How PostHog Used Inkeep to Instantly Resolve 1/3 of All Community Questions (In a Developer-Friendly Way)

TL;DR

PostHog is the all-in-one platform that helps developers build successful products. With 190,000+ customers and 97% word-of-mouth growth, their success is built on a simple principle: create tools developers love to use.

1. Prior to Inkeep, PostHog's technical community forum often saw delays in replying to nuanced developer questions. Their users could wait hours or days for human responses. And their engineers had to split time between building products and supporting customers.

2. Cory & the PostHog team wanted to explore integrating AI into their website to automate community support – but without the traditional "chatbot feel".

3. PostHog used Inkeep to launch the initial version of Max AI, an intelligent user assistant that would answer customer questions in PostHog's community forum.

4. After initial success with a 33% auto-resolution rate, PostHog expanded the use of Inkeep to their website, internal teams, and within their user-facing product.

Problem: Answering Technical Customer Questions While Maintaining Quality

PostHog had seen massive growth and faced a customer support scaling challenge: their community forums were growing rapidly, and engineers were struggling to respond promptly due to the volume of questions.

Community questions competed for attention with support tickets, and the queue often became noisy with repetitive inquiries already answered in the docs or in other threads.

But with a technical user base, PostHog was concerned their audience would instantly write off a chatbot due to the stigma that automated responses tend to be unhelpful or inaccurate.

For this reason, Cory & the PostHog team looked for an AI assistant that could be implemented without the traditional chatbot experience. So it was extremely important to the PostHog team find a format that customers would appreciate while also giving high quality outputs to maintain trust with their customers.

"When you see an AI chatbot, you just assume, 'Oh, this is gonna be a terrible experience' because most of them are."

Solution: A Customizable Platform to Build Delightful Customer Experiences, Starting Directly In The Community Forum

PostHog wanted to build a unique solution for answering community questions that would scale with their growing user base. This meant a customizable platform where they could implement their own UI while relying on the AI to provide high quality responses.

Inkeep was the perfect solution because of the customizable nature of Inkeep's AI user assistant via rules, instructions, outputs, and interface.

As an initial starting point, PostHog integrated Inkeep in their community forums (as opposed to adding a chatbot on their homepage). That's because users posting to a community forum don't have the expectation of an instant response from humans. This way, they could show an answer immediately – but only if AI was highly confident in the quality of response.

This low-risk, high-reward strategy meant:

1. Customers wouldn't expect an instant response due to the nature of a community forum interface
2. Real-world testing with actual user questions
3. User delight when the AI delivered good answers quickly

By using Inkeep, the PostHog team deeply customized a customer experience that adds value to users in the community forum.

This did this by:

1. Having Inkeep's LLM index PostHog's docs, tutorials, blog, handbooks, GitHub repos, support threads, feature requests, and bug tickets within Max AI.
2. Then, as an initial LLM output, there's a "confidence score" not visible to users.
3. If the "confidence score" is high enough, Max AI would show the answer to the user.

"If Inkeep is highly confident in the quality of response, we show the answer as a reply in the community thread immediately. If not, we simply don't post the AI response at all. In a community forum, this works great because people don't have an expectation of an instant reply. So when we are able to serve a relevant answer before they even have a chance to leave the page, it's an opportunity for us to delight the customer right away. This reduces the brand risk of unleashing a poorly-tuned AI bot that annoys users with inaccurate or unhelpful responses."

Results: 33% Auto-Resolution Rate

6 months later, the numbers show:

• 759 total resolved threads in their forums
• 247 resolved by "Max AI" (their branded Inkeep assistant)
• 33% auto-resolution rate

But the real win wasn't just support deflection. As Cory explained: "To me, each one of these questions is an opportunity for our brand to spark joy with someone."

And even when AI can't instantly solve the question – or when the customer provides feedback that the answer wasn't helpful – that feedback is routed to the Docs Team.

The Docs Team then reviews the question and looks to see if the documentation needs to be updated. This is often the case for a multi-product company with 10+ products constantly under development.

Beyond Forums: Expansion To New Use Cases & Teams

Once the forum implementation proved successful, PostHog expanded Inkeep across their ecosystem to include:

• Website Search: Inkeep's chat interface provided direct answers, acting as an enhanced search tool that served custom responses sourced across multiple technical docs, while being seamlessly integrated into PostHog.com without a floating chat trigger icon.

• Internal Slack Bot: The "#ask-max" channel allowed employees to query documentation and their publicly-available internal company handbook.

• Product Integration: PostHog integrated Inkeep into their product's LLM, automatically routing documentation-related user questions directly to Inkeep's API.

The product integration in particular allows PostHog's AI team to spend more time building their core product instead of needing to tune AI to answer use case questions and provide customer support.

The Technical Implementation Details

If you're interested in learning more on this implementation, Cory from PostHog has an awesome article about it on their website. But at high level, here's what made PostHog's implementation successful:

Custom Ranking Signals as PostHog passes metadata to help Inkeep prioritize answers:

1. Moderator/employee responses get higher weight
2. AI-generated answers are excluded from training
3. Thread quality signals improve over time
4. Original posters can mark an answer as helpful or unhelpful, and the question is routed to humans for review when necessary

Knowledge Sources where Inkeep would ingest from:

1. Documentation and website
2. GitHub repos
3. Community forum threads

UI Customization so the team invested heavily in making Inkeep feel native:

1. AI presented as a forum reply rather than just a chatbot
2. Seamless integration with existing UI components
3. Custom loading animations
4. Brand-consistent styling

Lessons for Implementation Teams

PostHog's journey offers clear lessons for other accuracy-obsessed companies:

1. Start with low-stakes implementations - Forums were perfect for testing without risking the primary user experience
2. Invest in customization - Making AI feel native to your brand matters
3. Use AI to identify gaps - Failed queries highlight documentation needs
4. Think beyond deflection - Each interaction is a brand opportunity
5. Trust specialized tools - Building AI infrastructure doesn't need to be your core competency

What's Next

PostHog continues to explore new ways to leverage AI for customer support. While they haven't yet integrated with their Zendesk ticketing system (due to data privacy considerations), the success across forums, website, Slack, and product integrations shows the potential for AI-powered support done right.

See how PostHog's investment in AI customer support helped them resolve 1/3 of community questions instantly.