If you have ever brought your data science team together to solve a problem and after three hours of deliberation gotten nowhere, you’re not alone. Data scientists are notorious for spending insane amounts of time in circular discussions. I can say that because: 1) I am one, 2) I work with them all the time, and 3) I’m absolutely guilty of participating on more than one occasion in a discussion that ultimately went nowhere.

Don’t blame the data scientists — that’s the way they work. It’s management’s responsibility to make sure time is well spent in group discussion. So, if you’re an analytic manager responsible for group problem solving with your data scientists, let me show you how to make sure you don’t waste everyone’s time.

What’s the problem?

Group problem solving with data scientists takes very strong facilitation. The goal of facilitation is three-fold: make sure the objectives of the session are met, make sure the session ends on time, and make sure people don’t attack each other in the process.

The biggest mistake managers make when bringing a group of data scientists together to solve a problem is not taking the facilitation serious enough. They assume they can just gather everyone, state the problem, and let the discussion take its course. This likely won’t work with data scientists, so you need to put in some time upfront to build an agenda.

The second biggest mistake managers make is wasting time solving the wrong problem. So the first major objective on the agenda should be consensus on exactly what the problem is. Don’t underestimate the value of doing this step. I’ve opened up many meetings with a clear problem statement only to realize that not everyone’s solving the same problem. This usually happens when you tell the group what the problem is without their buy-in.

So, even though you’re clear on the problem, start the meeting with a brainstorming session on exactly what the problem is. Ask everyone in the group what they think the problem is and record their responses. You’ll invariably see variations, which need to be reconciled. You’ll also see people trying to solve the problem (which hasn’t been clearly defined yet). Put these ideas in the parking lot and steer them back toward clearly defining the problem. You cannot solve a problem until everyone in the group owns the problem definition.

What’s the answer?

Once you have group consensus on the problem, it’s important to shepherd them through the process of solving it. This is harder than it seems, because data scientists don’t like too much structure when they’re trying to solve something. They think it confines their solution space, which is true; however, there are time constraints as well, which they won’t respect as much as you. So, you must move them distinctly through a three-step process of: discover, analyze, and conclude.

The discovery phase is the most natural in the problem-solving space: what are some possible solutions to the problem? Run this as a brainstorm, just like the earlier session on problem definition. Encourage openness — there are no bad suggestions. However, keep them away from over-solving in one area. The idea is to brainstorm as many possible solutions that the team can generate within a specific amount of time. Once the allotted amount of time for brainstorming is up, this agenda item is done and it’s time to move on.

The analysis phase is where you organize the results from the brainstorm. Tactically, this should be pretty easy for the data scientists, but you must shift their gears to move in this direction. As noted earlier, the natural tendency is to stay in brainstorming mode, so make it clear that you’re in a new space now. The analyze session consists of putting the results into logical groups (affinities) and polling the group to see which group/solution makes the most sense. You could also synthesize the solution groups together into a super-solution if it makes sense. At the point where you have a sense for where the group stands on the different solution groups (or hybrid solution), it’s time to draw a consensus.

The conclude phase of the facilitation is usually quick and easy — it’s just making the team’s orientation explicit. Once you have a sense for where the team is you can make a statement like, “It seems like the group is leaning toward solution A; is that a fair statement? Can we all agree that this is the best path to follow?” If there’s any reticence at this point, it should be addressed. It’s also important to remind the group that it’s okay to concede with a direction, even if you don’t agree. Note any and all dissensions formally. The record will not only validate the dissenters (making it easier for them to concede), but it may also serve useful if the consenting path proves unfruitful.


As intellectually stimulating as circular discussions are, there’s no place for them in a problem-solving session. When time is of the essence and problems surface, it’s imperative as an analytic manager to successfully navigate your data science team to the other side of the problem.

Before bringing the group together, prepare yourself for a strong facilitation with a tight agenda. In a very structured manner, get clear on the problem and then get clear on the solution. Otherwise, you’ll be taking the scenic route to nowhere.