AI Ride Matching That Learns Your Preferences on Rapido
An AI preference-matching layer for Rapido that learns what a rider wants instead of asking every trip
✳The problem they went after
Rapido riders care about air conditioning, driver language, vehicle condition and driver rating, and as the platform expands into autos and cabs they reapply those filters at every single booking. The repetition wears people down, rides still come back mismatched, and cancellations follow when the ride does not meet what the rider expected. The gap between expectation and default outcome costs satisfaction and retention.
✳What they actually did
Preferences get captured once across five categories: amenities, driver rating, language, vehicle condition and driver profile, with riders ranking up to five in priority order. Every available driver then gets a match score shown as a plain percentage, 92 percent or 75 percent, at the top of the booking suggestions. The weights keep adjusting based on which suggestions riders accept, post-ride ratings, and stated cancellation reasons. When supply is sparse, the flow falls back to the nearest available ride.
✳What you can take from it
Worth stealing
- Preferences captured once and ranked in priority order
- Match score surfaced as a plain percentage
- Weights learn from acceptances, ratings and cancellation reasons
- Nearest-ride fallback protects booking availability when supply thins
Read this if
Speccing a personalisation layer on top of a marketplace matching flow
The transferable bitA ranking layer that can starve the core booking flow is a worse product, so the spec caps the downside with a fallback rule before it talks about upside. The projected impact is also framed as a set that moves together rather than a single hero number: acceptance up 10 percent, satisfaction up 8 percent, cancellations down 12 percent, retention up 6 points.
✳The submission itself
Open the PDF →Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.
