TravelX

Where they were

TravelX helps airlines recover revenue from unsold seats: they buy back passenger tickets and resell them to last-minute travelers. The central business decision is knowing which flights will fill up and which won’t. When they came to us, that prediction depended on manual heuristics. They had the data and a growing engineering team, but no data science capability to turn it into something smarter.

What we found

The heuristics couldn’t tell the difference between a flight that would barely fill and one that would overfill. That meant missed opportunities on one end and wasted offers on the other. The data to build a real model existed, but nobody internally had the expertise to do it.

What we built

A classification model that predicts flight fill probability, cutting prediction error by more than half compared to the previous approach. The model runs in a batch pipeline that processes ~400 flights per day, giving the operations team fresh predictions each morning. We also ran knowledge transfer sessions covering ML fundamentals (model evaluation, calibration, performance metrics) so the team could interpret the results and maintain the model going forward.

What they can do now

TravelX makes inventory decisions based on model predictions instead of rules of thumb. The team understands what the model measures and why. They can evaluate its performance and detect when it drifts.

“We didn’t need a team that sells AI smoke and mirrors. We needed someone who looks at our data and tells us what actually works.” • Andrés Bruzzo, TravelX

Why they asked for more

The first model worked so well that Andy asked for a second: a per-passenger pricing model to optimize buyback conversion. Two months, one ML Engineer, and TravelX went from heuristics to a production model that runs their core business decision. Not every engagement needs to be a multi-year partnership.