How to get AI analytics right
Enterprises of all sizes and across virtually all markets are scrambling to augment their analytics capabilities with artificial intelligence (AI) in the hopes of gaining a competitive advantage in a challenging post-pandemic economy.
Plenty of anecdotal evidence points to AI’s ability to improve analytics, but there seems to be less conversation around how it should be implemented in production environments, let alone how organizations should view it strategically over the long term.Start with a plan
CEO Val Swisher, AI can be applied to analytics in three ways: as a descriptive tool, a predictive tool, and a prescriptive tool. Descriptive AI is used to describe something that has happened in the past, usually by grouping data into clusters to detect patterns and outliers. This allows enterprises to answer the question, “What happened?” Predictive AI takes descriptive results and attempts to apply them to the future, again using massive data mining and storing. This answers the question, “What could happen?” Prescriptive AI then takes all this data and resulting analytics to help guide the process to a desired outcome, answering the question “What should happen?”
Depending on your objectives, you’ll need to pepper your analytics with varying levels of these three flavors of AI. But how is this done and how can they be scaled to production levels quickly and efficiently without losing control?
In a recent article on eWeek, SparkBeyond U.S. data science head Ryan Grosso offered up a number of tips to help “bridge the gap between analytic aspirations and ability.” Heading the list is the need to develop in-house analytics talent (as in, human talent) capable of managing the data science tasks AI requires. As well, you’ll need to create hybrid teams with expertise in various domains to replace the often siloed hierarchies that take root in complex organizations. The key here is to train data scientists and business executives to speak a common language. Only then should you select and deploy the proper AI-driven analytics platform, preferably one that can be tailored to your needs rather than requiring changes to your processes or business model.
Reading is fundamental
Even for AI, however, the more difficult it is to gather and parse all this data, the more costly and error-prone the analytics platform will be. This is why one of the key elements in any AI strategy is to get your data house in order, say Manveer Sahota and Chris D’Agostino of DataBricks. One way to do this is to combine legacy data warehouses and lakes under a unified management system that leverages the scale of the former and the flexibility of the latter. This enables the kind of fine-grained control and governance to maximize data availability to intelligent analytics tools without jeopardizing privacy and security.
Post a Comment