A big thank you to my blog readers and Twitter followers for sharing my plea to see SAP InfiniteInsight earlier this month in the social media channels. SAP immediately reached out to me and SAP Director of Product Marketing, Julien Sauvage, provided a wonderful walk-through and inside look at this amazing predictive solution. Here is an overview of what was shown.
Predictive Analytics for the Masses and the Data Scientists
Since the acquisition last fall, SAP has been integrating KXEN functionality into SAP Predictive Analytics. SAP Predictive Analytics basically is an add-on to SAP Lumira (formerly SAP Visual Intelligence). There is a convergence happening with differing levels of features for data scientists and business analysts. The driving vision is to extend the value of data science by making powerful predictive algorithms available in three simple clicks or less – and – with no need to understand the concept of predictive model development. The ultimate mission, deliver predictive analytics for the masses.
As requested, Julien showed me both the SAP InfiniteInsight basic business analyst user interface and the advanced data scientist workbench functionality for each step of the CRISP-DM lifecycle process. For business analysts, there are several point-and-click features to add predictive functions for visual analysis. These options include but are not limited to Find Influencers, Create a Forecast, Create a Regression and Find Association Rules. Influencers are variables that have the most impact on an outcome. For example, a marketing manager may use Find Influencers to better understand where to target advertising spend. Most people know how forecasts and regressions are used. Association is commonly referred to as basket analysis – things that are often purchased, happen or exist together (think Amazon Recommender).
Another SAP InfiniteInsight feature Julien presented was a Contributions by Variable window. I really appreciated that functionality since a business analyst could easily understand it and apply this new knowledge to many other analytic use cases. In the Contributions by Variable view, similar to Find Influencers, the variables that have the most effect on an outcome are shown in the order of influence weight. Often if no other predictive findings are conclusive, the influencer variable information can still be used in improving decision making processes.
Moving on to the features a Data Scientist would utilize, Julien showcased the SAP InfiniteInsightworkbench. The workbench contains functionality that would typically be found in traditional data mining and predictive toolsets. Out-of-the-box workbench components include Favorites, Algorithms, Data Preparation, Data Writers and Models. There is also an option to add more “components” allowing for reuse of shared knowledge within an organization. As you can see in the image below, you begin with a data set and apply various predictive algorithms.
Stepping through each phase of the CRISP-DM predictive model development lifecycle, easy to understand screens are provided for each step of the process. A data scientist or even a power analyst user can choose predictive models, tweak algorithm settings, variables, run training routines and evaluate predictive performance or “lift”.
To evaluate predictive model performance, SAP InfiniteInsight uses industry standard k-fold Cross Validation. Predictive model performance results can be viewed via ROC Curve/Lift Chart, Confusion Matrix or exported scored data sets with appended predictions and prediction probabilities.
SAP InfiniteInsight has made predictive model integration practically painless for use in business processes, dashboards and reports via a step-by-step wizard. Keep in mind that integration of a predictive model into applications is where true enterprise scale and the power of predictive analytics comes to life in improving business decisions. The SAP InfiniteInsight Apply Model wizard features aid in loading new input data sets and provide scored result output data sets, statistics and visualizations.
SAP InfiniteInsight also can generate SQL code or User Defined Functions (UDFs) for a variety of popular databases including but not limited to SAP HANA, SQL Server, Netezza, Vertica, Teradata and Sybase IQ. If you have ever used the SAP Hana Predictive Analytics Library (PAL) or Predictive Queries (DMX) in SQL Server world, this code generator in SAP InfiniteInsight is comparable for integrating predictive models.
Compelling Predictive Automation Advantage
One of the key differentiators of SAP InfiniteInsight is the automation of data preparation and model building that historically has always been the most time consuming aspect of a predictive project. It was this ONE specific feature that attracted me to review this solution since I appreciate its significance having worked on several predictive projects since 2003. According to an Aberdeen study available on the web site , data preparation and model building can account for an average of 50% of the total predictive analytics lifecycle task time. Evaluated customers built and deployed predictive models in approximately 30% less time than their peers using other predictive tools. The bottom line = SAP InfiniteInsight’s automation capabilities combined with ease of use and straightforward application integration are indeed compelling and undeniably provide a competitive advantage.
I love what Julien showed me in SAP InfiniteInsight. It was exactly what I wanted to see and his demo exceeded my expectations. As a disclaimer, I have not played hands-on yet with this solution but could not resist sharing my initial findings. In the meantime, I suspect my only gripes might be possibly around pricing.
Thanks again to my readers for rallying to get me this awesome SAP introduction. Also thanks to SAP Director of Product Marketing, Julien Sauvage, for taking time out of his busy schedule to walk through the solution with me and answer my zillion questions.