Creating a great data team starts long before you get a bunch of data professionals on payroll and in the same room. For Tim Lum who is innovating business intelligence, analytics and big data at Virgin Atlantic, it is about finding skilled and curious people from a range of backgrounds and getting them to effectively engage in answering questions that create business value.
The first stage is the candid analysis of the skills that are already present within a company’s data and analytics team and the identification of gaps that need to be filled. “Ask yourself, where in my organisation do I not have these people, and start to build your team around that,” he told an audience at Tableau’s London offices.
At the event organised by Kubrick Group, Lum stated that the people who should be part of a data team are data developers, data researchers, stats people “who help to build models,” data creators “who build nice visualisations and hack at stuff,” and data business people “who interface with the business on how to get stuff through into strategy.”
He said that vacancies for any of those roles will then need to be filled with talented individuals from a wide range of backgrounds. Lum himself is not from a STEM background, as he is a sociologist by training.
“If the folks you hire are really just interested in the methods, that’s not good enough.”
Once those people have been invited to interview, he makes sure that they are inquisitive enough fit the culture that he has fostered with his team. He leaves ample time at the end of each interview for candidates to demonstrate how curious they are. Their continued candidacy is dependent on how inquisitive they are at the end of each interview.
“I can quiz skills in terms of code writing, but if they don’t ask questions, like how we start to pull in data, then that doesn’t help my general team culture,” he said. “If the folks you hire are really just interested in the methods, that’s not good enough.”
When a candidate is hired, Lum said it is important to focus on their wellbeing and career development “immensely”. He said this sign posts to them the level they need to attain while at the same time pushing and guiding them. This has the second advantage of encouraging the candidates to develop the autonomous drive to get themselves to that level.
To help new data team members get an understanding of the business, Lum said new hires with less industry experience should be mentored. He is also a big advocate of an open talent market within the organisation so that internal transfers can take place and then knowledge and experience from other parts of the business can benefit the data team. “They know something from different parts of the business and they bring that with them,” he said.
When engaging with the business in general, Lum said it is important that this happens effectively so that the business understands what the data team does. When that happens, the data team helps the business to understand what business processes they need to change.
As for the way of working within the analytics team, Lum has coined the term "insight-driven development," a method he developed while heading up an analytics team at Expedia for nearly four years until mid-2017. Essentially, it involves the analytics teams taking an overall business question and breaking it down into a smaller list of questions.
"Going through the list of questions, the data team can figure out what they know and don’t know."
If the question was how to reduce overbooking on flights, Lum said his team would start by looking at the present percentage of overbooked seats and then look at the reasons why someone might be offloaded from a flight and how often it happens.
“Going through the list of questions, the data team can figure out what they know and don’t know and whether the data even exists for what they are looking for,” he said.
If the data doesn’t exist, the team can make the decision to pause on that line of enquiry and find where the data is going to come from before delving any further. He said this is important because if you try to run models without the right data, they will not be accurate and will therefore have no business value.
“You start to build that insight once you know you’ve got the data for it. Make sure you have enough data to actually run a model and for when you have a hypothesis that allows you to do that. Then you work iteratively as a result in order to start building back,” he said.
“You wouldn’t trust an airline of only pilots or flight attendants.”
As words of warning about what to not do, Lum advised against trying to find a "unicorn" data scientist with a list of skills and experience as long as your arm (or wing span). This is an issue that has been raised by others in the data industry.
Firstly, it will be nearly impossible to pin them down and, if you do, they would be so in demand, it would cost an arm and a leg to hire them.
Furthermore, in his mind, it is not a good idea to fill an analytics team with people who are jacks-of-all trades. “You wouldn’t trust an airline of only pilots or flight attendants. Why trust an analytics team of only data scientists?,” he cautioned.