From Director of Machine Learning at Booking.com to Head of Data at Pennylane

Hi there! My name is Fanny and I’m the Brand & Content Manager at Pennylane.

I’ve been working with Thomas for a few months now and always wondered: that job he had at Booking.com sounds fancy. Director of Machine Learning 🙌. So why on Earth did he leave it to join us, a barely one-year-old French startup 99% of people had never heard of, with a Data Team of just 1 person (no offense, back-then-Data-Team 🙏)?

A couple of weeks ago, I finally asked him. I thought you’d be interested in his answers too, so here they are. 👇

Hi Thomas! Thank you for taking the time to talk with me. First, can you tell me a little bit about your position at Booking?

Hi Fanny! I stayed around 6 years at Booking. I joined as a Team Lead / Data Scientist; when I left, I was Director of Machine Learning. It was a full-time management position.

I managed a few teams that worked on the platform and the machine learning infrastructure that data scientists used. The role of this team was specifically to develop tools that would help data scientists do their job. To put it simply: we helped data scientists produce machine learning models.

Wahou. How many people are we talking about?

At the peak of activity — 40 people.

How did you hear about Pennylane for the first time?

Before joining Booking, I worked at PriceMatch [the first company Arthur Waller, CEO of Pennylane, cofounded. Booking.com acquired PriceMatch in 2015] with Arthur, Tancrède, Quentin, Thierry [all cofounders of PriceMatch and Pennylane]… and was glad to get back to the team again. I have known Quentin and Thierry for several years now — we are friends.

And what made you want to join?

I worked with Quentin and Arthur before, at PriceMatch. I knew what they had achieved, what they were capable of. We spend hours talking about Pennylane’s business model and its viability. From day 1, I was convinced that we would go far. Of course, I wanted to join!

I was also very excited to build something from scratch; to get my hands dirty again, to be “on the field”. I had started to miss that at Booking.

What were your first missions at Pennylane?

My very first mission was precisely to understand what our Pennylane-specific challenges would be — types of data and models, scale, technical constraints… I was very conscious of the fact that I came from a B2C environment, with very different challenges.

Both at Booking and Pennylane, I spent a lot of time thinking about what entails data, what kind of needs would emerge.

For instance, at Booking, we paid close attention to the website latency, which is critical for the conversion of visitors in e-commerce — and not so much in a B2B environment. Another challenge was the volume of data generated by the millions of daily visitors. At Pennylane, the scale of data is always going to be lower, and we invest a lot more in data quality.

Once I identified those challenges, my objective was to enforce the right tools and methods for data.

Since we’re working in finance and accounting, we have to gather data from many different sources, and reliably consume it in very different contexts — our app, business intelligence tools, advanced algorithms, etc. So it was critical that, from the very beginning, we had the right infrastructure that would give us this flexibility without making compromises on data availability and quality.

OK, very clear. Let’s go back to the beginning for a moment. What did the Data Team look like when you joined? What’s next for the team?

When I joined Pennylane in April 2021, the Data Team consisted of 2 people: Grégoire David and me. Today [mi-November 2021], there are 5 of us — we are expecting 3 more people by early January, and the team shall triple by the end of the year 2022.

Our first concern, as a team, was to recruit more data engineers — not analysts or scientists. Grégoire and I did a lot of data engineering ourselves; still today, a big part of the whole team is dedicated to engineering.

Why is that so?

I wanted to make sure data analysts and scientists would have access to quality, easy-to-use data. And that meant working hard on the engineering part of data first.

If we don’t spend the necessary amount of time to manage our data from the start, future analysts and scientists will spend most of their time compensating for data quality and availability issues. We do not want that. It’s a poor investment that would make us waste a lot of time in the future and would cause frustration in the team.

We voluntarily dedicate more time to data engineering than to data analysis to ensure our analysis provides value.

Only now are we reaching a point where it’s interesting to analyze data — because we reached the technical maturity to ensure its reliability and availability.

What are some of the Data Team’s challenges today?

There are 3 of them:

• Infrastructure — how do we identify and build the infrastructure that best fits our needs?
• Insights — how do we make sure every team gets access to reliable, actionable data?
• Intelligence — how do we add a spark of intelligence to our platform?

One of the main challenges we’re facing regarding “Intelligence” is: how do we automate data entry for accountants?

It doesn’t sound like it but it’s a fascinating subject because it embraces so many things! For instance: how to read invoices automatically and interpret their content, how to automatically match an invoice with a bank transaction…

Accounting may not be glamorous but it’s a very demanding intellectual exercise on our end, and I love it.

I have one very last question — but it’s a bit personal, you don’t have to answer. You stayed in the Netherlands for 6 years; wasn’t it too hard to move back to France?

It was a big change and we thought it through. Today my wife, our two kids and I are really glad to be closer to our friends and family

And if I had wanted to stay a couple of more years in the Netherlands, I could have since Pennylane is a remote-first company.

No, to be honest, I’m really glad we moved back. We live in the Lyon countryside now [a region famous for its gastronomy], and I couldn’t be happier, with all the food and cheese we have available!

If you’d like to talk more about data (or French cheese), you can reach out to Thomas on LinkedIn.

And if you want to learn more about job openings in Pennylane’s Data Team, I encourage you to visit our Careers page. :)