What analytic managers can learn from neural networks

Discover how to put neural management to work for your data science team when traditional models fall short.


Credit: Lights of ideas image by Saad Faruque, CC BY-SA 2.0

If you're managing a data science team and you don't know how a neural network actually works, you may be in worse shape than you realize.

First of all, if you're going to manage a team of data scientists, you better know a lot about data science. If you have any doubts about this, you'll just have to trust me on blind faith; this is a non-debatable issue for me. And since neural networks comfortably reside within the realm of data science, this is something every analytic manager should know. However, over and above content knowledge, analytic managers can learn a few management techniques by studying how these magical data structures work. This is what I refer to as neural management.

Neural management: A management model based on neural networks

A neural network is a programming model that's loosely designed around the way our brains work. You can use almost any programming language for the model.

When I was in college, I designed one in Visual C that learned how to solve 1 + 1 = 2. It was a small network of only a few nodes: two input nodes, one output node, and four nodes on a hidden layer, between the inputs and the output. Each node on the hidden layer was connected to each input node and the output node. The nodes on the hidden layer were initialized with random values and programmed to adjust themselves based on whether it was producing the right output, given the two inputs. A small training set was constructed with a series of correct inputs and outputs. Once trained, the neural network could successfully output a two every time both input nodes were presented with a one. In the real world, we don't need a neural network to solve this simple problem; in practice, neural networks are much more sophisticated and employed when traditional techniques for predicting outputs (e.g., logistic regression) fall short.

Neural management was inspired by the way neural networks operate. It's a philosophy that's most aligned with an iterative or agile style of management. With reference to the post, lever, and balance method (PLBM), we'll post on effort, use time as a lever, and scope as a balance. Alternatively, you could use time as a post and effort as a lever; however, the purest way to employ neural management is using effort as a post. Each node on a hidden layer in the neural network represents a data scientist. There are no special nodes or role definitions — all data scientists perform all functions. Like neural networks, this management model can be deployed when traditional models (e.g., hierarchical waterfall) fall short.

Firing the synapses: putting neural management to work

Neural management exploits the power of a high performing team, so it's important to nurture it as such. Anytime you post on effort, it's an indication that the team's dynamic is your highest value. Select your data scientist nodes carefully, with an acute awareness of how cultures and personalities will mesh.

You'll also need a good coach to tend to the human dynamics of team performance. An extremely high level of communication is required, so make sure everyone is not only physically collocated, but also proactively coached into talking with each other. Remember, data scientists are typically introverts by nature, so just because you sit them next to each other, doesn't mean they'll actually talk to each other.

Execution resembles an agile methodology, but there are subtle differences that make it characteristically different than agile. First and foremost, I discourage the idea of self-organization with data scientists. In many cases, data scientists simply won't self-organize, so this team should be managed tightly by a good analytic manager and coached by a good analytic coach. Also, there should be iterations, but strict timeboxes are not necessary. As I mentioned earlier, I prefer to use time as a lever instead of a post.

Once the team is set, divide the scope into equal chunks of reasonable functionality, and then prioritize as you would on an agile project. Then shoot for a period of time (say, a month), but feel free to extend this a bit if you're close to finishing the target chunk of scope. If you're not even close to finishing, you'll need to adjust the size of your scope chunks. Continue this process with the next iteration, making adjustments to time and scope until you feel comfortable with the delivery. After a few iterations, you'll have a good sense of how long an iteration should take and how much scope will be delivered. In this way, you can train your neural team to deliver results.


Neural networks are not only an important programming model for analytic managers to understand, they also represent a management paradigm that can put your data science team in a position to solve problems other teams cannot solve.

If you don't understand how neural networks work, take some time today to do the necessary research. Once you have the basics down, consider structuring your data science team to post on effort, use time as a lever, and scope as a balance. It may not feel right at first, but it sure is a good feeling to crack a code nobody else can. Neural management just might have the answer you're looking for, so give it a try.