Dr. Jeffrey McGill also contributed to this article.
The manifestation of big data and analytics is prescriptive analytics. How does your business make use of the data it collects? Yuri Levin and Jeff McGill discuss the integration of prescriptive analytics and what it means for businesses.
In a previous Knowledge Exchange interview, Anne Robinson described the analytics sequence:
Most users of business analytics have experience with the first two in this line-up, but what about the last?
Prescriptive analytics proposes decisions that can improve performance. These can be small decisions like how to schedule staff to minimize costs while meeting service requirements, or large decisions like which projects to undertake to maximize benefits to the organization. Whatever the context, prescriptive analytics offers real potential to deliver value.
In isolation, there is nothing new about ‘prescribing’ the best decisions – fields like Management Science, Operations Research, and Industrial Engineering have been developing such optimization methods since the beginning of the last century. Unfortunately, many of those methods, which seemed so promising in theory, required levels of data quality and quantity that were unattainable for many enterprises. This has changed dramatically in recent years – there is a rapidly expanding range of methods that are now feasible because of more and better data and readily available tools that can do the analysis. Good examples of some tools can be found is this white paper.
There have been great successes in predictive analytics over the years but also many failures. Two important lessons have emerged.
First, the world of management is far too complex and chaotic to realistically expect ‘optimal’ solutions – in most cases it is more reasonable to hope for ‘better’ solutions. Prescriptive analytics solutions should make base-case suggestions that experienced managers can use to inform their thinking. They may choose to accept the model’s suggestion or adjust the solution to take into account factors that the computer could know nothing about.
Second, integration of the descriptive and predictive components is critical – you are not ready to move on to prescriptive tools until you understand what has been happening and can estimate what might happen in the future.
Prescriptive analytics is the final phase in the development of management analytics – if you have a good handle on what is happening and why (descriptive analytics) and a good model that can estimate future outcomes for various possible management actions (a type of predictive analytics), then you may be able to use an optimization tool or model to search through the ‘space’ of management actions for those that lead to better outcomes for the organization. Integration of the Descriptive –> Predictive –> Prescriptive information chain is a key to success.
There is a special requirement on the predictive model mentioned above: it does not passively estimate what is likely to happen given past trends, rather it actively incorporates management decisions into predictions. If such a predictive model has credibility with managers, they can run scenarios with different decisions to explore outcomes. In effect, they are trying to search by trial-and-error through the space of solutions for a better one. But if the predictive model has that level of credibility, why not turn over the job of searching through decision space to the computer – they’re really good at that sort of work!
Big data, analytics and skills
It is important to recognize that big data and analytics are not the same thing. There is great excitement over the possible uses of WWW sources and data warehouses, and analytics has a role to play in developments, but analytical methods can work very effectively with ‘little data’ as well. Something as simple as a spreadsheet model coupled with a simulation add-in can achieve real benefits and serve to convince managers of the potential for larger solutions.
The skill levels required should not be underestimated – successful implementations of prescriptive analytics require understanding of modern data retrieval and cleaning technologies, statistical analysis, mathematical model-building, and optimization. Typically a team is required to cover that range of expertise, hence communication and cooperation is critical. Ultimately, the decision to expand the use of analytical tools in any organization is a strategic one, and there is a significant challenge in encouraging senior management awareness and support for such a strategy.