My biggest mistakes as a data-science team lead
Mistakes, I've had a few
A month ago, on a sunny Friday in September, I had the pleasure of attending the Baltic Data Science Day in Vilnius, Lithuania. That was a great event, with fantastic panel discussions and lots of active audience participation. Hats off to the organizers for kicking off what I hope will become an annual Baltic tradition.
The first panel of the day was on people leadership. Picture a bunch of data leaders and an incisive moderator. There’s Slido, you post questions, and upvote the ones you like. The most-upvoted questions get answered.
One question that made it to the top of Slido rankings was this: “What’s the biggest mistake you’ve made as a leader?”
Unfortunately, I didn’t take notes (silly me), so I no longer remember what any of the panelists actually said. I do remember, though, that the question made me think. How would I answer that question?
I think I’d start by saying, gosh, lots of them. Mistakes are unavoidable, whether you’re a team lead or an individual contributor. The best you can hope for is having strong feedback loops in place. Once you inevitably screw up, hopefully, somebody quickly tells you that you screwed up, and you correct course.
Then I’d tell you this.
Mistake #1: Mis-situating leadership
There’s this thing called situational leadership. The basic idea is that you should adjust your leadership style to the person you’re leading. Sounds trivial. It’s not.
All business ideas must, by law, be presented in a 2x2 consultant’s table, so here goes. Put the “motivation level” of the person you’re leading on the x-axis and their “skill level” on the y-axis. You get something like this:
The key situational leadership idea is that the way you lead should depend on the skills and motivation of your direct report:
If you’re leading a highly motivated top performer, just stay out of the way (a.k.a. delegate). If your direct report has the skills but not the motivation, understand where the lack of motivation is coming from and provide support. If someone is motivated but lacks the skills, be the coach that this person deserves. In the unhappy quadrant of “low skills, low motivation”, you need to be hands-on: Provide a clear action plan, pay close attention to execution, and provide lots of immediate feedback.
I knew the theory when I started out as a team lead. But my thinking was a little bit like this. Let’s say I have two folks on my team, Alice and Bob. Alice is a highly experienced data scientist, currently doing a great job on a new ML model. Bob is an analyst, fresh out of university and apparently struggling to find the motivation to calculate some new metrics. How should I lead Alice and Bob?
Here’s the naïve take on the situational leadership model: Alice is in the top-left quadrant, so I should stay out of the way, while Bob lies in the bottom right, so I’d better be hands-on:
The problem, though, is that people don’t stay in the same quadrant. Maybe Alice is awesome at prototyping new ML models but finds it really tough to deploy them in production. Suddenly, your hands-off approach feels like abandonment. The reality is that humans are not some fixed points on a two-dimensional surface. We’re a bit more complicated than that.
As a result, the real world looks more like this:
Alas, it took me a while to appreciate this. Not only do I need to adjust my leadership style to the person I’m leading, but I also need to calibrate my approach to a given person over time.
Mistake #2: The firefighting trap
Here’s a common way to think about people leaders vs individual contributors: Individual contributors solve code problems; people leaders solve people problems.
That’s how I initially approached my new role. Find out what the key people problems are. Understand how to solve them. Solve them. Repeat.
I don’t think that’s entirely incorrect, but it’s an incomplete frame. Your job as a leader is not to solve problems but to create value. A problem-solving attitude is helpful, but that can make you lose track of where the value is actually created.
For example, it’s almost a management cliché that you should focus on your stars, not your underperformers. However, in the daily trenches of managerial work, you naturally navigate to the most burning issue. “Alice is killing it, while Bob is struggling, so let me focus on Bob” sounds reasonable. However, if you spend all of your time in 1:1s with Bob trying to improve his performance, Alice will feel ignored, and her performance may deteriorate. Suddenly, you’re left with zero stars.
Easier said than done, of course. If you see a problem, the natural instinct is to go after it. However, always putting out fires can be a trap and make you less effective as a leader in the long run.
Mistake #3: Flying blind
You’d think a data science team lead would be extremely data-driven. Well, I wasn’t.
The key reason for that: I thought that collecting data and setting metrics for my team would be too difficult. However, it doesn’t have to be that way.
If you’re working at a product-driven organization, your data team is likely either:
Building business intelligence tools (e.g., dashboards, metrics, data infra);
Building data-science products (e.g., ML models, recommenders, decision systems).
For (1), you can check usage and reliability: How many users do these dashboards have? How many monthly views? What’s the uptime for your data infra? You can also do simple surveys to measure user satisfaction and the key data pain points.
For (2), my strong view is that each data-science product should have a clearly defined success metric. For example, if you have an ML model that’s used to forecast sales, your success metric is something like “average forecast accuracy.” You do some work to understand what level of accuracy is necessary and then aim towards that goal.
It took me a long time to appreciate the importance of metrics for data teams, and it remains a growth area. Nevertheless, I strongly believe that having metrics for data teams is possible and really worth it.
So what have I learned?
That’s what I would reply to “What’s the biggest mistake you’ve made as a leader?”. Like, you know, if I could give a one-hour lecture on the topic.
At the end of the day, mistakes are unavoidable. I think you just do your best to stay humble and listen to what people around you are saying. Learn from that.
Then, make better mistakes.