages offered by our company during their pitch to customers. We have 10 to 15 different active travel packages, each with a detailed around 1-2 page specification document outlining their features, destinations, pricing, and inclusions.

The main objective is to verify that the agent accurately described all the travel package details and features during their conversations with customers. What would be the best prompting approach to determine if the agent’s conversation with customer complies with prescribed guideline?

Given this is to be tested on all conversations and guidelines are not huge, we are not approaching the problem with RAG. Rather we are approaching it by writing Mega Prompts (~2000 words) to detail the scenario and passing the whole conversation.

What would be a better optimized way to approach this problem? Any insights or suggestions would be greatly appreciated!

Welcome to the community!

What is your optimization target? Operational cost, development cost, speed, or accuracy?

100% accuracy is almost impossible.

Here’s a potential problem with “megaprompts”

One cause of inaccuracies is context confusion. This happens when you have a long context that contains many similar passages. If you put multiple products into your context (prompt) and they are structured fairly similarly, it’s possible that the model “looks” at the wrong passage when generating a response. To prevent this from happening, it’s a good idea to elide everything that is not relevant to the current user interaction.