I don’t know who or which analytics company started it, but we often hear that “prescriptive analytics” is at the top of the analytical food chain. I’m sure anyone who is even remotely involved in marketing analytics has seen that pyramid chart, where BI and dashboard reports are at the bottom, then the descriptive analytics next, and then predictive analytics, and the prescriptive analytics on top. I, too, often differentiate different types of analytics, as they all serve very different purposes at different stages of marketing (refer to “How to Outsource Analytics”). However, I wholeheartedly reject the notion that so-called prescriptive analytics should deserve the position at the top of the pyramid.
If the chart meant to display “difficulty” of conducting various types of analytics, it may make some sense. Unfortunately, the users do not see it that way, but they see it as a roadmap. As if one must graduate lower-level status to move on to a more advanced level of analytics (I don’t really care for the term “advanced analytics,” either). That kind of hierarchical order surely is not, and should not be the case in the analytics business.
Because that would be like saying a doctor should not prescribe any medicine until she exhausts all possible procedures and options, in a pre-set order, to treat the patient. That would be just stupid. Doctors should be able to provide a cure if there are obvious visible symptoms, without conducting the most complicated (and expensive) tests first. Likewise, analysts must be able to prescribe solutions even from a one-page report, or a note on a napkin, for that matter. That means if we really want to use the term “prescriptive analytics” somehow, it should be attached to all types of analytics, at all times.
Now let’s break down what “prescriptive” in analytics means, anyway. We often say getting the “insights” out of data or reports is the essence of modern-day data science. But being prescriptive means analysts must go one step further and suggest what the next steps should be. If one sees apparent or not-so-apparent troubles or issues brewing, what are we going to do about it? Answering that question would require multi-dimensional thinking for sure, and such activity is a lot more difficult than building statistical models. For one, it is not purely about statistics and data manipulation, but about business practices, as well.
That is why I said the pyramid chart may make some sense if it’s meant to represent difficulty levels. The main reason why finding a decent data scientist is so difficult is deriving insights out of mounds of data and suggesting next steps are not just a mathematical challenge. Dealing with small and large data is a difficult task in itself, learning programming language is a lot like learning a foreign language, and mastering statistical knowledge is definitely not for everyone. Knowing how many people are severely allergic to foreign languages “and” mathematics, we are talking about a fraction of the population in America who would pass those two hurdles in the first place. Now we are adding that the analysts must possess business acumen to advise senior-level executives without using any scientific jargon (refer to “How to Be a Good Data Scientist”).