INSEAD Lunchtime Talks - How Lazada uses Data

Eugene Yan, VP of Data Science at Lazada, spent a lunch hour at INSEAD in April 2018 doing something rarer than another AI hype talk: explaining what machine learning actually does inside a real online marketplace, line item by line item. About 100 students showed up despite the midday slot, which says something about how hungry business students are for concrete examples instead of buzzwords.

Yan picked two case studies and went deep instead of wide. The first was automated classification of user reviews, a project that trimmed the manpower needed for that task by more than 90 percent and saved the company five figures a month. The second was product ranking, the unglamorous plumbing that decides what shows up first when you search for, say, a rice cooker. Tweaking that ranking logic lifted conversion rates by 3 to 8 percent and pushed revenue up somewhere between 5 and 20 percent. Neither example involves anything resembling a chatbot or a flashy demo. Both involve a lot of careful measurement and iteration on systems most shoppers never notice.

The rest of the talk was less about Lazada and more about Yan himself, and he was upfront that his career path was idiosyncratic rather than a template. He walked through what a typical day looks like for a working data scientist, which is apparently less glamorous than job postings suggest, and he pointed students toward free MOOCs he'd actually completed, plus side projects and volunteer work with DataKind as ways to build skills without a formal program.

What stuck with him afterward wasn't the food, though he made a point of calling the cafeteria lunch scrumptious. It was the students who tagged along to keep asking questions, not about becoming data scientists themselves, but about how to use data thinking as future analysts or managers. That's arguably the more interesting audience: people who won't build the models but will have to make decisions based on them.

Yan wrapped the whole thing into a short recap on his own site, mostly as a record and a citation point, slides included. It reads less like a victory lap and more like someone quietly grateful that a room full of business students wanted the boring, useful version of data science instead of the sci-fi one.