Juniper caught a retention drop before launch day
A guardrail metric stopped a popular redesign mid rollout, and a one click rollback saved most users from seeing it.

A redesign everyone loved
Juniper makes a habit tracking app. The team spent four months on a new home screen. Internal reviews were glowing, early user interviews were positive, and the launch was scheduled with a press push behind it.
How the rollout was set up
The redesign shipped behind a flag with a gradual rollout. Primary metrics were daily check ins and session length. Guardrails covered day 7 retention and crash rate, with automatic pause if either moved beyond its bound.
What the guardrail saw
At 10% of traffic, day 7 retention looked flat, which is expected because the metric takes a week to mature. Kerf uses a leading indicator for it, the share of users who return on day 2, and that dropped by 4% against holdout.
- The rollout paused itself at 10%.
- The owner was paged with the metric, the interval and linked replay clips.
- Replays showed users could not find the streak counter, which had moved into a menu.
Everyone in the room loved the new screen. The data did not, and the data was right.
Head of Design, Juniper
Results
The team rolled back in under 90 seconds, restored the streak counter to the home screen and relaunched two weeks later. The second version beat the old design on check ins.
Juniper now treats a guardrail breach as a normal part of shipping, not an incident.

“A guardrail caught a 4% retention drop on a launch everyone loved. We rolled it back in one click, before most users ever saw it.”
More customer stories

Northwind cut checkout drop-off by 18%
A retailer with 2,400 employees moved every checkout change behind a flag and a test, and found the fix was a smaller form, not a redesign.

Halcyon ships weekly instead of monthly
A health scheduling platform replaced a manual release process with gradual rollouts and automatic rollback.

Brightpath replaced three tools with one
An education platform consolidated flags, experiments and session replay, and the arguments about whose numbers were right stopped.