The future is now: cognitive computing throughout the enterprise today
Transformation and integration
Oftentimes, transforming data is necessary to fully integrate data for specific applications or analytics uses cases, which Knudsen notes is distinct from simply blending data for a particular dataset. Transformation is often required in instances in which a particular application requires a data model at variance with that of a particular dataset. According to Nanduri, traditional extract, transform and load (ETL) or ELT paradigms are “being destroyed and completely disrupted. We’re no longer in the world of schema and data models only. We live with data that is both SQL and NoSQL. When we talk about sensors, sensor data is probably coming in JSON. Twitter feeds are probably coming in JSON.”
Automating transformation for a specific application or analytics engine with cognitive computing, then, is also dependent on the data themselves and their underlying semantics. Semantic text analytics can not only denote what the data means by converting data to numeric tokens, but also descry the “similarities of other text and then, by using statistical concepts, you join the data,” Nanduri explains. Those statistical methods rely on computational algorithms that resolve the best way to integrate the data by determining their points of correlation on an internal graph designed for that purpose.
The notion of semantic understanding begat by cognitive computing becomes even more preeminent when implementing data quality prior to transformation. Semantic text analytics effectively “type” data as they are parsing through them to identify and understand their underlying semantics, which greatly informs the meaning of the data. By leveraging those capabilities in conjunction with semantic models known as ontologies, it is possible for the underlying cognitive mechanisms to discern what the data is in accordance to an enterprise knowledge graph that validates names, terms and other points of identification. Moreover, the incorporation of additional taxonomies or classifications of data in a hierarchical format is instrumental to matching records and effecting data quality. Those taxonomies help users clarify how they see their data by classifying them in terms that are relevant to the business so users “can look at the data through a lens that makes sense for what they understand in relation to their business,” Knudsen says.
The synthesis of those semantic technologies with NLP provides further utility for data quality. Although the concept of data quality is inherently related to effective data governance, its importance increases when it becomes associated with relevant business usage of customer or product data. “The really mature companies that we work with are thinking about this from the perspective of a customer journey,” Knudsen acknowledges. “Obviously, they want their clean customer lists, but also as a customer walks through the onboarding process, as they walk through the customer service process and as they buy more product and their experience becomes expanded, they want to see what the customer journey looks like.” NLP can assist in detailing that information by ensuring quality data bereft of ambiguity, duplications, incompleteness and recentness. Nanduri mentioned that Metaphone, a NLP algorithm, can utilize clustering techniques to determine a single entity (such as a product or customer) from varying subsidiaries, alternate spellings, abbreviations and other facets of natural language, so that data quality is effected “in a fraction of a second.”
Deep learning technologies also present a viable alternative for effecting data quality. Because of its proclivity for learning based on experience and determining pattern recognition, deep learning can be used to parse through data in disparate locations to determine what Porter terms a “golden record” of specific information for business use to provide the basis for disambiguation. That approach is beneficial when parsing through multiple databases, IT systems and file types at the sort of scale that would seemingly defy human involvement—or make it so time-consuming that the benefits of using cognitive options are readily apparent. Porter, who implemented such a solution for a well-known customer in the financial services industry, recalls, “We were processing over a hundred million of these items a day on that system. That was a massive system, and we can actually process it in 17 minutes.”
Improved user experience
Implicit to the utility of cognitive computing in all of the fundamental data-centered processes of data quality, data modeling, transformation and integration is an improved user experience. Much of cognitive computing’s virtue is that in multiple instances, it allows business users themselves to traverse those steps prior to leveraging an array of applications or analytics mechanisms. In virtually all of those examples, cognitive computing produces advantages at scale with an expedience that would otherwise not be possible. “We’re applying machine learning techniques to areas where we can provide a better customer experience and more scalability,” Knudsen says. “We think about cognitive as being a great tool in our toolkit to parse this data, to interpret and to include the higher-level machine learning recommendations and automation that we’re doing.”
Such functionality may not immediately revolutionize society, economics and healthcare the way some of the more extravagant applications of AI are projected to, but it certainly delivers the consistent business value organizations require to successfully monetize data. Moreover, cognitive computing is able to implement those advantages today, which marks an irrefutable return on investment that is difficult to match. “The future is now for getting information from data,” Nanduri says. “Cognitive computing has changed the way we do business. We’ve gotten exercises that used to take 28 days down to a few minutes. As a result, the business consumer is now much smarter.”