Prescriptive analytics: Just what the doctor ordered, but a little too early for all

The ability for systems to make decisions, not just predictions, is at the heart of prescriptive analytics. While often regarded as the final analytics frontier, with multiple benefits to bring to organizations implementing it, rolling out prescriptive analytics can come with some challenges. In our 2017 Trends in Data Platforms and Analytics report, we referenced the fact that prescriptive analytics was an evolving discipline. Here we look at the some of the reasons why this form of advanced analytics is still nascent and only currently adopted for select use cases, in certain industries.

The 451 Take

We are still at the early stages of adoption for prescriptive analytics. While it will eventually become as commonplace as other types of business intelligence, that isn't likely to occur for a number of years, for the reasons outlined below. In the meantime, companies considering it should examine the offerings available from the handful of vendors delivering prescriptive analytics, and bear in mind hidden expenses such as data management, which is required to get data fit for purpose, and consulting, which will drive up costs.

Predictive analytics is now commonly understood, but it is fair to say that prescriptive analytics, which is implemented at the same time – or after – its predictive counterpart, is not. Put simply, prescriptive analytics turns a prediction generated by predictive analytics into a decision. In other words, prescriptive analytics is all about enabling companies to answer the questions, 'What should be done?' and 'What can we do to make this happen?'

Prescriptive analytics can provide companies with a new way to improve the confidence they have in their decision-making processes and business outcomes. Implementing this type of analytics can therefore result in a number of boons: improved efficiency, reduced costs, the ability to create new revenue opportunities, and the ability to take customer satisfaction and loyalty to new levels.

Prescriptive analytics in action
A simple, real-world example could occur in the credit card application process. A person applies for a credit card online. Based on the information received, the system decides a prescribed course of action (i.e., whether to make a special offer, approve the application or defer the offer).

Outside of the financial services industry, prescriptive analytics has been mainly adopted in oil & gas, healthcare, and manufacturing and logistics. For example, this analytic approach is used to prescribe how and where to drill for oil in order to enable operators to produce more oil, more predictably, and at a lower cost. Healthcare providers have also adopted prescriptive analytics in order to improve the clinical care they provide patients, and in so doing achieve better patient satisfaction and retention rates. However, while there are more citable implementations and use cases than a decade ago, when prescriptive analytics first came into being, only early adopters and organizations with high analytic maturity were using this discipline.

Prescriptive analytics is difficult. Indeed, it could be argued that it is the hardest form of analytics to get right. For starters, companies need to have sufficient data volumes and types to make correct prescribed decisions. Only a handful of organizations and industries have the massive volumes of data required to get something useful out of prescriptive analytics, although the rise of the data-driven enterprise will mean this issue is likely to evaporate in 5-10 years.

Companies also need an array of technologies, and sufficient computing resources and infrastructure. The increased speed and memory size of computers, along with a growing acceptance of cloud infrastructures, has certainly helped to reduce prescriptive analytics' processing demands. Nonetheless, it still requires a significant amount of horsepower because of the optimization, simulation, complex event processing, neural networks, recommendations, heuristics and machine-learning algorithms involved.

Furthermore, off-the-shelf prescriptive analytics offerings available for general purchase and use are in the minority. IBM, SAS Institute, FICO, Ayata and River Logic are the main purveyors of prescriptive analytics. There are a few other players focusing on specific use cases, such as CognitiveScale and LeanDNA. However, by and large, prescriptive analytics is still something provided either by an enterprise software veteran or a mature pure play (Ayata, River Logic). Startup activity has yet to get into gear, although we expect that to change in the next few years.

Companies that want to roll their own, using a combination of existing products, should also be aware of the cost involved in this approach. It will involve crafting a software stack to analyze potential decisions, the constraints on each decision and the ultimate business outcome of each scenario, which can be prohibitively expensive and highly time-consuming. Furthermore, consulting to tie together all the requisite components, or customize them for a particular company's business, can also be required, which will add another expense and further length to the project.

Lastly, there is ongoing monitoring and maintenance to consider. While the use of machine learning and other closed-loop feedback technologies in prescriptive analytics means that these systems will adapt and learn as they are used, real-world deployments will still require some regular revisiting. Why? Businesses are not static and immutable. They change over time. Therefore – at a bare minimum – organizations need to ensure that the rules and hypotheses they use as a foundation for prescriptive analytics continue to match the goals of the business. If they are no longer relevant, the prescribed course of action will be out of date and useless.

Krishna Roy

Senior Analyst, Data Platforms and Analytics

New Alert Set

"My Alert"

Failed to Set Alert

"My Alert"