Benj joined Cleo back in October 2019 as a Senior Data Scientist. Back then, there were only three or four Data Scientists at Cleo. Fast forward to now, and Benj has just been promoted to Head of Data Science, leading a team of 10 Data Scientists and growing— we love to see it 💙
Can you tell us a bit about your background and how you came to land a job at Cleo?
I’ve worked in data roles for 9 years now, with a real range of industry experience. I’ve worked in startups smaller than Cleo, right through to larger corporations.
Before working as a data scientist, I actually completed a PhD in computational biology. You may be thinking “where is the link between natural science and data science at a fintech scale-up?” Especially when my PhD research was tracking flocks of homing pigeons with GPS.
However, it was actually my studies that sparked my interest in a couple of things:
- Using collective intelligence to benefit society. At Cleo, that looks like using previous user behavior data to create intelligent products that people love to use.
- Natural sciences were increasingly doing more mass data collections, so I had to learn to code to make sense of the data.
In terms of why I chose Cleo specifically, a large part of why I joined was to do with our mission and seeing the positive impact of my work. I love that we can put people’s financial history data into their own hands, in a way that’s user friendly and interactive.
Another reason for joining was the fact that Cleo is a tech native company. Barney (our Founder and CEO) was a Data Scientist, which means that there’s a greater emphasis on doing more with data and using machine learning at Cleo than you may typically find at other startups.
How has your previous experience prepared you for the transition into Head of Data Science?
I gradually started to manage other Data Scientists since joining Cleo, which has been a great experience. The fact that I’ve been here a while (in startup terms) means I’ve often been seen as the go-to for answering questions about many of our existing models. Over time, it made sense to have internal communications as a main part of my role — now I spend a lot of time engaging with other colleagues about what improvements they need to see across machine learning and how we can measure it.
Data science is highly valued at Cleo. We move quickly, model deployment is self-service, and you can switch between product areas. I’m proud to say that Cleo is good at keeping great data scientists on board, and I think a large part of that is down to the variety of problems available to work on.
What are you most proud of from your time at Cleo?
I built a multi-armed bandit system for optimizing notifications. Our great team of writers is churning out fresh copy all the time, but we don’t know which versions of budget insights are going to be the most engaging ahead of time. I adapted a Bayesian multi-armed bandit algorithm for this task, which makes sure new versions get tested, and once it has enough evidence it promotes the most popular.
I also helped to create a machine learning service to optimise how we retry user subscription payments, using our paycheck prediction. This means we’re not retrying for payment every day, instead we’re more likely to retry just after we think someone has been paid.
As a team, we’ve put a lot of thought into how we evaluate our chat systems. When I first joined, this was very qualitative and reactive, whereas now we systematically check the quality, find areas of improvement, and use active learning to label data.
What are you most looking forward to in the next 6-12 months?
It’s hard to choose just one thing, so I’m going to go with two:
- I’m really excited to see how the chat advisor evolves as we improve the ecosystem of models and tooling around it.
- We’re looking to double the number of data scientists working on chat and personal finance insights, which means we can make smarter decisions and improve our products for our users.