Careem Ride Booking Flow: Rebuilding Conversion with ML Prediction
Redesigned a seven-step ride booking flow using ML-based prediction, cutting steps for the majority of users and lifting conversion.
Amit Ranjan
Co-Founder at Potli
From their time as
Lead Product Manager- Rides
Careem • 2020 - 2021
Overview
Amit was leading product for Careem's core ride-hailing vertical when COVID hit and the business effectively stopped. The P&L collapsed, and with it, the usual constraints on what could be changed.
The Story
Amit was leading product for Careem's core ride-hailing vertical when COVID hit and the business effectively stopped. The P&L collapsed, and with it, the usual constraints on what could be changed.
The ride booking flow had accumulated complexity over years of iteration. It ran to seven steps, was embedded in users' mental models, and had proven difficult to change under normal operating conditions. But it also had a structural problem: roughly half of users who entered the flow did not complete a booking. The click-to-ride rate had hovered between 0.48 and 0.52 for an extended period.
Amit saw the market pause as an opening to do the structural work that would have been too disruptive to attempt otherwise. The team cleared the roadmap and focused on the booking flow.
The redesign replaced the step-by-step approach with ML-based prediction. Rather than asking users for inputs at each stage, the system used past ride data to predict what they needed, reducing the flow to three steps for roughly 97% of use cases. New users or unusual trips still went through the full flow, but the majority experienced a significantly leaner path.
The click-to-ride rate moved from the 0.48 to 0.52 range to 0.56 to 0.58. At the scale of Careem's ride business, each percentage point of improvement represented $40 to $50 million in top-line impact.