Building a Fin AI friendly knowledge hub

Hi Cassandre, do you already have a help centre in place or are you preparing to create one?

On the basis you already have one, my recommendation would be to start with the problem/s you are trying to solve. E.g. are you finding Fin is giving incorrect answers to customers? Or hallucinating too often? Or not giving useful answers? Once you have some examples, you can then work out the best way to address them.

That’s how we are ensuring we keep Fin as helpful to our customers as we can - constantly monitor the answers it is giving and make tweaks either to the content in the help centre or to its guidance, to improve those answers for next time.

One useful part of this is to create a process that helps you review Fin’s answers. E.g. create a workflow that assigns conversations that only Fin has been involved in to your support queue for review, or maybe a slot in the week where you go into the Fin queue and review the chats.

Good luck!

Hey ​@Cassandre Laurens Fonseca !

I recently finished optimizing our Knowledge Hub and I’m happy to share what’s been working well for us! For a little background, my organization is a property management platform that serves multiple customer types (Retail, Bulk, Investors, and Residents). I’ve focused on keeping everything cohesive, and our Knowledge Hub is structured similarly to our Help Centers to support that.

Here’s what I did before rebuilding the Knowledge Hub (This played a major role in deciding how to format the KH)

• Each customer type is set up as its own brand within Intercom. I really like utilizing brands within our workflows to keep the routing clean and ensures conversations route to the correct team.
• Each brand has its own Help Center, so the Messenger only shows help content relevant to that customer.
• Within each Help Center, we have collections for each product or service.
  • One of our customer types has an all-inclusive Help Center with all content (since you can assign articles to multiple Help Centers, this works well).
• We also use audiences to assign content based on customer type. This ensures Fin pulls from the right information. For content that applies to multiple groups, we simply assign multiple audiences.

Knowledge Hub Structure:

> Main folders for each customer type
> Subfolders for each product or service (closely mirroring Help Center collections)
> Articles and snippets are organized within these subfolders

This setup has made content updates extremely efficient. It’s easy to locate applicable content, make edits, and ensure everything stays aligned.

Example:

• Our Retail customer type has its own brand and Help Center.

• The Help Center contains collections for Product 1 and Product 2, each with relevant articles.

• All Retail articles are assigned with the Retail audience.

• In the Knowledge Hub, there’s a Retail folder with subfolders for Product 1 and Product 2, where all applicable articles and snippets live.

How I tackled the project efficiently:

1. Reviewed Help Center collections and reorganized them to ensure the structure made sense.

2. Created corresponding folders and subfolders in the Knowledge Hub to mirror the Help Centers.

3. Reviewed each article (yes, every article 😶) for accuracy, AI readability, and audience assignment. I also made sure every article was linked to the correct Knowledge Hub folder. Audience and folder assigned can be adjusted right in the article details side menu.

It was definitely a lot of work to go through everything manually, but completely worth it. The end result was a well-organized, up to date Knowledge Hub that's easy to maintain and directly improved Fin AI’s resolution rate. For ongoing maintenance, I have a monthly routine for reviewing Fin’s metrics and auditing conversations. I also have close contact with Product and Support Teams to ensure our content is updated when there is a change in product or a new release.

We aren’t a B2B company, but wanted to share what’s been working for us :) 

 Hi ​@Cassandre Laurens Fonseca,

A couple of questions first:

• Are you using Intercom to host your help center/knowledge base, or another provider?
• How much content do you currently have documented for Fin to use?

Next, some learning opportunities provided by Intercom:

• There’s an upcoming webinar on this exact subject (on July 10th), register here.
• This “Getting Started with Knowledge Hub” course in Intercom Academy covers how to “Optimize your help content for Fin, teammates & customers”

Finally, some direct answers:

Do you have any recommendations or best practices on how to structure a Knowledge Hub for a B2B company?

• The best way to organize content by audience (prospects, clients, partners, internal teams).

There are a lot of ways to tackle this. I prefer to write articles and organize help centers around specific features or offerings of your produce/service. Most of the time, a customer is seeking an answer to something blocking them when they’re on your help center. If there’s an article directly speaking to their problem, then it doesn’t matter if they’re a partner, user, or prospect.

• Formats and navigation that make information easy to find and up-to-date.

Help articles should be very easy to navigate visually, so various sized headers, numbered lists, bulleted lists, graphics, videos, etc.

• Tools or workflows you’ve seen work well for maintaining and scaling a B2B knowledge base.

I’ll keep these recommendations within the Intercom ecosystem:

• Review all negative Fin CSAT conversations for poor responses and content gaps
• Review as many negative CX score conversations poor responses and content gaps
• Have your human support teams tag poor bot responses 
• Have your human support and other client-facing teams share common questions to inform new content creation
• Review all automatic Fin suggestions (DO NOT ACCEPT WITHOUT READING AND EDITING)
• Have a knowledge manager — you need someone who is responsible for your help content. They don’t need to write every single article, but they should be able to gauge the usefulness of current articles and identify what needs to be updated and what new content needs to be made.

Hey ​@Cassandre Laurens Fonseca ! I’m a knowledge manager at Intercom and you’ve been given some great advice from the community here 😍

Multi-brand is a useful way of separating help centers for different customer segments (prospects, clients, partners, internal teams) and targeting this content correctly with Fin, while managing everything from one workspace. However, you can use audience rules to achieve this as well if each segment is tracked in Intercom using a custom data attribute. Another helpful way of organizing your content internally is with tagging. This is quite flexible because one article can have multiple tags (belong to multiple customer segments/internal teams) but can only belong to one folder. So if you do have some shared content, filtering by a tag will help you find or update everything with that tag no matter where it lives.

As others have mentioned, I find it very helpful to align the Knowledge Hub structure with the Help Center, and ours is also organized by product area. This feels intuitive to navigate and related features are kept together which makes auditing easier as products evolve.

Maintaining and scaling a knowledge base is very much an ongoing and collaborative effort! In this guide, we share the best practices and workstreams we’ve developed to keep things ticking along nicely. 

Hope some of this helps on your knowledge management journey 🚗