At the Deep Learning in Retail and Advertising Summit, leading minds working in AI from the likes of ASOS, Vodafone, Argos, WalmartLabs and many more came together to share their latest research and cutting-edge applications of AI and deep learning in e-commerce as well as optimisation behind the scenes in warehouses.

Joining us from Royal Mail was Kat James, Senior Data Scientist, who spoke about offline applications of recommender systems. I was fortunate enough to sit down with Kat and record an episode for the Women in AI Podcast, where we spoke about her current and past work, as well as her place in the industry more generally.

You can listen to the podcast here, and read a preview of the interview below:

Tell us a bit about your work at Royal Mail

I am a senior data scientist at Royal Mail and I’ve been there for about 9 months now. From a data perspective, Royal Mail is an absolutely fascinating organisation (and challenge). We are legally obliged to deliver approximately 50 million letters to 24k addresses 6 days a week within 1-3 working days. As you can imagine, we collect data from all stages of the operation so a lot of our work is around improving efficiency, working with tech such as IoT devices and vans and forecasting how the operation will perform in both the short and long term. It’s a hugely exciting challenge, but as you can imagine, no 2 days are the same.

How did you begin your work in AI and in recommender systems?

I completed my PhD in 2014 in statistical genomics, where I focussed on building statistical models to deal with the application of new sequencing techniques to rare tropical viral diseases (in my case HIV-2). At the time, I was aware of Data Science as a career option, but it was only when I started looking for my first position that I became aware of how closely aligned the skills I had from my PhD were to those needed to be a data scientist. My first role was at British Airways, where I was almost immediately given the challenge of personalising destination for marketing comms. This problem was best solved with a content-based recommender and it was love at first sight from then on!

You mentioned the content-based recommender you worked on at BA - when I think of these systems I think of online shopping, for TV and music suggestions etc., so how are you using them at Royal Mail?

We are playing about with where we can use them at Royal Mail at the moment. We have implemented a recommender system to help the B2B sales team, which is probably our most traditional use but we are also working on using recommenders as information filtering algorithms, hoping to simplify making data-driven decisions. We are a single company with 1400 delivery offices acting to a certain extent as single entities so thinking of each operational unit as a user allows us to develop recommenders to personalise how we are communicating data to colleagues.

What challenges are you currently facing in your work, and how is deep learning helping you solve them?

Our main challenge, which is probably relatively unique to RM is that we have too much data. There isn’t an area of the business that isn’t crying out for help from our team and combining advanced data science techniques and legacy systems is a constant challenge for us. Deep learning helps us to make sense of the huge amounts of data we have to play with, especially when faced with very diverse data streams and constantly changing requirements.

Hear Kat answer the following questions on the podcast:
  • What are AI’s implications for the retail industry as a whole?
  • Which other industries are you most excited so see implementing AI for a positive impact in the next 5 years?
  • What do you think we can do to encourage more girls into AI and deep learning?
  • What do you think the key skills are for a career in AI/DL?
  • What’s next for you?