Operationalizing advanced analytics – Experiences from current BARC user surveys and projects

Many companies are reaching their limits with their respective analytics strategies: Although, they have BI, analytics and data mining platforms at their disposal and use these to generate useful results and findings to make better-informed decisions. However, these analytics systems are frequently separate to the other existing systems, which leads to problems: The results, tables, evaluations, and analyses are often only used for one-off initiatives, or repetitive manual work is created for multiple use. To run, for example, a campaign for promoting a product, analysis and data mining tools are perfect for generating the customer information required (how probable is it that the customer will buy, which customer should be reached via email, postal mailing, or by phone). Unfortunately, this is followed by a discontinuity in the media in the further process and in relation to the operating systems: The CSV tables generated by the tool, must be put in a form that supports the operating systems (e.g., campaign systems, call center management systems, CRM system, etc.), in order to achieve the actual added value of the analysis – cost savings and increased revenue. The key is in the operationalization of valuable advanced analytics artifacts: Their transfer from (advanced analytics) labs to stable IT solutions with regular maintenance cycles means advanced analytics can bring genuine benefit to a company by directly supporting the operative business processes.

BARC user surveys indicate that companies are having difficulties with the operationalizing advanced analytics, as more than 50% of the surveyed companies that specified they currently use advanced analytics are prototyping use cases or evaluating prototypes at the moment. Only 22% have completed this phase and are deploying advanced analytics on a regular basis for aiding the decision making process (15%) or have automated important business processes (7%).

Figure 1: "In which phase of implementation are you currently in?" LOB led and IT & data scientist led companies where surveyed (n = 74; BARC user survey, "Advanced & Predictive Analytics – Data Science the line of business", 2017)

Figure1: “In which phase of implementation are you currently in?” LOB led and IT & data scientist led companies where surveyed (n = 74; BARC user survey, “Advanced & Predictive Analytics – Data Science the line of business”, 2017)

The transfer of prototypes from (advanced analytics) labs to stable IT solutions with regular maintenance cycles conceals specialist, technical, and organizational challenges, and said challenges must be considered by the data labs, IT, and departments involve in this process.

Labs (systems of innovation) and factories (systems of record) are frequently separate units

Data science labs can also be perceived as system of innovation. They represent a space for spirited experimentation and discovery. Trial and error are therefore part of the working process and solutions are assessed according to the added value they potentially create as well as how innovative they are. Solutions are frequently discarded or used as a benchmark for additional iterations stages. In contrast, the classic BI environment can be perceived as a system of record. Its focus is on stable and highly qualitative solutions that serve the resolving of recurring analytical questions. Efficiency is achieved by standardization, automation, and secure processes. Within the context of advanced analytics, there is an additional task for these systems of records: If highly relevant models or use cases are identified in the lab, these must be integrated in the corresponding systems. These could be scheduling BI systems or operative ERP or CRM systems.

Specialist, technical, and organizational challenges are inherent to the transfer of prototypes to stable environments with regular maintenance cycles and must be understood and implemented by the data scientists, IT, and lines of business involved in the process.

Operationalization steps appear as separate phases in the analytical cycle

The analytical cycle is usually used for structuring advanced analytics projects. It consists of the following phases:

The remit definition documents the objectives of the analysis and highlights the remit in terms of strategic and operative dimensions. This forms the reference for the entire project. Data understanding, selection, and preparation generate a solid data basis for the implementation of the remit. The objective of the modeling and model evaluation is the identification of a suitable mathematical model. This may require several iterations of data processing. The outcome evaluation aids quality management and is the basis for the decision of how far should be proceeded with the approach. All steps are examined here again, and a decision is taken whether the results should be distributed as a one-off investigation, the analysis discontinued, or the solution should be operationalized. Model deployment encompasses the steps for integrating the model into the existing IT structure, the model scoring, the possible creation of user interfaces, or the modification of existing applications and organization of the model management and model evaluation. Finally in continuous model evaluation, the results of the analysis are regularly tested and the solutions revised where necessary. This includes the periodic retraining of algorithms based on the latest data.

The final two phases, model deployment and model evaluation, constitute operationalizing advanced analytics. If an analysis is conducted more than once, and operations should be scheduled according to the latest data, then operationalization is necessary. This task is realized together with the developers of the solution, data scientists, and the operators of the IT environment, IT, or the BICC.

Figure 2: Prototyping and operationalization in the analytical cycle

Figure 2: Prototyping and operationalization in the analytical cycle

Advanced analytics managers report about different approaches with the operationalization of advanced analytics

In the BARC fireside chats for advanced analytics decision makers (https://barc.de/bi-leaders), the subject of the operationalization of advanced analytics results as well as the discontinuity in the media between the lab environment and operative systems are becoming increasingly important. Companies are coping here using a variety of options: Some encourage their data science teams to pass on their data mining prototypes to IT departments with thorough documentation and be available as a point of contact when the models are integrated into ERP, CRM, and other systems. This means the user can be supplied directly with important information during his daily business process. In other companies, data scientists and IT already work closely together in the prototyping phases from an early stage; in addition, data scientists are bearing increasing responsibility for the maintenance of operationalized solutions.

New software support for the operationalization of advanced analytics

It is undisputed that BI, analysis, and advanced analytics technologies aid companies better than ever in their decision-making process, based on a variety of functions and visualization capabilities. However, an increasing number of companies want to not only be effective, i.e., reach the desired result, but increasingly wish the reach the result more efficiently, with minimal effort. One key may lie in a better interlinking of analytics and the operational area. Software vendors have recognized this fact and are now developing interesting concepts, methods, and technologies in order to exploit the potential of analytics on a much wider and immediate basis, i.e. in the everyday business process and operational environments. So, for example, Apteco is offering an analysis and campaign solution consisting of a column-based database, a data integration module, and advanced analytics components for analyzing customer information to optimize and realize marketing campaigns. Once the user has defined the relevant customers and their characteristics in the analysis tool, this information can be transferred directly to the module to plan and implement campaigns. Data mining platforms such as those from Dataiku, IBM, SAS, or Tibco Statistica can provide advanced analytics models as a service by means of API for external applications and systems. The meaningful operationalization of analytics enables companies to reduce costs, and identify opportunities sooner and more quickly.

Find out more about the exciting opportunities and best practices of the operationalization of advanced analytics at the Data Festival on April 17 and 18 April in Munich (https://www.datafestival.de). And make a note about the BARC Congress for Business Intelligence and Data Management on November 20 and 21, 2018 in Würzburg, where you can network with other participants, exhibitors, experts, and like-minded professionals, and swap notes in moderated workshops (http://barc.de/congress).