Your support team faces customer concerns and tickets daily so they understand how the product is being used/where users are having the most problems.

You can use this to draw patterns and see what your users are having the most trouble understanding.

Break these down into FAQs or byte-sized questions and answer them on your knowledge base directly.

Don't leave anything to assumptions. Make sure your entire product is explained easily from the smallest features to more complex features.

It's important to have detailed help articles that explain the product in-and-out but sometimes the user might be looking for a very small clarification in which case going through entire articles can be frustrating.

Create a separate section just for FAQs and answer more simpler questions about the product. Here, you can keep the answers concise and redirect the users to the main article in case they want to read the whole article but don't have to.

Typically, when users are searching for answers in your knowledge base, they use questions or keywords in their queries.

While creating your knowledge base or tutorials, you can name the article as how the user would ask the question.

For example, if it's an article about sending an email without using CSVs, instead of writing the title as "Send email without CSV", write it as "How to send email without using CSVs?". A user is more likely to search the latter and hence easier to find it as well.

Having a detailed knowledge base is the first step but making it accessible to your users is as important.

Link your knowledge base wherever necessary i.e. within products, emails, on your website, and train your agents to share relevant links with users so they're aware of the knowledge base.

A direct indicator of the fact that your knowledge base or help docs need to be updated is to look at the queries the bot wasn't able to answer.

Go to the Monitor dashboard and check out Open queries. These are the queries that haven't been answered by the bot.

You can make it a point to check these open queries every day to see which questions is the bot not able to solve and update your knowledge base based on exactly how the question has been asked.

Downvotes are an indicator that the bot isn't able to give accurate answers. So, it's naturally desirable to have as little downvotes as possible.

You can view this from the Analytics dashboard. Select the time frame to see how many answers were downvoted in that period.

A high downvote percentage is an indicator that the bot isn't understanding the product or service well enough to find solutions. You can take this as a way to upgrade and improve your knowledge base.

Response rate is the measure of the percentage of queries that the bot has been able to answer correctly. A higher response rate indicates that the bot has been able to understand and answer most of the queries posted by agents or customers.

You can view this from the Analytics dashboard.

A lower response rate means there are gaps in the knowledge base and it needs to be updated as per the questions that the bot hasn't been able to answer.

Threado AI Agent Assist helps your agents solve tickets faster.

You can take a look at agent engagement from the Analytics dashboard and compare how many tickets have been assisted by AI.

A smaller ratio of tickets resolved to tickets resolved by AI shows that the bot is able to answer customer concerns more accurately and quickly.

A higher ratio could mean that perhaps there are gaps in what the bot knows about the product, meaning you can use this to improve your knowledge base.

You can leverage the bot to figure out gaps in your knowledge base but you also need to retrain the bot on your updated knowledge base for it to answer queries accurately.

Whichever data source has been updated, go to the Train dashboard and retrain the bot.

For example - if new URLs have been added or existing URLs have been updated, you need to add those URLs for the bot to be trained on them.

📌 Note - For PDFs and CSVs, if the file has been updated, you need to delete and add the updated file. For Slack and Discord channels, edit which channels you want the bot to be trained on. In the case of integrations like Zendesk, Freshdesk, etc., the bot needs to be trained on updated or new tickets manually. Learn more about training the bot on different sources.