I have had a varied career which has helped me in the wide variety of things I need to understand as a chief data scientist. My undergraduate degree is in Mechanical Engineering, a field that involves computer simulation, linear algebra, tensors and numerical methods. I was less interested in automotive and aerospace applications and so earned my PhD in Computer Science. My thesis investigated statistical methods and designed experiments for tuning a type of AI algorithm based on the swarm intelligence of social insects. I moved to pre-sales consulting, tuning and integrating a social network fraud-finding solution for clients in insurance and retail banking. I loved the fast pace of pre-sales combined with data complexity, so spent several years at big four accounting firms running forensic analytics teams. I saw teams struggle to deliver analytics with changing data, changing requirements and evolving business rules. I wrote my book, “Guerrilla analytics,” to help scientists and analytics practitioners leverage principles from software engineering to deliver their work in these difficult circumstances. For the last four years, I have been building the applied data science capability at Sainsbury’s. I have the pleasure of leading an awesome team of scientists, engineers and product management to build the algorithms that automate incredibly complex decisions.
A standout highlight is founding and growing Sainsbury’s applied data science capability. I started with one mathematician, no infrastructure and a business full of potential, but limited knowledge of how to progress. I am proud that we are now a significantly-sized team with our own budget, leveraging cloud native infrastructure and delivering quantified benefits in the millions. Writing “Guerrilla analytics” has certainly opened doors for me. A book is a huge investment of time, but it helped clarify my thoughts on the best operating principles and practices for analytics and data science as well as growing my network internationally.
Get the foundations right. An understanding of applied mathematics is essential. An understanding of aspects of computer science is essential. Learn to communicate in a commercial setting. Work for companies where you learn from others and where leadership is tackling the obstacles to agility and access to data.
Yes. My aim this year was to demonstrate that complex decisions can be automated quickly if you put the right people, process and technology together. I’m proud to say we have delivered several valuable cloud-native algorithms in months and restructured to double down on these successes.
I will be driving towards greater convergence of science/mathematics and product development. The cloud-native abstraction continues. Common algorithms are available cheaply at scale and sometimes partially pre-trained. Complex functions, SQL databases and even MI dashboarding are serverless. The old (but recent!) ways of working with scientists in silos writing scraps of code handed off to engineers will change. I see scientists, product managers and engineers all working together off shared product data, using the technologies above to provide the lens that each discipline requires. This means scientists, product managers and engineers all learning about one another’s disciplines and new ways of working that promote this learning.
There can be mis-managed expectations around what industry requires versus what candidates have trained in or heard the hype about. We look for great data scientists, engineers and product managers with the right attitude that aligns to our values of courage, openness, working together, drive, accountability and growth. We help them grow to be disciplined, agile, applied scientists through rotations, internal training, structured training and access to our group-wide community of practice.
I am optimistic that organisations will ignore the hype around artificial intelligence and make pragmatic advances on extracting efficiencies and opportunities from data. Cloud providers have lowered the barrier to entry of data storage costs and increasingly the algorithm/compute barrier. All that remains is cultural acceptance of the changes this will bring.