Tracking Sales Processes Efficiently with AI-Enabled Chat Bot:

In the fast-paced world of sales, staying on top of every detail in a lengthy email chain can be a daunting task! In this case study, we explore how AI powered chatbot is simplifying communication, and enhancing collaboration within sales teams. Let's delve into the scenario and understand how this innovative solution tackles the challenges faced by sales professionals.

Picture this! You are a salesperson who just takes over other's projects. Typically, the only way to transfer a sales pipeline is to forward lengthy email chains and manually explain the current status to others.

However, an AI-driven chatbot could undertake this task and provide you with the information from emails lying in your mailbox.

Now, let's take a look at how it works!

Workflow

Our potential customers use Outlook to deal with daily emails and here is our workflow:

1. Link your Outlook with our application, and the application will preprocess the emails and embed them into the database.
2. Ask questions in natural language on the website.
3. Leverage the hybrid search algorithm to fetch the most relevant email messages and reconstruct into email conversation
4. LLM generate a response using the provided prompt, including the user's question and relevant emails
5. Run Outlook Data Loader daily to update incoming and outgoing emails

Previously, We just released a compelling user case on our Slack channel that highlights our company's data focus. Dive into our story first: https://us.fixstars.com/case/llm-slackbot

While the foundation remains the same, following specific customizations have been made for this email demo.

Email preprocessing

Clean the original email

It's a familiar scenario: email grows longer with each reply, since it bundles up the entire conversation history. It is not efficient to reconstruct the conversation in the Q&A phase later.

To address this, we employ email preprocessing, extracting plain text from the email body and trimming previous replies.

Summarize the length email

Sometimes the conversation is just too long since sales people always send messages back and forth during the sales chain! It's hard to track from the initial message to the very end and because of the hardware constraints, we can not use unlimited token length at this moment.
To tackle the issue of lengthy email conversations, we use large language model(LLM) to summarize the lengthy email.
This way, we can concatenate email messages as much as possible.

Hybrid search

Hybrid search is a technique that combines multiple search algorithms to improve the accuracy and relevance of search results. It uses the best features of both keyword-based search algorithms with vector search techniques.

Reconstruct email conversation

When an email is loaded into the database, we store both the email content, corresponding embedding, and its metadata, which includes the message ID and conversation ID.

When a user asks a question, we retrieve most relevant emails regarding with the user’s question, and reconstruct emails that are under the same conversation ID, arranging them from the oldest to the latest.

With all these changes, we can now use this chatbot to track the sales pipeline more easily!

Here are a few more examples:

Note all the conversations in this blog are examples

Reference

1. https://ai.meta.com/llama/