Predictive analytics in an inherently difficult task and often takes specialized skills. While not easy, the business results of predictive analytics can be significant. 68% of companies say they use predictive analytics to create competitive advantage while 55% say that they increase revenue. KXEN is a software company that specializes in making predictive analytics easier to use by automating predictive analytic processes and some data preparation tasks. Like other predictive analytics companies, KXEN targets uses cases in risk and fraud prevention, operations and customer service, but given its end-user focus, it is natural that the company seems to be finding a niche on the customer-facing side of business in areas such as sales operations and marketing.
The key to KXEN’s strategy is what might be called a commodity or good enough approach to modeling. That is, end users do not need to know advanced statistics to use InfiniteInsight, the company’s flagship platform. The user feeds data into the KXEN engine, and the system dynamically creates and validate models. One selling point for the company is that the engine is able to ingest hundreds of thousands of variables and automatically sort through the data set to find the right predictor variables. In a traditional analytic process, by contrast, a trained statistician would define the variables to be used in the model and often go through a data reduction process prior to building a predictive model. InfiniteInsight, avoids this step entirely. To test the model, it uses only half of the data to build the model, and the other half of the data to validate it. It avoids over-fitting by cross-validating the original training set with the validation set.
The variables with the highest predictive power according to the KXEN algorithms are the ones that are subsequently used in the actual production system. For instance, consider a next-best-offer prediction that includes opening a new checking account, offering a new credit card or offering a home equity line of credit. Each product offer would be modeled separately and may have different drivers. When a customer reaches the company through the call center or a website, each product will be scored according to the customer-related variables that are most predictive of that offer. If a person has just been married, that may be a better predictor for opening a new checking account than if the person just bought a new house, which may suggest a home equity line of credit. InfiniteInsight integrates with third-party business rules engines that are a necessity for almost any type of real-time operational analytic system.
As demand for analytics becomes more important to organizations, application vendors can choose to build analytics into their applications or strike partnerships with companies such as KXEN to provide the necessary intelligence. Such partnerships represent a big opportunity for BI and analytics vendors since emerging cloud-based companies often focus on applications themselves and not analytics. For example, KXEN partners with salesforce.com to provide predictive applications on the salesforce.com AppExchange. At the same time, KXEN has its own Cloud Prediction platform that offers applications for predictive offers (also called next best offer), lead scoring, retention and case routing. This hedge is the smart play. Our benchmark research into next-generation business intelligence shows that companies are split on how they will deploy their next generation of systems: 38% said it will be part of a specific business intelligence system, 36% said it will be driven through Microsoft Office, and 34% said it will be embedded into the application. With the rise in mobile intelligence and the importance of operational intelligence in today’s organizations, it will be interesting to see how these numbers change in our next generation business analytics research, which we will conduct in 2013.
The latest release of KXEN’s flagship InfiniteInsight, version 6, which became generally available in March, adds capabilities for social intelligence as well as campaign intelligence for marketers. For social intelligence, version 6 provides capabilities to explore social graphs to identify connections among people and find top influencers in a category. It can also put social attributes into a predictive model for purposes such as predicting social media paths to increase the effectiveness of a viral campaign. Another product called Genius enables point-and-click campaign modeling so marketers can run analytics on just about any size of campaign. This is becoming important in the world of digital marketing since smaller, more targeted campaigns are needed to lessen to the noise in the consumer’s digital environment. It used to be that only large direct mail campaigns would get a unique model, and that model had to be built and interpreted by trained statisticians. Once the model was optimized, it would then be translated into database language, the database would be scored and the target prospects selected. This took time and high-priced talent. Today, many models are needed in a much shorter timeframe. Commodity modeling approaches such as Genius help marketers quickly optimize their campaigns without having to involve a statistician. Such time-to-value is a key buying criterion is today’s fast-paced markets and for KXEN’s client base of more than 400 companies of various sizes.
This week the company announced location intelligence as a native feature of its InfiniteInsight platform. Location awareness enables the system to understand the location of a particular person and use this information to help predict the most relevant offers to that person at that time and place. Using the location technology, the company also offers co-location and geographic path analysis techniques by which the location intelligence can look at similar events occurring within a certain area or look at a time sequence of events occurring in multiple places. Such techniques can help, for example, to root out crime or provide real-time route optimization during heavy traffic times. Our benchmark research on location analytics, that we are completing, suggests that location information has been an underappreciated source of intelligence, and while it is beginning to gain some early traction, people’s lack of location analytic skills is still an obstacle.
Predictive analytic models are only as good as the quality of the input and therefore data pre-processing is a key consideration for predictive analytics. Our benchmark research into business technology innovation shows that data preparation and quality are critical challenges and time-consuming activities impacting analysts in 42 percent of organizations. KXEN has basic tools for data preparation such as checking for missing variables, classifying variable type, encoding of continuous variables and outlier detection and handling. Its social graph capabilities can also link people with many identities, though their ability to clean the data set and merge these identities automatically is still unclear. Data preparation is an area where other tools still may be needed since they often include more advanced data preparation capabilities.
Analytics was ranked as the top technology innovation priority by 39% of participants in that research, more than twice as many cited the second and third highest priorities of collaboration and mobile technology. In addition the most critical capability to satisfy business analytics is applying predictive analytics in almost half (49%) of organizations. Analytics is a broad category, and predictive analytics is perhaps the most complicated in terms of systems and organizational integration. KXEN has developed an approach that automates much of this complex world of predictive analytics. Its advantages include providing organizations with a common language framework for understanding predictive analytics.
The primary arguments against KXEN’s approach are that the quality of its models may not be as strong as those done by a trained statistician and that the breadth of use is not as wide as some of its competitors attain. While these arguments have validity in certain circumstances, we note that lack of skills is the primary barrier to dissemination of predictive analytics. In many situations, commodity models that address this skills gap at the front line of the organization are better than current approach of randomness and gut-feel.