PostHog · AI product · Senior · 45 to 60 minutes

Should PostHog use AI to fix onboarding?

PostHog is a product analytics and developer tools platform that helps teams understand user behavior and build better products. Answer this product management case and get it graded line by line against a weighted rubric.

AI product strategy
Onboarding optimization
Experiment design
Systems thinking
Prioritization

The situation

The VP of Growth asks you, a senior PM, to recommend how PostHog should improve new-user onboarding. Teams are signing up because they want product analytics and related tools, but many stall before they see value, especially during setup and first-use steps that require product knowledge, implementation choices, and navigating fast-changing documentation. The growth team has collected a friction log from failed onboarding sessions, and it suggests repeated patterns of confusion rather than a single broken screen. At the same time, AI onboarding assistants are getting attention internally, but there is skepticism. A brittle assistant could give wrong setup advice, drift out of date as docs and product flows change, and create support burden or trust issues. Assume sign-up to successful onboarding conversion is 8 percent today, and only 25 percent of converted users reach a meaningful activation milestone within 14 days. Leadership wants a recommendation this quarter on whether AI should be part of the answer, for whom, and how to de-risk it without slowing core product work.

What to produce

Write a 1 to 2 page product strategy memo.

Your answer must

Define the user problem using the friction log, including the most likely failure modes in onboarding and the evidence you would validate first.
Recommend whether PostHog should invest in an AI-assisted onboarding approach now, and clearly identify the target user segment and job to be done.
Describe the product approach at a level that covers where the assistant would get its knowledge or context, how it would stay accurate as the product changes, and what should not be handled by AI.
Lay out a sequencing plan for the first two releases, including what you would launch first, what you would defer, and why.
Propose a success framework with leading and lagging metrics, guardrails, and an experiment design that can distinguish better onboarding from novelty effects.
Address second-order effects on support, trust, maintenance cost, and the ability to reuse the work for other AI capabilities later.

Constraints

You have one quarter to show a credible signal before asking for a larger investment.
The team available is one PM, one designer, and up to four engineers shared with growth.
Any solution must minimize manual upkeep of product knowledge because docs and setup flows change often.
You cannot accept a noticeable increase in incorrect setup guidance or unsafe exposure of customer data.

Format

Use short sections with headings: problem, recommendation, product approach, sequencing, metrics, risks.

How PostHog actually played it

PostHog traces its AI onboarding wizard from a growth team friction log of failed onboarding sessions to a 5x conversion and 2x activation gain. The lessons are about feeding context rather than hardcoding knowledge and fixing context drift at the source. A rare AI product case with outcome metrics.

Start AI features from observed user friction, not from the technology
Do not hardcode knowledge that can drift, fetch it as context
Reusable context lets you compose new skills cheaply

Built from We used context engineering to 5x conversion and 2x activation by Edwin Lim.