Consider the following statistics: 

1. Google search volume for the term has increased three-fold since 2014.
2. Press coverage has grown exponentially. Here are three examples:
   • Forbes has written 37 articles containing the term "prescriptive analytics." The first time it appeared was in November 2014, and three of the 37 posts have been published in the last 30 days.
   • Information Week has already published almost as many articles containing prescriptive analytics in 2016 as it did in all of 2015.
   • Of the 867 LinkedIn posts that refer to prescriptive analytics, 75% were published within the last year.
3. While a year ago only specialized analysts covered the category, it’s now covered by all major analyst firms such as Gartner and Forrester.
4. Gartner predicts that its use will grow from 10% of organizations today using some form of prescriptive analytics to 35% by 2020, making it the fastest growing category of software.

This growing amount of attention is due to the realized value of prescriptive to optimize decision making and planning processes by prescribing actions (versus simply predicting outcomes, displaying data, etc.).

Back to arguing with analysts — let’s now have a look at how Gartner defines prescriptive analytics.

What is Prescriptive Analytics?

Gartner defines prescriptive analytics as:

“…the application of logic and mathematics to data to specify a preferred course of action. While all types of analytics ultimately support better decision making, prescriptive analytics outputs a decision rather than a report, statistic, probability or estimate of future outcomes.”

Just as it sounds, prescriptive analytics prescribes action. However, Gartner makes matters slightly more confusing by breaking it down into two categories — 1) optimization and 2) decision automation through real-time execution of heuristics (i.e., rules). 

Ok, now we have two more definitions to explore that relate to prescriptive analytics — let’s have a look at those to see where Gartner went wrong.

The Optimization-based Approach to Prescriptive Analytics

In using the term optimization, Gartner is referring to applying mathematical optimization to solve various problems in different types of organizations (i.e. commercial, government, non-profit). In other words, most decisions in an organization have multiple possible answers.  Some are relatively straightforward but most have thousands or millions of possibilities — and they all must meet various objectives simultaneously (i.e., financial goals, customer service goals, employee retention, etc.).  Furthermore, they typically hav­­e cross-functional impacts, are constrained by business realities (i.e., budgets, demand curves) and require fast decisions. 

Applying prescriptive analytics through optimization enables users to wade through all these factors and find the path that meets the most objectives given the defined business .

Decision Automation: The Rules-based Approach to Prescriptive Analytics

In Gartner’s second definition, it claims that prescriptive analytics can be achieved through the application of rules for real-time, automated decision making.  This often requires the combination of predictive analytics (i.e., statistics) with rules that are often delivered through an execution or business process management system. Essentially, it means that when something happens, the system will decide what to do on the fly, given the rules that have been plugged in. Just picture a workflow — rules are predefined and so are the possible outcomes. (Note that this is quite different from optimization).

Here’s an example: Someone applies for a credit card online. Based on the information received, the system decides whether to make a special offer, approve the application, decline the offer, etc.

This example is very clear — however, it’s not clear where the prescriptive component lies.  Much like in day-to-day execution systems (order management, manufacturing execution, CRMs, etc.), the real-time execution component is simply following rules.  Unlike optimization, the rules are not exploring all paths but rather choosing between paths that are predefined.

The real decision is not which rule path is followed — the real decision is what the rules are to begin with. In other words, the prescriptive analytics part of real-time decision automation lies in determining what the rules should do, not what the result is. Potential trade-offs, business constraints and objectives are taken into account when defining the rules, not when executing the rules (i.e., making decisions in real time).

In summary, according to Gartner’s definition of prescriptive analytics, this automated decision making is still considered prescriptive analytics (i.e., some math plus some prescribed action). So it seems the problem lies in Gartner distinguishing between optimization-based and rules-based approaches to prescriptive analytics. What Gartner calls a rules-based approach to prescriptive analytics is actually nothing more than decision automation of pre-defined rules — it is in no way prescriptive analytics.

In fact, optimization (and to a lesser extent related approaches such as goal-seek or Monte Carlo simulation) is the only approach to prescriptive analytics, whereas Gartner’s rules-based approach uses data that ideally comes from optimization in the defining the rules for real-time execution / decision automation. 

Optimization is complementary to Decision Automation

Let’s go back to the example of the credit card application.  The real decision is how to set up the rules – and this is where prescriptive analytics comes in. If an applicant shows up that looks great, what should the company offer her?  What if there are already too many similar applicants — should the offer be withdrawn?  What if the applicant doesn’t meet all the requirements? At what time should the company use personnel resources for live interaction with the applicant?  Is there a point when that capacity runs out?  Should the company change marketing tactics based on who is showing up to apply? 

These are the real decisions that take the form of rules when translated into automated execution.

In the end, prescriptive analytics is really about making decisions. Optimization is the approach used to prescribe action, including defining rules.  Decision automation is about executing on those rules that can (and should) be defined through prescriptive analytics.

Optimization might be used if decisions define strategic actions (i.e., which products/services to offer and in which market?) or tactical plans/policies (i.e., how many staff do we need of each type to meet our SLAs?  How should we deploy our marketing or trade promotion dollars?), or operational decisions (i.e., how do we allocate online ad dollars on a day-to-day basis? What is the best way to schedule patients in the operating room?).      

The Value of Leveraging Prescriptive in Your Planning

Now that we’ve re-worked the Gartner’s definition, let’s jump to the topic of market interest.  As organizations have gathered their data, used descriptive/diagnostic analytics to understand what is happening and applied statistics to look into the future, they have realized there is still a significant gap between the information they have and the implications on what they should be doing.  

Prescriptive analytics not only solves this problem, it brings tremendous value in the process.  For example: 

• Organizations are able to increase service levels (i.e. on-time in full service) by 10%+ without any additional resources.
• Companies can increase profits by 1-5% of revenue.
• Risk is better understood and can be explicitly connected to business outcomes.
• In some areas, addressable costs can be reduced by 15-20%. 

Historically, many of the applications of prescriptive analytics have been limited to a single function (i.e., optimize truck deliveries).  Today, however, prescriptive analytics is being applied across business functions to support a much broader range of operational, tactical and strategic decisions.  

What do you think… is the hype real?