Econsultancy: Please describe your job.
Ben Chamberlain: I am a senior data scientist at ASOS, where I lead the customer understanding team. We build systems that generate customer intelligence from proprietary data.
We work with a load of teams inside ASOS to help put everything that we know about our customers into making better products.
E: Whereabouts do you sit in your organisation?
BC: At ASOS, data science is part of Digital Product and I report to Andy Berks, our Digital Product Director.
E: What kind of skills do you need to be effective in your role?
BC: There are three classes of skills that data scientists need to have. People have different strengths and weaknesses, but everyone needs to have at least some knowledge in each area.
The first is the ability to get things done quickly and effectively with a computer. This might be downloading and testing some new software, fixing a broken program or getting two databases to communicate.
The second skill is statistics / machine learning. While a PhD in stats is not a requirement for the job, it is necessary to understand the fundamentals well. Machine learning is unusual because very bad algorithms can often look very good when they are not tested properly. Of course, they don’t work when put into production, but often a lot of money has been spent before you realise. This is what data scientists call the ‘danger zone’.
The final skill is business knowledge. This includes things like knowing which problems data science can help with and which to avoid and choosing the appropriate amount of resource to allocate to projects.
Ben Chamberlain, senior data scientist at ASOS
E: Tell us about a typical working day…
BC: My typical day is a nice blend of research, writing code and discussions with stakeholders. As scientists, we have to stay in touch with the latest ideas, so the team spend a good amount of time at the whiteboard talking through new research.
I usually have a couple of meetings with the engineering teams and business executives to work on experiments that we would like to run, as well as writing a fair bit of code. There’s also the occasional special ASOS surprise thrown in. Today, for instance, our new tech bar is opening, so the whole office is finishing at three and we have a DJ to help us celebrate.
Another good example is last week, Duncan Little, one of our scientists, traded modelling data for modelling clothes and went into the studio to shoot our new ASOS Lookbook.
E: What do you love about your job? What sucks?
BC: ASOS is a great place to work. It’s young, energetic and still growing 20-25% year on year. There are only a handful of businesses with global HQs in London that have huge customer bases and great data. This means that as data scientists we work on the company’s most important problems and our work has global reach. The great employee discount keeps my girlfriend happy too.
It’s hard to think of things that really suck, but if you pushed me, I’d say that there are no gyms nearby. They recently made all ASOS internal gym classes free, but trying to get a spot before work is like trying to get a ticket for Glastonbury – you sit refreshing your browser every ten seconds from 8:59 the previous day and it’s still a long shot. That said, we’re currently building a new ASOS gym area that should be ready next year. First world problems…
E: What kind of goals do you have? What are the most useful metrics and KPIs for measuring success?
BC: Our goals are very much project based. We work with a lot of different teams within ASOS and they all have their own KPIs. The first project that I did at ASOS looked at how we could predict the customer lifetime value and how we could use that information to improve shopping frequency and average basket value.
Our most recent work looks at how customers engage with ASOS communications and how we can use this information to increase email open and click-through rates.
E: What are your favourite tools to help you get the job done?
BC: I program a lot in Python. I use Jupyter notebooks for exploratory work and the Pycharm integrated development environment for larger software projects. Google recently open sourced a python library for deep learning called Tensorflow, which we have all found really helpful.
Our production code uses Apache Spark, which is great for distributed processing when we need to churn through 10TB of web logs every day.
E: How did you get started in the digital industry?
BC: I was working for QinetiQ, which is a British defence and security company, as a signal processing scientist. I didn’t really do any signal processing though and spent most of my time writing graphical user interfaces in C++, which nobody would do any more.
I remember at that time reading a New Scientist article, which surveyed scientists and asked them which subject they would choose if they could change specialism. Artificial Intelligence won and shortly after that I met Simon Maskell, now professor Maskell, who was working on AI applications for intelligent systems. I begged him for a place on his team and started working with him on a form of customer understanding for the intelligence services.
From there it was not such a long hop to ASOS.
E: Do you have any advice for people who want to work in AI?
BC: AI, not for the first time, is really hyped right now. This won’t last and people drawn into the hype are likely to be disappointed. I would say, ask yourself if you have a real passion for this type of work. If you do, then there are now lots of great online resources to help people learn. Two that leap to mind are Geoff Hinton’s Coursera course and Nando De Freitas’ Oxford lecture series, which is available on YouTube.
It is quite a competitive area and so doing things like entering Kaggle contests, writing open source projects or doing an internship can help differentiate you from other candidates when it comes to job applications.
For more on data science, read A Day in The Life of.. a data scientist in an AI company.
If you’re looking for a new role in digital marketing, why not check out Econsutancy’s jobs board.