Business Analytics – Prescriptive analytics

INESC TEC Science Bits – Episode 1

PODCAST INESC TEC Science Bits (22:52 — 31.4MB)

Guest Speakers:

Bernardo Almada Lobo, Centre for Industrial Engineering and Management

Gonçalo Figueira, Centre for Industrial Engineering and Management

Keywords: Business Analytics | Data mining | Big Data | Machine Learning | Statistical Analysis | Business decision making | Mathematical Optimisation

CONTENT SUMMARY

What is Business Analytics? 

A set of analytical methodologies that enable the creation of business value. It is evidently data-driven and it should impel business action. Even though the ultimate purpose is to make better decisions and empower decision making, the research and development on analytical methods involve different points, and are employed in different settings: to get improved information and  build on top of large sets of data; to get better predictions; to focus on managerial problems and to reach better decisions.

How does a company or organisation apply Analytics

Data is one of the most crucial points and having quality data is mandatory. Otherwise, the results will be useless. The type of Analytics will depend on the main issues and challenges the company wishes to address. If a manager wants to understand what is happening in a plant, in a production line, or in a specified machine, one should apply Descriptive Analytics. Predictive Analytics can be useful to understand what might happen or what is likely to happen. Lastly, Prescriptive Analytics will actually propose solutions, such as how to produce or when to perform maintenance actions.

Sequence in which the different types of Analytics should be conducted.

In certain ways, Prescriptive Analytics benefit from the results of Predictive Analytics, which, in turn, uses the output of Descriptive Analytics. The practical application of Prescriptive Analytics is much more challenging, since it suggests comprehensive solutions to the problem, not just insights/foresights about the system. Sometimes, we resort to Prescriptive Analytics right away, because the previous layers (descriptive and predictive) are already at a good level, or quite straightforward (they simply do not require sophisticated analytical methods). In other cases, we find out that it is not exactly like that, and we ought to strengthen the descriptive/predictive components. 

Analytics is just one of the stages of Business Analytics

A Business Analytics project is an endeavour with several building blocks. First, we need to put in place the ability to understand a business problem and determine whether said problem is ready for an analytics solution. In other words, we start by framing the business issue. Then, we need to translate the “what” of the problem into the “how” of the analytics problem; in this case, we are actually formulating the problem statement as an analytics problem. There is a trade-off to make between Accuracy / Simplicity / Speed / Flexibility / Scalability.

Application examples of Prescriptive Analytics in Business or in Industry

Optimisation problems emerge everywhere. Moreover, the real-world issues are difficult to solve. In certain cases, the number of possible solutions in the search space is so large that it hinders the thorough search for the best answer – we call them combinatorial optimisation problems.

At INESC TEC, we have been solving managerial problems in several fields (manufacturing, healthcare, retail and mobility), with a special focus on Operations Management. Some of the problems are more strategic in nature, such as network design (where to fix facilities, plants, warehouses, etc.,); others may be tactical – such as operations planning (related to workforce, capacity and distribution, inventory management, maintenance policies for asset management, etc.) – and operational in nature i.e. vehicle routing, production lot sizing and scheduling, among others.

Prescriptive Analytics as a set of mathematical models that ultimately can help the managers and CEOs in decision-making

Prescriptive Analytics (also known as optimisation) is not only about mathematical models. One of the approaches is mathematical programming. The second approach is based on heuristics that describe the problem through an algorithm – essentially, a program that produces solutions very quickly, despite not guaranteeing optimality. Analytics will definitely create value on guiding tactical (and strategical) managerial decisions. With more access to useful data, companies are gradually using more sophisticated analytical methods. Nevertheless, it is possible to verify that advanced analytics are not sufficiently adopted, especially Prescriptive Analytics.

Low adoption of advanced analytics, especially Prescriptive Analytics.

One reason for the low percentage of implementation is the considerable number of unsuccessful prescriptive analytics projects. A successful prescriptive analytics project requires a buy-in of all stakeholders involved, from top-level management to data owners, employees affected by the change and end-users of the new solutions and analytics processes.

In many situations, managers are reluctant to apply recommendations from a system/model of which they cannot understand the core reasoning. There must be no concealed elements behind the technology/tool. Managers ought to understand that “the system will help making faster decisions” rather than “the system will make better decisions for them”. It is not realistic to presume that human intuition and the expertise factor can be eliminated.

Tools for Prescriptive Analytics

The core is optimisation methods, which include a variety of different paradigms, like mathematical programming, constraint programming, heuristics and metaheuristics. Then, there are other techniques, like multi-criteria decision making methods.

In addition, it is possible to have a hybridisation between two optimisation methods, like metaheuristics and mathematical programming, the so-called matheuristics. Alternatively, we can even combine optimisation with other methods e.g. combining optimisation with simulation is one of the most interesting approaches.

Future developments in Prescriptive Analytics

This field is moving forward in a variety of ways; one is researching the combination of simulation and optimisation. Talking specifically about optimisation, over the last three decades, we have witnessed a substantial amount of work on approximate optimisation methods (the so-called heuristics and metaheuristics). Problems got more complex, so we had to give up on optimal solutions in some cases. However, these methods are also being combined with exact optimisation methods, to accelerate the time we take to get optimal solutions.

Prescriptive Analytics and the Industry 4.0

In the context of Industry 4.0, problems are getting even more complex than before. Systems need to be decomposed in sub-systems, and decisions must to be decentralised. The activities are being automated and the decisions are made in real-time. To achieve that, solving a mathematical model, or even running a metaheuristic is not a viable option, because it requires several minutes. Machine learning (ML) is a key tool for that purpose, because we could make the process of learning new rules automatic. So far, ML has been mainly used for descriptive and predictive analytics. Now, prescriptive analytics is expected to become a central topic in ML, which will be cross-fertilised with optimisation techniques.