Agile Data Science: A Portfolio Approach to Managing An Analytics Team

I recently concluded a two-year stint managing a team of ten highly skilled analytics professionals spread across three different locations. There were of course many challenges but the team over-achieved on just about every measure of success one can imagine. The team’s wins include

  • completing on-time and under-budget a data-matching project that delivers tens of millions of dollars of savings to the Australian government every year for the foreseeable future;
  • development of an analytics data lake using open-source technologies to support ad hoc but data- and compute-intensive R&D workload for data scientists and engineers in the organisation;
  • development of a new and state-of-the-art entity-resolution engine that is being deployed to replace a problematic legacy rule-based system;
  • publications in scientific venues;
  • media appearances in IT News and New Scientist;
  • thought leadership in domestic and international partner engagements, including novel public-private partnership on data-sharing and joint risk-modelling; and
  • millions of dollars in external funding wins in competitive processes.

With that record, imagine my surprise when a team mate commented at one of our last team meetings that she didn’t know what my expectations were for the team throughout that last two years and how we actually worked as a team. This got me thinking about my own management style, about things I take for granted and things I do without thinking. Many of them are now second nature to me but they may not be completely obvious to others, especially those who haven’t had the experience of managing and coaching multiple analytics teams. Much has been written about the challenges of managing a team of independent-minded data scientists and engineers, a task sometimes described as akin to “leopard-herding”. Is our success a fluke or is there a method or two behind our madness?

Truth be told, the team’s successes didn’t come about from a master plan I somehow concocted using my strategic acumen. We did many things right as a team and got lucky on many occasions but, at the end of the day, there were actually only two elements in our management philosophy: being opportunistic and taking a portfolio approach towards project selection.

Being opportunistic allows us to be agile and respond quickly to shifting priorities and changing resource profiles. (We talked about opportunism in more detail in this blog post.) Having a portfolio of projects across the risk-reward spectrum allows us to engage in a diverse set of tasks that satisfy, in a balanced way, each team member’s need for achieving results and learning new skills while providing some management comfort that, in (mathematical) expectation, the team will achieve some definite and significant results over time. This makes good intuitive sense to me. From a risk management perspective, it would take extreme bad luck for every one of the projects in the portfolio to fail so the downside risk from year to year is somewhat capped. Most years we will have our fair share of successes and failures, from which the team will learn and grow. And, once in a while, we will hit a good year where just about everything we try worked and one ends up looking like a master strategist. 🙂

Like most good advice, the analytics management philosophy we have just described is simple to say but hard to execute. The main ingredient needed to get this agile data science management recipe to work is a culture that accepts failure as a necessary part of trying new things, both within the team and the broader environment in which the team operates. In practice, this means getting the right bosses in place or finding the right bosses to work for. The second ingredient required for the recipe, something I admittedly didn’t do well, is constant communication with team members about the overall philosophy and approach, because an individual that operates within a team that works opportunistically on a range of projects with different risk-reward characteristics can easily get disoriented without a compass to point them in the right overall direction. Work on that culture, communicate constantly even when you think you’re already understood, and follow the opportunistic, portfolio approach to project selection and execution and you may yet stand a chance of taming the leopards.


2 thoughts on “Agile Data Science: A Portfolio Approach to Managing An Analytics Team

  1. Congratulations on you and your team’s incredible achievements Kee Siong! Great summary of your secret to success as one of leaders in Data Science in this country.


Leave a Reply to Alex Cancel reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s