Part 2 in a series of four on the Intelligent Enterprise Machine Learning Framework

Since CRISP-DM was first conceived in 1996, there have been many changes, especially with big data, ML uses and IT advancements.  Large, rapidly changing data sets, streamed data, real time output and self-learning models were not part of the Data Mining landscape twenty years ago. Therefore, the requirements to deliver many modern ML projects do not fit well with the old framework.

CRISP-DM does not adequately cover the operationalization of the analysis in the business environment, where model performance must be continuously monitored and controlled to automatically maintain peak performance.  Therefore CRISP-DM does not cover the whole project lifecycle necessary for modern ML projects.

CRISP-DM can also be difficult for citizen data scientists and business users to understand because it is not business orientated.

Three Reasons for a New, Modern Machine Learning Project Framework

1)Model deployment/operationalization/control phase is crucial:

  • SAP Data Manager and SAP Predictive Factory facilitate the operationalization of SAP ML projects that use SAP Predictive Analytics. This powerful technology is being replicated in Smart Predict, in SAP Analytics Cloud.
  • Other SAP ML projects can operationalize the models in the database, e.g. using PAI with S/4 or other SAP applications.
  • CRISP-DM does not cover the operationalization phase of a real-world project in sufficient detail.
  • Therefore, a framework is required that covers the operationalization of the models, including the monitoring of performance and maintaining/controlling the models over time.

2)Citizen data scientists and business users are often confused by the phases and tasks used in CRISP-DM, especially as they can be difficult to integrate into existing organizational performance strategies.

  • Therefore, a framework should be easily understood by citizen data scientists and business users, and easily integrated into an organization’s existing enterprise performance management initiatives.

3)At SAP, we strive to help every customer become a smart, best-run business and make the world run better.  Therefore, we need a modern ML framework that will support us to deliver the intelligent enterprise and use our customers’ data assets to achieve their desired outcomes faster, and with less risk.

Characteristics of an SAP-Specific Machine Learning Project Framework

And so, instead of trying to fit a modern SAP ML project into a framework designed for older, traditional data mining processes (such as CRISP-DM), a specific SAP ML project framework is required.  This new project framework should be:

  • Be flexible, so that it is applicable for SAP Predictive Analytics, SAP Analytics Cloud, R, Python, TensorFlow, and PAL, so that anybody who currently uses CRISP-DM can easily use this new framework—from citizen data scientists to professional data scientists.
  • Support the operationalization of the models into the production environment, create a “control” plan to continuously monitor performance, and take the appropriate action when the model performance deteriorates so that it can be maintained continuously.
  • Be easily understood by citizen data scientists and business users, and capable of being integrated seamlessly into existing business performance initiatives.

Enterprise performance management (EPM) is an area of business intelligence which monitors and manages an organization’s performance according to KPIs.  EPM provides a framework for organizing, automating, and analyzing business methodologies, metrics, processes, and systems to drive the overall performance of the organization and it is at the heart of the SAP Intelligent Enterprise ML framework I propose in my next blog.

Learn More

For an in-depth look into the intelligent possibilities for your business, review the August 2018 Forrester Consulting study, Powering The Intelligent Enterprise With AI, Machine Learning, And Predictive Analytics, commissioned by SAP.