Closing the Loop on $483K
A feature shipped, and then everyone moved on.
Context Travel's booking flow includes a moment of friction: a client's preferred tour date isn't available in a small-group format. Rather than losing that booking outright, the platform surfaces the closest alternative dates where group availability exists: a design decision meant to keep the client in the funnel instead of bouncing to a competitor.
The feature shipped, worked as intended, and moved into the normal cadence of new work. It had been live for a while, but nobody had gone back to actually measure what it was doing once it met real clients making real decisions.
Closing the loop during cooldown.
During a cooldown period between projects, I went back to close that loop. A click on the alternative-date link converts at roughly 54 times the site's baseline rate, a strong enough signal on its own to justify pulling the order data and tracing those visits through to what people actually booked.
The interesting part wasn't just that people booked. About six in ten of them could have chosen the private-tour route instead, at roughly four times the price, and picked the group date the feature surfaced anyway. That's a direct signal the feature was doing its job: giving people a real alternative instead of losing them at a dead end.
"I'm big on capturing engineering proximity and real outcomes in my documentation. This was a clear case where I had the access and the reason to check."
No group tour existed on their original date at any price. Price wasn't the driver here; there was nothing to be price-sensitive about.
A private tour was bookable immediately on their original date, at roughly 4x the group price. They chose to wait for the cheaper group date instead.
Figures reflect actual completed bookings pulled directly from order data, not a forecast.
Most work gets judged on how it looks. This is judged on what it did.
Most of my work gets evaluated on whether it looks better or tests well. This is one of the few times I can point to a specific interaction pattern and trace it straight to revenue. No attribution modeling required, no assumptions about causation.
It's also proof that I don't consider a feature done just because it shipped. Going back to check whether something actually worked is part of the job, not extra credit.
"I found out a feature we'd shipped (one that shows clients alternative dates with group availability) had actually recovered $483K in revenue across 259 orders. Nobody had gone back to measure it, so I did."
Michaela Hoffman, Principal UX/UI Designer