Great AI support starts with great documentation. To train Fin effectively, you need more than just a Help Center; you need a living, evolving knowledge management system that keeps pace with your business and meets customer expectations for fast, reliable answers.
At Fin, we’ve refined our approach to knowledge management through hands-on experience, training Fin to achieve a 97% automation rate for informational queries which are directly served through our content. In this guide, we share the best practices and workstreams we’ve developed to help you implement a knowledge system that delivers results for Fin, your team, and your customers.
If you're just starting to build your knowledge base, we recommend reading this guide first. Once you're up and running, come back here for ongoing knowledge management tips.
Align knowledge with product changes
What we recommend:
Partner with your product team to establish a process for keeping support content in lockstep with product releases. This includes:
Asking product managers or engineers to share details of every product in build (these can be scrappy internal notes!)
Translating technical product details into customer-facing help center articles.
Updating screenshots, navigation paths, and embedded links whenever the product UI changes.
Running targeted audits for major updates to ensure all impacted knowledge is current.
Example:
Each week we ship 10-15 product releases which involves creating and updating around 90 articles. We use Operator to audit our knowledge base and propose updates which we can review and publish for each release in less than an hour. A significant or widespread product release may require 2-3 hours to update relevant content across internal and external sources.
Use Fin's content recommendations
What we recommend:
Fin will surface recommendations when it identifies content gaps based on conversations that required escalation. These may include article edits, removing duplicates, fixing contradictions, or adding entirely new articles/snippets from answers the support team provided. These are highly actionable suggestions which require minimal effort.
We recommend reviewing these suggestions weekly to decide whether to accept, revise, or reject them, then updating your content accordingly.
Example:
A weekly review might yield 10-15 Fin suggestions. Most are small updates that can be implemented in an hour.
Enable your support team to flag issues
What we recommend:
Encourage your team to flag content gaps or errors they encounter in the course of helping customers. This could include:
Fin giving an incorrect or unhelpful response.
Customers being unable to self-serve an issue.
Outdated screenshots or information in articles.
Set up a simple submission process (e.g. a ticket form) and set aside time to action these improvements weekly.
Example:
At Fin, support teammates often surface 15-20 content improvements per week. Each suggestion typically takes 15 minutes to action.
Fix underperforming content and optimize for AI
What we recommend:
Even if your content is factually correct and up to date, it may still perform poorly when used by Fin. Unlike human readers, AI relies heavily on clear structure, unambiguous phrasing, and strong alignment with customer intent. Monitor Fin’s resolution rate with each piece of content to spot frequently used but underperforming content (these are prime candidates for improvement, and small changes here can have a big impact).
Focus on the top 20% of content by Fin involvement rate.
Flag items with a resolution rate below 90% as candidates for optimization.
Ask Operator to review and optimize each piece of content against a content readiness checklist so it better aligns with the correct topic, uses common customer phrasing, removes ambiguity, and is easier for AI agents to parse.
Example:
Every month we identify 10 articles with a high involvement rate but low resolution rate. Each one typically takes 5 minutes to restructure and publish an optimized version using Operator.
Capture expected product behavior
What we recommend:
Sometimes a customer question uncovers a behavior that seems confusing or unexpected at first, but is actually how the product is designed to work. When this happens:
Ask your support team to send the customer a macro which explains the product behavior and tags the conversation (e.g.
expected-product-behavior).Ask Operator to review all conversations with that tag from the past month to find the product behaviors, then check if they're documented in your knowledge base.
Once Operator has updated content with all missing product behaviors, ask it to tag those conversations (e.g.
product-behavior-documented).
Example:
At Fin, around 20 undocumented product behaviors are tagged in conversations each month. Reviewing these conversations and capturing the product behavior in our knowledge base takes less than an hour with Operator and dramatically improves Fin’s ability to handle nuanced customer queries without escalation.
This is our Operator prompt:
There are [X number] conversations tagged with expected-product-behavior but NOT product-behavior-documented. Find and read them in full to extract the product behaviors and check the knowledge base for each one. If a product behavior is undocumented, propose updates to relevant content to capture and explain the behavior. Fetch the full content of the most relevant articles to verify what's already covered before proposing updates.
Track and prioritize knowledge work
What we recommend:
Use a task management tool (e.g. Superhuman Docs, Trello, Asana) to prioritize and track:
Product updates and content required.
Incoming content suggestions/improvements.
Status of article audits, updates, or new content creation.
This helps you collaborate and share knowledge tasks, ensuring you have a clear overview of ownership and progress made.
Summary: Best practices for training Fin with great knowledge
Area | Key Activity | Recommended Frequency | Average Time Required |
Product Updates | Write/update content | Weekly | 30 hours |
AI Suggestions | Review and action AI content suggestions | Weekly | 1 hour |
Human Feedback | Review and act on teammate suggestions | Weekly | 5 hours |
AI Optimization | Optimize content for AI (using AI!) | Monthly | 1 hour |
Document Product Behavior | Add explanations for identified expected behavior | Monthly | 1 hour |
Knowledge Tracking | Maintain a task board or tracking system for collaboration | Continuous |
|
When these workstreams are in motion, knowledge management becomes a dynamic, shared practice that continuously evolves to meet the needs of your team and your customers. With this strong foundation in place, Fin is equipped to deliver exceptional support at scale.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts