22 January 2017
Sometimes “Data Science” triggers associations with “academic”, as in: theoretic, not practical, of limited practical usefulness, etc. There are two sides to that concern. From one end, Data Science often resembles R&D work in that you never know if and when it will pay off. A pharmaceutical company will test many compounds, few of which will eventually hit the market. It’s a bit short-sighted, but you could consider all the other efforts “wasted.” Similarly, Data Science projects don’t always pay off: you need to manage those efforts much like you manage a portfolio of compounds that a pharmaceutical company is testing in its early stages. Mitigate costs and risk by repeatedly evaluating progress, and be willing to cut your losses by canceling projects that don’t bear fruit.
On the other side, Data Science does bear resemblance with academics. Similar methods for experimental testing and control are applied. But hey, there is nothing as practical as a good theory. Deming, among others, started championing the scientific method a long time ago: gathering data using objective methods, using hypotheses and statistics to test effects of management interventions, etc. We do not want to “compromise” here, as the scientific foundation provides compelling evidence that can greatly help a company move swiftly toward their more profitable sweet spot.
To ensure that Data Science projects are well aligned with broader management goals, it is important to be aware of corporate strategy. Do your best to get clarity on “global” strategy as you plan and prioritize your resources. This alignment is best accomplished by an objective assessment of the current position in the marketplace, maybe as the outcome of a BI audit. Analyze what the company considers “success” internally, where profit is coming from today, and where future growth can be expected.
One of the models that I have always found useful for this purpose was proposed by Treacy & Wiersema (1997). Like all models, it isn’t “correct” per se. All models are wrong, some are just more useful than others, as George Box noted. Treacy & Wiersema discern three fundamental strategic directions: Operational Excellence, Product Leadership, and Customer Intimacy. Some combination of two can certainly be possible. Their premise is that you shouldn’t attempt all three, though, because that harms your ability to distinguish yourself from the competition.
If there is (relative) clarity from corporate headquarters on the overarching strategic direction, great! However, if your company does not express its strategy with razor sharp clarity, then take a stab, and work from there. It is perfectly fine, in fact desirable, to defer refinement of Data Science project requirements iteratively. In an Agile fashion, these analytic projects can serve as building blocks. Separate projects likely will surface consistent findings, and this corroboration then points to opportunities for commercial success. As you surface more empirical evidence, and fine-tune strategy and execution, more opportunities will present themselves. All these project outcomes can be placed in light of commercial developments like market shifts, M&A activity, etc.
Getting clarity around strategy, and making (explicit!) choices in light of your department’s direction, helps you build efforts on top of each other. Many roads lead to Rome.