PostHog · AI product · Senior · 45 to 60 minutes

Which AI problem should PostHog solve first?

PostHog is a product analytics platform that helps software teams understand user behavior and improve their products. Answer this product management case and get it graded line by line against a weighted rubric.

AI product strategy
Prioritization
Risk management
Metrics design
Systems thinking

The situation

Your VP Product has asked you to recommend how PostHog should approach AI-powered features across its product suite. Customer interest in AI is rising, competitors are adding AI labels to many products, and several internal teams have proposed ideas such as natural language help, automatic analysis, and in-product assistants. Leadership wants to move quickly, but there is concern that a generic AI layer would create demos more than durable value. There is also risk that weak outputs, missing product context, or unreliable behavior could reduce trust in a tool that teams use to make product decisions. You need to propose where to place the first serious AI bet, how narrowly to define the initial problem, and how to build it in a way that is useful even when the model is imperfect. Assume the company can fund one major AI initiative over the next two quarters, but not a broad platform program plus multiple end-user features. Your recommendation should cover what to build first, what product context or state the AI would need, how to handle failure cases, and how PostHog would know whether the feature is actually working after launch.

What to produce

A 2 page product strategy memo.

Your answer must

Choose one primary AI use case to pursue first, and explain why the user problem is specific enough to solve well and valuable enough to matter.
Describe the minimum product context and state the AI would need inside PostHog to generate useful outputs, and how that shapes the user experience.
Outline a launch approach that includes key failure modes, user trust guardrails, and what the product should do when the AI is wrong, unavailable, or low confidence.
Define success metrics for adoption, quality, and business impact, plus a post-launch monitoring plan that would tell you whether to iterate, expand, or stop.
Address sequencing and second-order effects, including what you would explicitly not build yet and why.

Constraints

Stay within PostHog's existing product ecosystem, do not propose a separate standalone consumer AI product.
Do not assume model outputs are consistently correct, the feature must remain safe and understandable under imperfect performance.
You can recommend only one major AI initiative as the first bet over the next two quarters.
Your proposal should be conservative about data use and customer trust.

Format

Use clear section headers such as recommendation, target problem, context needed, launch and guardrails, metrics, and sequencing.

How PostHog actually played it

The post follows PostHog's AI feature work in three stages: choosing what to build, implementing it and improving it. It stresses validating that a problem is specific, using app context and state, planning for failure and monitoring effectiveness. A practical framework for scoping AI features.

Validate that the problem is specific and valuable before building
Plan for failure with guardrails and monitoring
Keep measuring effectiveness after launch

Built from What we've learned about building AI-powered features by Ian Vanagas.