How Fixing Courier Forecasting Saved Glovo €30M a Year
How I identified a systemic problem across 8 markets, navigated human resistance, and scaled an ML-driven solution that transformed courier operations across South-Eastern Europe.
A Different Kind of Role
As the Supply Excellence Manager for the SEE region, my mandate wasn't just to "keep things running." It was to find the levers of efficiency that were invisible to the day-to-day operations teams. In a high-growth environment like Glovo, scaling often masks deep operational inefficiencies. My role was to look past the growth curves and find the technical fractures in our supply chain.
A Systemic Problem Nobody Had Named
Despite having a sophisticated Machine Learning model for forecasting courier demand, 80% of our managers were manually overriding the data. This human "gut feel" was creating massive oversupply or critical shortages, resulting in millions of euros in lost efficiency and frustrated couriers who weren't earning enough per hour.
The Friction Point
Managers didn't trust the model because they couldn't see the "why." They preferred their Excel sheets over the black box, even when the data proved them wrong.
The Approach
Understand Before Fixing
I spent two weeks shadowing local operations managers in Bucharest, listening to why they feared the ML model. I realized the problem wasn't technical; it was psychological.
Find the Champion
Instead of a top-down mandate, I found one skeptical manager in a mid-sized city who was willing to run a "Shadow A/B Test" for two weeks.
Let the Model Run
We locked the inputs for 14 days. No overrides. The results were so undeniable that the champion became my biggest advocate for the next 7 markets.
The Numbers That Made the Argument
One City. Then Fifty. Then Eight Countries.
Scaling wasn't a matter of pushing a button. It was a regional roadshow. I developed a rollout playbook that paired data scientists with local ops leads, ensuring that as we moved from Romania to Bulgaria, Serbia, and Croatia, the local nuances were respected while the ML integrity remained absolute.
From SEE to the Entire Organization
What started as a regional project was eventually adopted as the global gold standard for Glovo. By proving that Machine Learning could outperform human intuition when properly managed, we unlocked a scalable blueprint for efficiency across 20+ countries.
The Hard Problem Was Human
Find the one person willing to go first
Mass adoption is an outcome, not a strategy. Real change requires finding the smallest unit of proof and scaling the success stories, not the instructions.
Let the results do the persuading
Data doesn't change minds; results change behavior. When a manager sees their KPIs turn green without the 2 AM manual adjustments, the product sells itself.
"Change at scale almost always starts with one person in one city who is willing to go first."