Turning the Promise of Prescriptive Analytics Into Reality

Predictive analytics offers plenty of value: it can tell us which prospects are likely customers, what might cause them to buy, or when projects are likely to run into cost overruns.

A recent survey of business leaders found that although 99% of Fortune 1000 companies are investing in data and AI, only 30% feel they have a well-articulated data strategy.

This is remarkable when you think about it. For the past 50 years, businesses have been amassing ever more sophisticated systems to collect, process, and analyze data with the goal of producing actionable insights and improved decision-making. Yet at the end of the day, 7 out of 10 of them don’t even have a strategic framework for using all the data they have.

When you consider the approach most companies are taking, this is also unsurprising. Today, companies have excellent tools for predictive analytics, which enables them to view past patterns to predict future results. Predictive analytics offers plenty of value: it can tell us which prospects are likely customers, what might cause them to buy, or when projects are likely to run into cost overruns. Such insight can also produce action. It can guide marketing tactics, help with preventive maintenance, and optimize sales processes.

But while business leaders may be rewarded for incremental gains, the insights of predictive analytics cannot tell you the advantages and consequences of major strategic decisions: where it makes the most sense to open a plant, what large-scale capital expenditures will be most valuable, or whether to launch an adjacent product line. And maybe the most elemental thing they cannot do is to tell you is what changes if you make a decision.

Those kinds of insights fall not into the realm of predictive analytics, but rather prescriptive analytics. The difference is that the former can tell you what is likely to happen, while the latter tells you how your activities can affect that outcome. At its best, it can even suggest the actions that you should undertake to have the best results. According to Gartner, it can give you answers to really important questions, like “what can we do to have this happen” or “What should we do?”

While prescriptive analytics is not exactly a new concept, it has so far had only limited (though occasionally spectacular) implementations. Amazon, for example, uses it to preposition products before customers are even aware they’ll want to buy them. Airlines use it to increase the uptime and efficiency of their maintenance operations. And the oil and gas industry has used it to find the most likely locations to drill wells.

While these isolated examples are successful, they also tend to be specific and don’t necessarily generalize. Most CEOs would like to have prescriptive insight to make decisions on a daily basis, but in order for that to become reality, we’ll really need three things.

Recommended Data Science News DataRobot Core Unveiled, Complete With Capabilities For The Expert Data Scientist

The first is data.

The reason prescriptive analytics works for geophysicists looking to locate a natural resource is that they can deploy ocean surveys, seismic graphic data, capital cost information, and so on to fuel the models that make decisions. They have literally terabytes of structured and unstructured data to process to understand whether Site A is a more likely candidate than Site B.

Until recently, most companies in other fields have not been so data-blessed. But that is changing. As they and third parties amass more and more data, the gap has narrowed, and we are approaching a tipping point where this kind of analytics is much more feasible.

The second problem is models. In order to perform prescriptive analytics, you need a modeling framework in which to place the data. Right now, if you have the time and data, you can the dots using existing data and widely available technology like Excel or business intelligence (BI) platforms. But the process is both time-consuming and costly. However, the future is bright for companies willing to explore this approach on a larger, more systematic scale — and quite possibly by making available such models on an open-source basis.

Of course, will also need tools to manipulate these models and analytics packages to support them. However, the overall concept is not something new: we simply need to move from the current, restricted applications to the more widespread, shared model common in many other fields.

When that happens, we will be on the brink of a much wider application of the kind of prescriptive analytic frameworks that CEOs have always wanted and even expected from the data they have. With them, they should be able to break the current logjam which makes it difficult to bring the massive amounts of data we have collected to bear on a much wider range of strategic problems.

Top Telehealth Update: Top Telehealth Trends for 2022

[To share your insights with us, please write to sghosh@martechseries.com]