I spent the better part of Labor Day flat on my back. The sad part is that I wasn’t performing any sort of tricky gymnastics or lifting large, heavy boxes — I was just having lunch and my back seized up.

After several hours of pain, a trip to the emergency room, and some good painkillers, I was able to walk again very slowly. This wasn’t the first time my back has gone out; in fact, I’ve been dealing with this for several years. The majority of the time I’m perfectly fine, but once in while — and often without warning — my back just gives out. My chiropractor says I’ll be dealing with this for the rest of my life. Once the back is injured in the way mine was initially injured, there’s an “evergreen weakness” in that area.

This idea of an evergreen weakness also comes up in my consulting work. Ask yourself: Does your data science team have an evergreen weakness? If so, there’s something you can do about it. As unpleasant as it sounds, the best way to get rid of weakness on your data science team is to let go of the problem.

Not everyone is a data scientist

Weakness throws off balance in a team. If you think about the collective competencies in your data science team as spokes on a wheel, a weakness would represent a short spoke. Think about how your wheel would turn if one of the spokes was shorter than the others. This imbalance takes a toll on the rest of the team members; it’s not fair to the short spoke or the rest of the team.

The reality is data science is a hot field, and everyone’s rushing in to get a taste of the good life; unfortunately, it’s not such a good life if you’re not built to be a data scientist. The problem is you’ll never get a weak link to admit their own inefficiencies after years of schooling, months of searching for a good job, and weeks of interviewing with your company.

I’ve been on several data teams where one or two people really shouldn’t be there. At first you try to help them along, but after a while you just adapt to the weakness. You and the rest of your team informally acknowledge that you’ll have to pull the weak link’s weight, and you try to be friendly while keeping them away from all the serious work. It’s an uncomfortable position for the rest of the team members, because nobody wants to say what everybody knows. It’s the leader’s responsibility to identify and ultimately remove the dead wood. Here’s how it’s done.

Identifying the weak link

If there’s a weak link on your team, you need to identify that person. Your team coach (if you have one) is the best person to handle this task. If you don’t have a team coach, bring in an internal or external consultant to run a team building session. Although you have a general objective of building the team, you also have a specific objective of possibly identifying a weak link. Share this with your consultant, but not with your team, as making this objective explicit will interfere with the intervention. The way the intervention is executed depends on your strategy for balancing the competencies within your data science team.

If your strategy is to have a team of individual specialists, then identifying where the weakness is may be a little more obvious. For instance, if your mathematics and programming are strong, but your graphics have a hard time connecting with your users, you probably have an issue with your data artist. In this case your coach or consultant must be very delicate in how the group discusses its shortcomings, as it’s easy to equate a functional deficiency with a specific person. Instead of talking about the group’s competencies as a whole, focus the discussion on individual contributions. It’s fair to celebrate those who have gone over and above their call of duty; however, if you notice that a lot of programmers and mathematicians are also building infographics, you should ask yourself why your data artist isn’t doing this job.

If your strategy involves building a group of well-rounded individuals, then your intervention will take a different course. In this case, it’s okay to discuss group strengths and weaknesses, but this won’t uncover a weak link; if there is a weak link on your team, this will become evident in the pre-work (which won’t be shared with the group). Ask each team member to rate other team members’ competencies (math, programming, graphics, etc.). If you honor the confidentiality of the survey, you’ll get honest answers from your team. If someone is consistently ranked low, this may be your weak link. Again, share the aggregate results with the team, though keep the individual responses private.

Repairing the chain

If you find a weak link, let them go tactfully but quickly. Don’t waste time trying to coach them. Most people who aren’t cut out for data science are good at something else. Find out what that is, and help them find a more appropriate job. Maybe you have a really good computer programmer that can’t handle high-level math; or, maybe you have a very logical thinker who’s good at math and programming, but struggles with the creativity required to build good data visualizations. They’ll do fine — just not as a data scientist. The quicker you get them off the team and into a more appropriate spot, the better off everyone will be.


Your data science team is only as good as its weakest link. If your weakest link is above the bar, then don’t worry about it. But some data science teams are staffed with people that shouldn’t be there. Sometimes it’s obvious, but many times because of social dynamics weak links fly under your radar for a long time. It’s your responsibility to know what you’re dealing with.

Even if you don’t suspect a weak link on your team, explore the matter by organizing a team intervention. You don’t want your team’s center to give out at the wrong time.