A one-person data-science function at TravelBird

TravelBird was a European travel e-commerce company that curated and sold trips. When I joined in 2016, data science was effectively a one-person function — me — expected to make an impact across every revenue-critical part of the business: marketing attribution, demand forecasting, portfolio strategy, and recommendations.

The challenge was breadth with real stakes. Each of these touched profitability or planning directly, and there was no team to specialise — so the work had to be pragmatic, shipped, and worth more than it cost.

I operated as a data-science generalist, picking the problems where a model would move a real number and shipping them end to end. The work that mattered most:

• A Markov-chain marketing-attribution model that reallocated spend toward the touchpoints that actually converted.
• A 12-month demand-forecasting system that moved the sales team from reactive sourcing to proactive planning.
• A portfolio analysis — leading a small team of analysts — that pruned the catalogue without losing revenue.
• A fix for the cold-start problem: replacing interaction-dependent recommendations with deep-neural-network representations built from item content and imagery.

Doing all of it solo taught me to be ruthless about where a model is actually worth building — the instinct I still bring to R&D.

The attribution model was the standout: reallocating marketing spend toward the touchpoints that truly converted added roughly €5M a year in net profitability. The portfolio work cut the catalogue by ~40% while keeping net revenue flat, and the deep-learning recommendations solved cold-start for ~40% of the catalogue that interaction-based methods couldn't reach. The forecasting system changed how the commercial team planned — from reacting to sourcing ahead.

It was my first experience owning data science end to end across a whole business — the foundation the later team-building and product roles were built on.