After ixigo relaunched as an AI-native app built around TARA, flight search-to-booking conversion on Android fell 8%. Why?

Introduction

In May 2026, ixigo relaunched its app as "AI-native". TARA, its multimodal assistant, became the centre of the experience: users can search by voice, text or taps in English, Hindi or Hinglish instead of relying mainly on filters and search bars. The relaunch also brought a personalised homepage, Trip Mode, agentic flows and a redesigned interface with dark mode.

In this scenario, after the relaunch, flight search-to-booking conversion on Android fell 8%. iOS is roughly flat. A big redesign changes many things at once, so the challenge is to find which change caused the drop, and whether it's temporary (users learning a new interface) or structural. I'll clarify, rule out external factors, break down the funnel, form hypotheses and test them.

Step 1

Clarifying Questions (3 minutes)

• Was the relaunch rolled out gradually or to everyone at once?

• Why it matters: A staged rollout gives a natural control group.
• Hypothetical answer: Staged over three weeks on Android by percentage of users; iOS went 100% on day one.
• Impact on approach: We can compare Android users on old and new versions during the rollout.

• How is "search" counted in the new app?

• Why it matters: In a conversational app, a single TARA conversation might trigger several searches. If each counts as a search, conversion per search falls even if bookings don't.
• Hypothetical answer: Each results page load counts as a search, including those triggered by TARA.
• Impact on approach: We need to check total bookings and conversion per user as well.

• Have total flight bookings fallen?

• Why it matters: Separates measurement issues from real loss.
• Hypothetical answer: Android flight bookings are down about 4% compared with the expected trend.
• Impact on approach: Part of the drop is real, part may be measurement.

• Which Android users are most affected?

• Why it matters: Device and network differences point to performance problems.
• Hypothetical answer: Not yet segmented.
• Impact on approach: Segment by device class and network.