We removed our best AI feature. Retention doubled.

Last quarter we had an AI feature that could predict, with surprising accuracy, when a user was about to stop using the app. It would send a perfectly timed push notification with a personalized prompt right before they were likely to drop off.

Engagement metrics looked incredible. Open rates were 3x our average. Session frequency went up.

But something felt off.

When we talked to users, we kept hearing the same thing: "I feel like the app knows when I'm avoiding it." One user said it felt like "a therapist who texts you the moment you cancel an appointment."

We were using AI to prevent churn. But we were also teaching users that they couldn't have a quiet day without being nudged. We were solving a business problem by making an emotional product feel transactional.

So we killed it.

Instead, we replaced it with something much simpler: when users came back after a gap, the app acknowledged the gap without judgment. "It's been a few days. No need to catch up. Start wherever you are."

No prediction. No nudge. Just patience.

Retention over 30 days doubled. Not because we were bringing people back more aggressively, but because the people who came back felt safe doing so. They weren't returning out of guilt from a notification. They were returning because the space was still there, unchanged, waiting without pressure.

The lesson: sometimes the most effective growth strategy is giving people permission to leave. Products that feel safe to step away from are products people actually want to return to.

Anyone else found that removing a high-performing feature actually improved the overall experience?