Data Science for the C Suite, part 1: Keep it simple

You’ve likely heard of the great modern accomplishments of Machine Learning broadly labeled as AI. The now old news of beating the Go champion, optimizing energy consumption in data centers, breakthroughs in computer vision and speech recognition, medical diagnosis, etc…. You’ve likely interacted with a beautifully built and tuned Deep Learning algorithm, like when you searched for a picture of a dog on your phone, or when Alexa understood and answered your 3 year old daughter question “what is 2+2”.

If you’re part of this connected era It’s quite impossible to not have experienced all the wonders of Learning Systems. For C Suite executives however, the chances are Data Science has not made its way to their decision making. Sure there are those weekly reports about a forecast, or the occasional analysis outlining a possible strategy. Those reports however, tend to fall short of exposing new information or helping an executive identify a blind spot.

How often does a CEO try to optimize a business objective through an out of sample prediction or by interacting with a dashboard that is connected to an optimization method? It’s fairly rare to have that.

For me introducing Data Science to the C Suite is about recognizing that some business decision making are ripe for optimization and improving.

In this 3 part blog I will highlight a tactic, in each part, for embedding data science inside a company’s C suite. I hope this will help you bring data science to your executives and enable the prowess and effectiveness of Machine Learning at that level.

Keep it simple

It is unlikely that your company’s CFO needs to access image classification, speech detection, or natural language processing algorithms regularly. This means many out of sample predictions a C suite executive needs can be formulated with parsimonious models and then passed through a simplex optimization for business strategy execution. For the sake of this discussion, I am defining parsimony to be non DL models, e.g. Decision Tree ensembles, Generalized Linear Models, etc…

Data science for the C suite is about enriching the decision making. With that in mind if this is a C suite first foray in data science driven decisions, the chances are that acting on a machine learning based strategy implementation requires trust. Trust, that the “black box” an executive is interacting with will lead to better outcomes. Hence starting with simple would be a good way forward to demystify how data science works.

How will this play out? For example, running capital intensive businesses requires periodic review of capital allocations. How the capital gets allocated and in turn which parts of the business to grow will make a difference between long term success and failure. For example the COO of a Real Estate development corporation, might be interested in prospecting lower cost and high value areas for new developments. An interactive Geo dashboard where prospects valuations are established through a Learning Algorithms will be a nice first step. But what if this dashboard allows infusing Macro economics to generate scenarios for cost of land, labor and material, interest rates, and an area’s economic prognosis? Now we have prospects costs and valuations tracking over time. The COO likely has a strategy in mind. Strategies can be turned into objectives, e.g. maximize 5 year net profit. We can now take this one step further with a simplex optimization that will recognize best opportunities based on such strategies.

To be sure, I have seen C suite executives working through those processes of capital allocations. Except such processes stretch over weeks of back and forth and in many cases they are non repeatable. What I am describing here is real time, happening through an interactive dashboards and workflows. This can be seen as a digital twin for a business process and decision making.

This is still a hard problem. If this is new to a C suite team, it need not to be abstracted even further with a Deep Learning model and a genetic optimization algorithm.

That’s not to say that a Data Science team should not progress to Deep RL and Genetic Optimization algorithms for building those digital twins. Until then it pays to keep it simple.

It takes time, trust and some early wins to get most executives comfortable with augmenting their decision making and strategies through machine learning. In part 2, I discuss another strategy for bringing data science to the C suite.