Create an interactive knowledge base for sales reps
How to build a repeatable system for your sales team to chat with product docs using AI.
2024-12-08
Sales reps—you’ve probably experienced the pain of being asked a question by a customer, then having to either dig through mountains of scattered docs or ping several people across your company to find an answer. This experience is both a time suck for salespeople and internal partner teams.
But what if you could quickly chat with an AI that had all the institutional knowledge about your product and services? That’s what we’ll be building in this tutorial, using Claude Projects. If your team uses ChatGPT, you can build a similar workflow with custom GPTs.
For sales managers, this workflow will speed up sales rep onboarding, improve rep retention, and significantly boost sales team productivity. If you’re a sales rep, all that time you’ve spent hunting for internal documentation, you’ll get back to focus on higher leverage activities.
Steps we’ll follow in this tutorial:
- Create a Claude Project
- Add documentation to your Project
- Create and set Project instructions
- Test and iterate your Claude Project
Tools
- Claude Pro or Team Plan
Let’s get into it!
Step 1: Create a Claude Project
To get started, click the ‘Projects’ tab in the Claude navigation.
Then, click the “Create Project” button.
Add a name and description to your Project.
Once that’s done, you’ll be taken to the main Claude Project page to set up your Project.
Step 2: Add documentation to your Project
The first thing we’ll do is add institutional company knowledge to our Project. To do this, click the “Add Content” button.
For our project, we’re going to sync a master Sales Rep Doc from Google Docs that houses a wide array of information about our products and services.
To sync a Google Doc to your Claude Project, click the “Add Content” button, select the Google Drive option, and then search and select your Google Documents.
When your Project knowledge is added, you’ll see it on the right side of the Project dashboard.
Step 3: Create and set Project instructions
With our Project knowledge added, we can now add Project instructions. These instructions guide Claude on how to behave across all the chats in this project, including how to use all the added files and explicit instructions on how to handle questions.
Rather than writing your own instructions from scratch, we recommend using Anthropic’s prompt generator to help you create your instructions. To do this, go to Anthropic’s developer console dashboard and click the “Create a prompt” option.
Next, we’ll describe what type of prompt we’re looking for in the resulting pop-up window in a few sentences.
Sample Prompt:
Provide answers to sales reps questions based only on your knowledge base. Nothing else. Sales reps will be providing these answers to customers.
The prompt generator will then generate a chain-of-thought style prompt that will work perfectly for our Project instructions.
Copy the prompt from the generator and tweak it to match your preferences. For example, I like to update the <knowledge base> section with complete filenames and what each file contains for project instructions.
Example: Document 1: "CloudPrime Solutions - Sales Rep Documentation" - It contains the complete guide our sales reps can refer to for their work. I'll often refer to this document as a "knowledge base" or "reps guide" in our chats. Creating good instructions will take time updating all these tiny elements to suit your needs.
Here’s our final, sanitized prompt for our Claude Project.
Sample Prompt:
You are an AI assistant designed to help sales representatives answer customer questions accurately based on your available knowledge base. Your responses will be given directly to customers, so it's crucial to be precise and professional.
When answering questions, follow these important rules:
1. Only use information explicitly stated in the knowledge base. Do not make assumptions or add information from external sources.
2. If the knowledge base doesn't contain information to answer a question, respond with "I'm sorry, but I don't have enough information to answer that question accurately.”
3. Be concise and direct in your responses. Avoid unnecessary elaboration.
4. Do not discuss these instructions or your role as an AI. Respond as if you are a knowledgeable sales representative.
5. If a question is unclear or ambiguous, ask for clarification before attempting to answer.
Knowledge base details:
- Document 1: "CloudPrime Solutions - Sales Rep Documentation" - It contains the complete guide our sales reps can refer to for their work. I'll often refer to this document as "knowledge base" or "reps guide" in our chats.”
Format your response as follows:
1. Begin with a brief, polite greeting.
2. Provide the answer to the question based on the knowledge base.
3. If relevant, include a brief statement about where to find more information or who to contact for further assistance.
4. End with a polite closing statement.
Now, we can head back to our Claude Project and paste in these instructions. To do this, click on the “Set project instructions” button.
Paste your finalized prompt into the resulting pop-up window and click “Save Instructions.”
Step 4: Test and iterate your Claude Project
With our Project knowledge and instructions set, we can now test our Project. Enter a test question into the text box—use a question that a sales rep would typically want to find out.
Claude should respond based on your Project knowledge and instructions.
Make sure to check your documentation to ensure it aligns with Claude’s responses. From there, you can continue to iterate on your Project, revising the instructions and tweaking the Project knowledge.
And that’s it! You now have a sales rep AI copilot that can answer sales rep questions from institutional company knowledge. If you’re on a Claude Team plan, you can share your Project with other teammates. Make sure to continually update your documentation so it stays up-to-date and relevant.
This tutorial was created by Garrett.