Data Science’s Last-Mile Problem in Corporate Strategy

Much has been written over the past few years about the challenges that data science teams face in corporate environments. If you Google “data science last mile problem” you’ll be met with dozens of articles and even a few companies focused on the problem of how to maximize the impact of data science projects.

The intersection of data science and corporate strategy is where I’ve spent much of my professional career. Based on my experience both as a strategy consultant and data scientist in industry, I believe that data science’s last-mile problem in corporate strategy is somewhat different but highly solvable. In this article, I’ll outline what I think data scientists and strategists can do to close this gap and maximize their impact.

What Data Scientists Need to Know About Strategy

Bias Toward Action

As a young strategy consultant, I worked with one of my clients to better understand their pricing behavior using a machine-learning model. One day, the CEO dropped by one of our working sessions to see how the project was coming along. The material we were going over was somewhat technical, and we gave an update in terms of the model’s goodness of fit and what we were doing to improve it. The CEO asked, “Well, how good does that metric have to get before I can use it?”

I stumbled through a response, but I had already learned an important lesson: Whatever you’re working on should lead to some action being taken, and you should be able to discuss your work in terms of that action. If additional work is unlikely to lead to a different set of actions, then it may be time to focus on other projects. If you’ve lost sight of the actions that will be taken based on your work, take a step back to clarify the purpose of your work and assess its impact.

All Models Are Wrong – Some Models Are Useful

We data scientists choose our profession because, among other reasons, the tools and techniques used in data science are interesting. We may feel a bias to reach for the most advanced types of models with which we’re familiar. But the most advanced models are not necessarily the most useful. This is especially true in corporate strategy. Models that use intuitive inputs and are easy to explain are more likely to be useful than sophisticated black boxes that few people on your team can understand, let alone maintain. Strategy consultants routinely use very simple models and frameworks to great effect. This is not to say that the full suite of data science tools cannot be brought to bear on corporate strategy problems. It can, but it falls on the data scientist to use the right tool for the job, whether that’s a support vector machine or a simple two-by-two matrix.

Ownership Is the Expectation

Technical knowledge and coding skills are important parts of the data science profession, but they are not always the primary differentiator between data scientists who have an impact and those who don’t. In my opinion, the primary differentiator is the ability to own a project from ideation to implementation to executive communication. The most impactful data scientists I’ve met are those who, in addition to applying their technical skills, can also manage a project, create a good PowerPoint presentation and explain the implications of their team’s research to upper management in a way that resonates and leads to action. If that doesn’t describe you, then seek out mentorship in those areas and ask for opportunities to exercise those skills. Regularly asking for feedback is a great way to assess your current opportunities for improvement along those dimensions.

What Strategists Need to Know About Data Science

Theorists and Experimentalists

Most scientists in academia have one of two primary mindsets: theorist or experimentalist. Each needs the other, and imbalances impede progress. I’ve seen both mindsets at work in the strategy offices of many of the companies.

Theorists tend to emphasize deeper thinking about the available data and connections to fundamental principles that can be applied more broadly. They tend to publish their results less frequently, and often to a more niche audience.

Experimentalists, on the other hand, produce new data in real time, typically with quick, surface-level analysis. They tend to publish their results frequently and to a broad audience.

If your team gets feedback about strategy-related research that is lacking in depth, it’s likely that the experimentalist mindset is overly dominant in your team. Conversely, if you hear concerns about “analysis paralysis” and a lack of actionable insights, you may be over-indexed on the theorist mindset. In both cases, the solution is to think clearly about the needs of your corporate strategy process and then carefully align your portfolio of data science projects with those needs.

Strategy Work Uses Machine Learning Differently

Regression and classification are often the first types of analyses that many data scientists are exposed to, and they are probably the most common types of models in use by data science teams. Although they can have many applications in corporate strategy, data scientists in a strategy environment will likely need to flex different muscles. Specifically, constrained optimization, stochastic simulation and forecasting are likely to be important because they deal directly with the uncertainty inherent in corporate strategy.

Strategists can help close the gap by being explicit about strategically important assumptions and the levels of confidence they have in those assumptions. Constantly ask yourself “what must be true?” for a given strategy to be successful. Then, you can work with your data science colleagues to quantitatively evaluate the implications for your organization.

The Scientific Method is Still Supreme

With all the buzz around data science, generative AI and other shiny new types of technology, managers and strategists might wonder if it could give them a crystal ball and fundamentally transform the way in which strategy is developed. The truth is it probably won’t. New technology should be used based on its ability to drive clarity of thought among decision-makers, develop and test hypotheses about the market, and interpret data and industry developments in the context of your organization’s objectives and values. In other words, the role of data science is not to replace the role of strategic thinking but rather to enable the scientific method to flourish within the corporate strategy organization.

If data science isn’t reaching its full potential within your strategy organization, it might be time to return to basics: Form a specific, testable hypothesis about an important question, and work with your data science team to develop an actionable strategy to test that hypothesis.