Using data science predictions to drive strategy

A couple of people asked me the same question, "how do you use data science predictions to drive strategy", and unfortunately for the majority of audiences, I don't have a cookie-cutter formula that will enable coin-machine operators to crank strategy using data science. But there are a few definitions and frameworks that one can use to leverage data science to support strategy decisions.

First, understand the elements of the two. Data science itself has a whole bunch of components on its own. I like this chart by Drew Conway: http://drewconway.com/zia/2013/3/26/the-data-science-venn-diagram, that emphasizes the multi-disciplinary approach in order to make statistics relevant. First, you need a good understanding of statistics - what are the underlying assumptions of a model, what does it tell you and what it does not. Second, you need some technical skills to make sense of data, improvise around data to create a usable input into stats models, as well as take the output and turn it into something usable. Third, business knowledge, and this is where its key to leverage business' leaders questions and opinions about key levers that will influence business. 

Strategy has been the mainstay of consulting firms and corporate strategy for decades. There is also a lot of business strategy decisions in decentralized teams across companies. While centralized teams often focus on longer term strategy (3-5 years and up), decentralized teams typically focus on short-medium term strategy (~1-3 years), how to beat the competition now and in the foreseeable future. Strategy frameworks typically contain many levers, each of which has some variables - whether from data or qualitative options.

The next question to ask then, would be, which levers and variables would data science predictions help us understand levers and estimate better. Data science is inherently backwards looking and returns a probabilistic result as opposed to a deterministic outcome, for the uninitiated, that's geek speak for there's a X% chance that a customer might buy, or a range of values that the market will grow into ($50-70M), as opposed to a definite number, yes or no based on a set of criteria. Data science is often used to automate decisions where there is a clear, incremental option - i.e. A/B testing, web feature selection, and we need to be careful not to expect these quick-wins in something that is ambiguous and uncertain in nature.

By selecting the right variables to predict, data science can add a forward-looking prediction on the next period (with diminishing accuracy over a 1-3 periods), which makes it a powerful tool in replacing historical plus broad-based growth estimates and brings us one step closer by having a better picture of what will happen if we don't change anything.

A second, more commonly utilized method, is to leverage high impact variables inherent in statistical models. Analyzing the highest impact variables in the model and creating programs that moves those variables that result in a better predicted outcome is the mainstay in most customer churn programs. Imagine doing the same with your revenue predictions and market data science models that allow your various business lines to figure out next best opportunity.

The best part is, you probably already have a strategy framework and understand your business levers and variables very well. If you have the right people in your strategy team - a statistician/data scientist, a good understanding of data that's available, a strategist, and someone who can overlap across statistics and business, the opportunities for creating value are immense and only limited by leadership mindshare.

Comments & feedback are appreciated.