How to train your Apollo AI Chatbot to respond to your chats

When setting up your Apollo AI Chatbot, there is nothing more important than being confident in its ability to provide accurate and helpful responses every time. In this guide, we'll share our best practices to help you ensure that it does. Let's dive in!

1. Add as much relevant content regarding your organization, products, and services to your data sources.

AI is only as good as the information it

It's crucial to provide your Apollo AI chatbot context about your organization, FAQs, and products/services is crucial. The more relevant information you include, the better equipped the AI will be to answer queries for you.

There are 5 Data Sources you can use:

• Plain Text
• Document Upload (we support .txt .pdf .csv and recommend using .csv for product lists and data matrixes)
• Shortcuts
• Knowledge Base*
• Your website (plain text scraping)

*When creating a Knowledge Base, you can copy/paste content from your website, FAQs, and other documents. Formatting isn't as important, however, if you would like to keep your Knowledge Base public, you can use AI Commands to convert articles into a customer-friendly format and in a tone that matches your business. If you do this, we recommend keeping each KB article to one topic.

Here's an example prompt you can use:
"Please convert this into a clear and concise help center article that is written in an informative yet personable tone:"

2. Stress test the bot using the preview.
Before unleashing your Apollo AI Bot into the wild (i.e. responding to your customers), it's pertinent that you stress test it using the preview feature in the AI Assist settings. Simply act as the customer and simulate real-life interactions to see how Apollo performs in different scenarios. This will allow you to identify missing information and areas for improvement, giving you the opportunity to fine-tune the responses before it goes live.

Tip: If you have added information to the KB and Apollo isn't picking it up, please ensure that you have connected your KB as a data source! Otherwise, it will not be able to access it. To do this, click the toggle next to Knowledge Base in the AI Assist settings > Edit and select the Knowledge Base from the dropdown menu.

3. Check for conflicting information in your data sources.

If the bot still fails to provide the correct answer, there might be conflicting information in the data sources. You can easily identify this by clicking the data source icon in the AI Assist Preview widget to see where the response was taken from. By identifying the conflicting information, you can take steps to resolve the issue, such as updating or removing conflicting data sources. This ensures that AI Assist has access to reliable and consistent information, leading to more accurate responses.

4. Can't get it to answer a particular question exactly the way you want it to? No worries, let's create a Shortcut!

When you come across a question that AI Assist answered incorrectly or could be better optimized, the best way to tackle it is by creating a Shortcut with the right information. By doing so, you're teaching the AI how to respond accurately to that particular question. Test it out by asking the same question again and see if the AI provides the correct answer this time.

Just make sure that the Shortcut is connected with the AI's data source!

5. Use Shortcuts for verbatim responses.

Sometimes, you may want AI Assist to respond verbatim, without any variations. In such cases, you can add respective question(s) to the Shortcut. This ensures that when those questions (or similar variations) are asked, the AI will provide the exact response you have specified. It's a great way to ensure consistency and precision in certain situations.

6. Once you're happy with the way it's responding, it's time to turn it on.

Which brings us to point 7...

7. Regularly review your chat history.

To ensure AI Assist is meeting your expectations, it's important to review the chat history regularly. This will help you identify any questions that were answered incorrectly or could be optimized for better responses. By keeping an eye on the chat history, you can stay on top of any issues and take proactive steps to improve the AI's performance.

If you've prompted it to capture customer information, you will want to review the chats anyway just so you can follow up with your leads!

In conclusion, training Apollo AI Bot to meet your expectations is a continuous effort, but you should quickly run through all the common scenarios to have it answer perfectly almost all the time.

With these strategies in place, your Apollo AI Bot will be well-equipped to deliver responses that meet or exceed your expectations.

If you need further assistance, our Customer Success team are happy to hop on a call with you to review your requirements and help you achieve your desired responses. Click here to book an onboarding call.

Happy training!