+3.7%
App conversion
+7K
Paid installs
20+
Customer interviews
Context
Atlassian Marketplace is a $6B+ ecosystem of 6K+ apps from 2K+ partners, used by 300K+ customers, from SMBs to large enterprises, to extend Jira, Confluence and other Atlassian products.
Problem
Marketplace growth had stalled. Customers struggled to find the right app among 6K+ listings, and many who found one dropped off before installing. Discovery relied on keyword search and editorial picks that ignored the customer's context.
My Role
Owned the growth strategy across the Marketplace funnel, from discovery to evaluation to installation. Used funnel analytics and 20+ customer interviews, with AI-assisted synthesis, to find the friction points and prioritize 5 ML/AI experiments. Worked with engineering, data science, design and partner teams.
What we found
Funnel metrics showed where customers dropped off; interviews explained why.
- Discovery: customers had to go looking for apps themselves, and the Marketplace had no context to suggest the ones they needed
- Evaluation: generic listings made it hard to judge whether an app fit their stack before trying it
- Installation: unclear listing details, such as trial terms and how to cancel, and a three-screen install flow caused drop-off
What we built
- Smarter discovery: find apps by the problem being solved, beyond keyword search, including new categories like AI agents
- Recommendations: ML-driven suggestions based on a customer's existing products and what similar teams installed
- Personalization: the most relevant apps first, based on each customer's products, team size and funnel stage
- Clearer listings: trial details, how to cancel and other key information up front
- One-click installation: cut from three screens to one click
- Sponsored ads: placements that help partners reach customers looking for apps like theirs
Decisions & Trade-offs
Parallel experiments vs. sequential
Ran all 5 experiments in parallel rather than one at a time. It took more coordination, but we learned much faster.
Metrics & Results
- +3.7% app conversion within 6 months
- +7K paid installs within 6 months
Links
Learnings
- Let customer friction set the strategy. Aiming each ML experiment at a specific friction point, found through data and interviews, is what turned models into conversion gains.
- Parallel experiments compound. Testing together surfaced interaction effects between experiments that one-at-a-time tests would have missed.
- Marketplaces grow as flywheels. Better discovery → more installs → happier partners → more apps → better discovery.