In this deep-dive session, I'll share how we at Uber tackled a critical business challenge: losing millions in revenue due to undiscovered mobile app bugs that were killing our conversion rates. I'll reveal how we built DragonCrawl, an AI-powered testing system that helped us detect conversion-blocking issues before they hit production.

The problem was painfully familiar to anyone running a mobile app or e-commerce business: Users would start a purchase flow, hit an unexpected bug or UI glitch, and abandon the transaction entirely. Traditional testing couldn't catch many of these issues because they often appeared only in specific contexts - certain devices, languages, or locations. Manual testing was too slow and expensive to cover all scenarios.

Key Takeaways:

• How we increased conversion rates by using LLMs to identify user-blocking bugs before they reached production, including specific patterns of bugs that most commonly cause users to abandon transactions
• Practical techniques for training AI models to think like real users trying to complete purchases, including how we used actual user session data to teach our system to spot conversion-killing UX issues that traditional testing misses
• Real metrics from our production rollout showing how automated AI testing caught more conversion-impacting bugs compared to our previous testing suite, with specific examples of high-impact issues that would have cost millions in lost revenue