Fin AI Agent uses machine learning to source answers from any support content, whether that’s Intercom Articles, Zendesk Help Center, or any public URL you provide. Before giving Fin content to digest, it’s worth taking some practical steps to ensure your content is optimized.
How did we prepare at Intercom?
Before we started using Fin AI Agent with our customers we took a few steps to ensure that our content was in top shape.
1. We moved customer friendly content from internal resources into our Help Center
Our internal knowledge base Guru contains a lot of sensitive product information, however there was a collection of troubleshooting guides in there that we knew would be useful to our customers.
Since we already had this content drafted, it made sense to give it to Fin AI Agent. Now customers with complex issues can troubleshoot with Fin, or at least self-diagnose their issue while they wait for a teammate.
2. We audited all our articles
To ensure that Fin AI Agent isn’t learning from out of date content, we needed to ensure that every live article was accurate and up to date. To do this we divided our articles up into product areas and gave the relevant team a week to check, update or retire each article. We have over 700 live articles, so this was a real team effort.
3. We tested Fin AI Agent for a week with a segment of our customers
We then analyzed what Fin AI Agent wasn’t able to answer and created new content to fill these gaps. Using Fin to work backwards to identify your gaps is a useful way to get insights, especially if you don’t have capacity to overhaul your articles or create a lot of new ones.
We recommend starting with a smaller subset of customers so you are able to monitor and review Fin's interactions and work on updating resources from there.
Enable Fin AI Agent on 2 different smaller groups - power users and customers only starting to use your product. This was very insightful for us in terms of customer wording and glossary of terms they use to ask the same questions.
What else can you do?
You can use our above steps as a guide to apply to your own content, but beyond those ideas here are some other steps you should consider:
Start with top performing articles
Visit your articles report, and look at your top articles, i.e. the ones with the most traffic and engagement. Employ the 80/20 rule here and ensure that your most popular articles are up to date.
Identify gaps in your content
Use your articles report and look at the searches with no results tab. Use this to ensure you’re not missing any content.
Talk to your support and sales teams and other stakeholders and get them to inform on important articles you may be missing.
Consider creating articles for topics that fall outside your help centre searches too. If it’s a frequently asked question, there’s no reason to hold back.
Think about Fin AI Agent as a new teammate
Ask yourself, "Hey I’ve just got an extra person added to my team, what type of queries do I want them to focus on?" For example you might want them to be able to fully resolve the top 10 most common queries your team gets.
If you want Fin to answer those top 10 questions, you need to ensure the help content it has access to provides all the relevant information.
Use the Fin AI Agent answer debugger
The Fin AI Agent answer debugger has been designed specifically to help you understand how Fin AI Agent generated an answer using your support content. It's available on each AI-generated answer via the Help Desk and you can improve how Fin handles support queries in the future by:
Creating new content, or
Updating existing content.
Use this one trick
Actually, there's no magic trick - but what we do know is that well written, well structured content and lots of it will win out.
Keep things simple, write in your customers’ language and make content easy to scan:
Explain what’s visible in screenshots or GIFs (Fin can only use text).
Large blocks of text will make your customers’ eyes glaze over.
Help people scan for the information they need in your articles by using tables, numbered lists, and bullet points to explain tricky concepts and break up space.
Finally, remember that a lot of customers will still search and read your articles organically. As long as you focus on writing high quality, clear and helpful content, you can be sure your customers are getting the best help wherever they look for it.
See examples of AI support content Best Practices.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts