SAP InfiniteInsight (formerly known as KXEN prior to acquisition by SAP in 2013) addresses a particular set of predictive analytics problems in several well defined markets (typically, but not exclusively retail, financial services and telecoms). The two very significant features of InfiniteInsight are the speed with which predictive models can be built and the reliability of those models. It is not however a general purpose machine learning or data mining toolbox, and relies on techniques which come under the heading of structural risk minimization (SRM). These provide a much better guarantee that resulting models will work into the future. The downside is that the technology is not applicable to many business problems – but the positioning is very much targeted at end users in marketing departments and doesn’t claim to be a general purpose platform.
The productivity boosts that come from InfiniteInsight are derived from the way it helps with data preparation and model development. This is where it really excels, and as any data analyst will tell you, data preparation particularly is the largest piece of the effort. What many in the predictive analytics community may not be so willing to tell you is that, depending on the problem and the nature of the data, many models are not reliable. InfiniteInsight addresses this problem as well as it can be addressed.
So who should use InfiniteInsight? Well despite the claims that marketers can add it to their arsenal of tools, in reality the user really does need to know what they are doing – despite the ‘it’s so easy’ claims. InfiniteInsight is easier to use than almost all other tools of its kind, but not that easy. This doesn’t detract from the value InfiniteInsight has to offer, it’s just that it still needs some level of skill to use.
As for applications, these are almost entirely centered around marketing, sales and the customer. Classification, regression, segmentation, forecasting and association rule processing are billed as the main techniques to deal with issues such as marketing response rates, customer churn, identification of selling opportunities and so on.
Obviously the creation of predictive models is just a first step – the models also need to be deployed in the context of business applications. To this end InfiniteInsight supports the deployment of scoring equations in the database and integration can also be achieved using a variety of languages (SAS, Java, C and PMML). Numerous databases are supported (Teradata, Oracle, SQL Server, IBM DB2 and Netezza) and essential management tools are provided to monitor data and model deviations.
In summary InfiniteInsight is effectively an application more than a predictive analytics toolset, and provided it is deployed in areas where its algorithms can add value (marketing, sales, customer care, social data) then it will almost certainly add significant value to the business operations.