Presenting recent books: “Business analytics” by Mischa Seiter
Problem solver: New insights
The availability of data has grown enormously - and with it the pressure on executives to use this data and gain new insights from it. In his latest book, Mischa Seiter addresses the question of how data can be efficiently obtained and used to solve business problems.
Mischa Seiter follows the business logic behind his business analytics approach and asks four key questions, each of which is dealt with in a single chapter:
- For what type of problem should business analytics resources be used?
- Which resources (data, IT and human) should be used?
- Which analytics algorithms are used to obtain evidences (data-derived insights) to solve the business problem?
- How should the evidences be prepared so that the senior management can optimally use them to solve the problem?
Framing – Narrowing down the problem
The first step is to pinpoint the business problem and come up with the associated analytics question. For example, a typical problem for many companies is a decrease in customer satisfaction. The business problem in this case is increasing customer satisfaction to a standard that is normal in the industry. For that to happen, it is essential to know the factors that influence customer satisfaction. So the analytics question is this: Which factors influence the satisfaction of customers? In order to use most limited analytics resources meaningfully, the relevance of the business problem needs to be evaluated. The benchmark for this is usually the effect that is to be achieved by solving the problem. Then the analytics problem is deduced.
There are three types of analytics that cover different fields:
- Descriptive analytics identifies unknown patterns, e.g.: What are the characteristics of our A-customers? Are certain products or services usually sold together?
- Predictive analytics constructs a forecast model, e.g.: When will a machine stop working? Under what conditions will a customer cancel?
- Prescriptive analytics constructs a model for the best outcome, e.g.: How to distribute spare parts in the warehouse to reach a certain service level? Which product price leads to the highest profit for a specific minimum sales volume?
Since the three types are consecutive stages, an analytics question will also contain elements of the preceding types – so a prescriptive analytics question also involves a descriptive element. These should then be considered as separate problems, since they require different algorithms.
The last step in framing is to narrow down the data domains needed to solve the problem. The data domains may need to be changed later in the business analytics process, for example if a sufficient result cannot be achieved with the help of the considered domains.
Allocation – Using resources
The necessary resources are then available to solve the analytics problem. The relevant resources for business analytics are data, IT and human resources.
Data forms the basis for analytics algorithms. It must be of a certain amount and quality. The quantity and quality requirements are derived from the applicable algorithms, which may not yet be fully clarified. Therefore, the sub-processes of allocation and analytics are usually iterative. The amount of data is relatively easy to define, but the quality requirements are more complex. Quality dimensions can be, for example, the completeness, correctness and consistency of the data.
Generally, the amount of data is so large and the analytics algorithms are so computation-intensive that high-performance IT is needed. This can be divided into six components: hardware, operating system, application system, data management and storage system, network and telecommunication system and internet platform. These components can be provided by the company’s own IT department or by external service providers. A key trend here is cloud computing, either as a private cloud, when the cloud performance is only available to a pre-defined group of users, as a public cloud, when the cloud service is available to all who pay for it, or as a hybrid cloud, when traditional IT, private cloud and public cloud are combined. The author expects the hybrid cloud to become the norm, especially when provisioning IT for business analytics.
In addition to technology, human resources plays an important role in providing expertise. And one of the most common mistakes when dealing with analytics is underestimating this resource. To avoid making this mistake, an overview of the necessary roles and skill profiles is necessary. These can be roughly differentiated:
- Experts on the topic who are directly involved. Take the example of a product manager who needs to introduce predictive maintenance to one of his products. The expert’s job is to define the problem in the context of framing.
- Experts in analytics, often also called data scientists, who specialize in dealing with algorithms. They are entrusted with all data-related activities, from preparation to analysis to the visualization of the results.
- IT experts, who are responsible for preparing and supervising the necessary IT resources for the analytics process. Regardless of whether this IT is operated in part or completely by external parties, the internal IT staff has to coordinate these partners.
Three tasks take priority with regard to human resources: firstly, determining the qualitative requirements for the defined roles. Secondly, determining the quantitative requirement based on the required speed of the analytics process. Thirdly, determining the organizational location, which is mainly a question of centralization in the company.
Analytics – Gaining insights
The third step in the analytics process is to gain data-based evidence and solve a predefined analytics problem. This process is divided into data preparation, data analysis and evaluation.
Data preparation is the most underrated part of the analytics process. Typical challenges include missing data that need to be supplemented, as well as the aggregation and harmonization of different data sources into one database.
The data analysis is usually automated using algorithms. Different types of algorithms can be distinguished. For descriptive analytics, text mining algorithms or social network analysis can be used. For predictive analytics, regression, classification and time series analysis are used, while prescriptive analytics involves optimization and simulation algorithms. The final step of this sub-process is evaluating the results. This is when the author discusses various options, including the examination of a decision tree’s forecasting quality using ROC curves (ROC: receiver operating characteristics).
Preparation – Solving the problem
After all, the insights gained from the data must be prepared in such a way that users can use them optimally. This preparation covers three topics: Visualization, mechanism clarification and derivation of application limits.
The visualization of evidence is an important prerequisite for their application, as an inappropriate form of visualization leads to users perceiving and interpreting the insights incorrectly. Typically, pictorial representations are used with various graphical elements, such as word clouds, Sankey diagrams, waterfall charts, sunburst charts or decision trees. For example, Sankey diagrams can be used to represent flows of material or energy, while sunburst diagrams can be used to represent hierarchical structures.
To optimally use the evidence gained, the mechanisms underlying the evidence must be identified. For example, an outlier analysis shows data instances that are different from others in the same matrix. But only by identifying the mechanism that is the cause of the outliers can the analysis results be adequately safeguarded.
Finally, gained insights cannot be generalized. The data and algorithms used determine their application limits. For example, if an evidence is based on the assumption of a particular customer behavior, this represents a limit – if customer behavior changes, the evidence is no longer valid.
Theory and practice
To clarify how the individual measures of the process are applied, three case studies have been added. One of them, the made up ‘Outfitters GmbH,’ accompanies the book throughout. Two real, but anonymous examples make up the final chapter of the book: the first deals with the analysis of patent data, the second with a predictive maintenance solution.
“Business Analytics: Using Advanced Algorithms Effectively in Corporate Management,” Vahlen, 2017, €49.80
Dr. Mischa Seiter is Professor of Value Creation and Network Management at the University of Ulm’s Institute of Technology and Process Management, Scientific Director of the International Performance Research Institute, and, together with Péter Horváth and Ronald Gleich, author of the reference book “Controlling.”
Author: Editorial team Future. Customer.
Image: Kzenon – AdobeStock