Integrating Data Labs And Factories To Operationalize Advanced Analytics

Carsten Bange

Although explorative business intelligence requires different processes and approaches compared to traditional BI, it still has to be integrated with existing IT and BI structures to ensure analytics solutions are operationalized. Data labs design new analytics solutions and factories operationalize these solutions, and while they differ in terms of goals and methods, they can also share resources and approaches to foster integration.

Challenges for advanced and predictive analytics projects

According to the BARC Survey “Advanced & Predictive Analytics,” the most common problems that obstruct advanced analytics projects are:

  • a lack of resources in business and IT;
  • difficulty quantifying the business value of advanced analytics; and
  • poor understanding of a data-driven business culture.
factors inhibit your Advanced & Predictive Analytics

Figure 1: Which factors inhibit your Advanced & Predictive Analytics projects? (n=89) Source: BARC Survey “Advanced & Predictive Analytics” (n=200)

All of these factors create challenges when applying advanced analytics and putting analytics solutions into practice by turning them into reliable products and services. To understand why this is a challenge, we will look at how advanced analytics projects are conducted and how data labs and factories operate.

Conducting analytics projects

The “analytical cycle” (Figure 2) is a standard procedure for managing advanced analytics projects. It can be broken down into the following steps:

  • A problem definition provides clearly documented goals and serves as reference for the entire project.
  • Data understanding, data selection, and data preparation create a solid database for the task at hand.
  • Modeling & model validation aim to identify a suitable modeling approach. This may require various iterations with regard to data preparation.
  • Results evaluation is important for quality management. All prior steps are reviewed and checked to determine whether the project meets the business objectives, then a decision about utilization of the results is made. The choices are no utilization (i.e., cancellation or start over), onetime utilization, or operationalization.
  • Operationalization means deploying the model and integrating it into relevant IT systems to improve decisions, products, or services.
  • In the continuous model evaluation, the impact of advanced analytics solutions is measured. Model adaptations are carried out in response to changing conditions. Since this also implies checking the business assumptions and data sources, the cycle starts from the beginning.

Data labs and factories

Traditional BI tasks have been standardized and streamlined for many years in many organizations. A defined output of reports, dashboards, or data for analysis is provided to users in a defined quality and a defined output format for the least possible cost. This approach resembles a factory. Factories focus on high quality and often on high volume throughput, which is achieved by standardization, automation, and stable processes. Traditional BI systems correspond to systems of record that are best run in a factory approach that, organizationally, is often part of the IT department or organized in a cross-departmental BI competency center (BICCs).

The new demand for exploring data with advanced and predictive analytics does not usually suit this approach. Therefore, data labs are created that, like a laboratory in a research and development organization, provide for agile experimentation and discovery. Trial and error, as a typical approach to finding innovative solutions, is key. Explorative BI with advanced and predictive analytics in data labs or analytics labs corresponds to systems of innovation. The labs use agile methods for developing prototypes. Analytical solutions are often quickly discarded when they are used for discovery purposes, do not show promising results, or only provide a benchmark for further solutions. This often makes the value-add or return-on-investment hard to quantify, especially if it has to be defined upfront. Funding for these initiatives comes from innovation budgets.

While often operating in isolation, the factory and the lab approaches must work together when it comes to integrating analytics solutions into operational systems in a stable and sustainable way. The analytical cycle shows which tasks are assigned to which approach (see Figure 2).

Labs focus on the stages of problem definition, data understanding and preparation, modeling, and results evaluation, which may feed back to problem definition. Usually this cycle is repeated until a viable solution is found.

But the goal of advanced analytics initiatives is to operationalize analytical discoveries. Model deployment and integration in productive IT environments are tasks for factories. Here, the factory approach for implementing and running a solution with a focus on data governance, reliability, maintenance, scalability, and reusability is important.

The handover from evaluating the results of a prototype in a lab to the factory for operationalization is the key aspect to consider when setting up advanced analytics initiatives. Whether this is working or not will make or break any analytics strategy.

different tasks for labs and factories
Figure 2: The analytical cycle, with different tasks for labs and factories

Common needs and resources

Despite their differing focus, labs and factories have commonalities. Labs require data engineers, often found in BICCs or in IT departments. They are important in the data understanding, data selection, and data preparation phases, which are the most time-consuming steps in analytics projects. On the other hand, do IT departments need data scientists for processes such as continuously checking model quality, model retraining, model adaptations, and support for users? Data scientists can also be useful when checking data quality. “Data artists” that are familiar with the visual representation of results are part of many BI teams but are also necessary for data labs that need to communicate complicated results to lines of business. Sometimes BI managers become BI and analytics managers, overlooking both processes and ensuring that labs and factories are integrated. Technology-wise, labs and factories are integrated when models or model results generated by explorative BI are displayed in BI systems or used in operational systems when processes are automated.

Conclusion

Labs and factories focus on different tasks in the analytical cycle, but both share a common goal – to turn insights from data into products and services. To do this, good communication between labs and factories is crucial, and this can be enhanced by sharing human resources between labs and factories.

Learn more: read other blogs in the “Enabling the Data-Driven Enterprise” series or review the BARC Survey “Advanced & Predictive Analytics.”

This blog was co-authored by Dr. Sebastian Derwisch, data scientist at the Business Application Research Center (BARC). Sebastian holds a PhD in economics and has extensive experience advising companies in the areas of use-case identification for data analytics, tool selection for advanced analytics, and the organization of data science teams.


Carsten Bange

About Carsten Bange

Dr. Carsten Bange is founder and managing director of the Business Application Research Center (BARC), an IT market analysis and consulting group he founded in 1999 and later merged with Le CXP and PAC to form CXP Group, the largest European IT analyst group. Carsten holds a PhD in management information systems, is a frequent speaker at IT conferences and seminars and has served as an analyst and management consultant on business intelligence, data management and digitalization strategy, organization, architecture and technology selection for over 20 years. He can be reached at cbange@barc-research.com.