Lifecycle Push Notifications for Lyft Driver Retention
Bringing back drivers who stopped driving on weekends
TL;DR
- Problem
- Drivers who earned at least $100 one weekend often didn't return the next, leaving fewer drivers for peak demand.
- What we did
- We tested two push messages built on each driver's own earnings, with no bonus attached, against a no-message group.
- Result
- The near-miss message showed the larger lift, +0.72% driver hours, though neither message reached statistical significance; the program review estimated $5.2 million in annualized impact.
- Industry
- Rideshare
- Timeline
- February–March test
- Engagement
- Prior in-house work at Lyft
- Tools & channels
- Push notificationsA/B testing
+0.72%
Driver hours, near-miss message
+0.56%
Driver hours, earnings reminder
$5.2M
Estimated annualized impact
Drivers were not returning the next weekend
Some drivers earned at least $100 in a weekend, then didn't return the next weekend. That left fewer drivers available during an important demand period. The question was whether a timely message could bring them back without paying for it.
Frame the same earnings two ways, with no incentive
Eligible drivers were split three ways: 34% received no message, 33% got a reminder of what they had earned, and 33% saw how close they had come to a higher earnings amount. Both messages used the driver's own earnings and carried no promotional payment, so the test isolated the framing.
Two framings compared with no message
| Group | Share | Driver hours | Rides |
|---|---|---|---|
| No message | 34% | Baseline | Baseline |
| Earnings reminder | 33% | +0.56% (p=0.176) | +0.51% (p=0.26) |
| Near-miss framing | 33% | +0.72% (p=0.080) | +0.45% (p=0.32) |
How a message is framed changes behavior
Drivers responded to a message with no money attached, and the near-miss framing did slightly better than a plain reminder. The review also found that sending these by hand wouldn't scale, which pointed the next work at event-triggered infrastructure for personalized messages.
How to read these numbers
Neither message met a conventional p<0.05 threshold, and the test did not show a significant difference between the two messages. The $5.2 million figure is a projected annualized impact from a later program review, not measured annual revenue.
Services used:Customer engagement and win-backLifecycle email and newsletters
