“We ABC tested these pages against each other and none of the alternative versions won against the control version. And so yes, that's the moment that we all agreed to listen to our gut feeling — the control version was not conveying the messages that we wanted to convey and was not user-friendly at all. We implemented the branded version, and a few weeks after we saw positive results.”

When Every A/B Variant Loses, Ship the One That Serves the User Best

Deezer redesigned their logged-out homepage and ran a three-way A/B test — none of the new variants beat the control statistically. Rather than staying put, the team listened to a qualitative signal: the existing page didn't represent the brand or serve users well. They shipped the branded variant anyway. Within weeks, it outperformed the old control. Data tests can fail to detect real improvements, especially for brand-driven UX changes that take time to compound.