Many people have written and presented on how to build a successful data science team. Finding the right talent, working on the right problems, delivering actionable insight are a few things highlighted as the key factors. One key factor I find that is often missed (and I personally believe a very important one) is finding the right leadership for your data science function.
In my 15+ years as a data scientist, some of my managers had an ability to make me (and the team) develop and stretch our capabilities and therefore produce outstanding and innovative work. In the highly competitive market of data science, companies cannot afford to lose the hard-found data scientists due to mediocre technical leadership.
What was it about their leadership that made a difference and inspired the data scientist in me?
It made me wonder. Apart from the typical traits of great leadership (like honesty, confidence, decisiveness, communication, vision), what was it about the leadership of my previous bosses that made a difference and inspired the data scientist in me?
Be a guiding light, not a fireworks show
The data science leader might have lots of new ideas that are fed from the rapidly developing data analytics industry. Working for a leader that constantly bombards the team with new ideas and requires the team to act on all of them, often feels like watching a fireworks show. Just as the team has their attention fixed at one point in the sky, it is pulled away by an explosion somewhere else. Being a “fireworks show” with a continuous flow of diverse bright ideas may impress the upper-management, but as a guiding light for the data science team, it is destructive.
Your data scientists thrive on solving those difficult problems – changing purpose and direction too frequently will undermine their sense of accomplishment. As with a fireworks show, where eventually the smoke blurs the vision and the noise is tiring, the team will lose sight of the longer term strategy and become fatigued with starting up new things but never following through.
Challenge the team intellectually, not with timelines
Often leaders of technical teams don’t have the same technical background or depth of their team. In the absence of technical skill, some leaders then resort to timeline challenges rather than technical challenges.
The really good technical leader figures out how to challenge the team intellectually, even if he/she doesn’t have the same technical depth as the data scientists. These leaders will work with the team to identify diverse, interesting and hard problems that will grow the technical capability of the team.
Make time for innovation and learning
You have hired a highly skilled, curious data scientist. The worst thing to do to them is pinning them down with tightly scoped-out or repetitive project work and not allowing time to investigate, discover and explore areas of interest. Innovation is seldom achieved within prescribed timelines and rigorous project management.
One of best strategies I have seen applied was to allow a small portion of the team’s time for working on their “pet” research project and share the challenges and discoveries of the project with the rest of the team.
Consult when making technical decisions
Many of the smartest data scientists consciously choose a technical career rather than a managerial one. Respect their knowledge and skill by consulting them when making technical decisions.
In particular, the leader of a data science team may often be the first point of contact of software vendors or providers to flog their new technology. Many vendors claim they have an analytics component in their tool, but in reality it doesn’t have enough depth for the data scientist to tackle the real world problems. Without hands-on experience, it can be hard for a leader to identify true capability from a sales pitch. A good way to validate claims is to run trials or proof of concepts with the data scientist to test the capability.
Seek out peer review of work
The best technical leaders are not afraid to have their team’s work peer reviewed. These leaders recognise that independent review is not only an important part to keep the team challenged, but also ensuring that the business decisions are made from sound data science practices. With the rapid developments in data science, it is necessary to check that the team has applied the new technology appropriately and learn from other’s mistakes.
Be supportive of the team to attend external technical events and publish or present their work externally. There are lots of ways to still protect the business’ IP whilst discussing the data science technology with peers. Additionally, gaining external recognition for innovative and interesting work is a great motivator for data scientists and will help your function to attract the best talent.
Be knowledgeable without the “name dropping”
There is a lot of hype in data science. New algorithms and software start-ups with exotic sounding names pops up daily. Most of these are basically just a variation of already existing technology. For example, the deep learning algorithms are not that brand new at all. It is only a special type of neural network. The exotic sounding random forest is in reality only an ensemble of decision trees.
The good data science leader will be conscious of this fact and attempt to understand what is behind the latest catchy name. Utilise your data scientists to investigate the mathematics behind new technology.
Celebrate mistakes
A big part of data science is about manipulating, transforming and reporting of data, and writing code. Mistakes are easily made. If one data scientist made such a mistake, chances are another one will too. The best technical leaders will create an environment where mistakes are discussed openly, learned from and measures put in place to avoid it from happening again.
One of the best strategies I have seen was to celebrate the mistakes in the same way as good performance: Give an award for the “best’ mistake that was discovered and learned from.
