BI, contrary to what common experience might suggest, includes the word intelligence. Decisions emerging from BI implementations often smell more of organizational politics than intelligent thought. BI tools focus more on visualization and calculation than logical discourse. In fact, the level of intelligence discernable in today’s BI could be compared to that found in, well, choose your own least-favorite business, political or religious fanatic.

These observations were among the reasons I titled my last book “Business unIntelligence”. Back in 2012, when writing the book, I pointed to IBM Watson, then knee-high to Alan Turing, which had the previous year triumphed at Jeopardy!, suggesting that: “Watson can truly be labeled a business intelligence system, focused on soft information rather than hard data. It plays to the information processing and storage strengths of computers to provide current and relevant information to supplement human skills in decision making.” History buffs can find my earlier and even more positive prognosis in this 2011 blog.

I have been disappointed. In the intervening years, artificial intelligence has become both the most intriguing and most hyped topic in every corner of the technology industry. I must admit to throwing fuel on the fire myself. Yet, as I’ve written elsewhere recently, AI has made only limited inroads into BI tools so far.

My expectation, as the above quote and blog show, was that AI would enable a new and urgently needed focus on soft information (most immediately text, but also images, video, and audio) in BI. The breakthrough in Watson’s DeepQA architecture was its ability to extract and structure meaning from textbooks, encyclopedias, and other documents that use natural language containing implicit meaning. Such meaning is highly contextual, and often imprecise in nature. In a business context, this points immediately to the often-challenging area of business definitions and requirements where ambiguity and assumed background knowledge are regularly blamed for failures in BI projects.

In other words, I saw the focus on metadata, or as I prefer to call it: context-setting information (CSI). In my January article here at eckerson.com, I touched on how automated CSI collection is beginning to appear in some metadata catalog products. There’s much more to be done here, but, in the meantime, my eyes have been opened to another approach to using AI in BI, one that eschews CSI altogether.

A new startup by the name of Outlier is using analytical and AI approaches to offer insights into the complex and voluminous data available from (mainly) click and social media sources to today’s businesses. Traditional dashboards can show such data in enormous detail in streaming perpetual motion. But what is really relevant? Are your Key Performance Indicators indicating items of value or changes in them? Can you eyeball trends and exceptions?

At a deeper level, what are the questions you could or should ask about the data? The traditional rejoinder would suggest that you must understand your business, model the data, and use those to define your KPIs, questions and analyses.

Outlier takes a different approach. It treats all data as time series without any knowledge of the meaning of the data or its relationships to other data and uses statistical techniques to identify trends and exceptions. It then adds basic AI reasoning to suggest what might be important and pose questions to the business person that they might investigate further. In addition, using AI learning techniques, it uses information about what the user actually does investigate to improve its future suggestions.

The counter-intuitive leap here is that there is no modeling up front and no formal context-setting information gathered about data meaning or relationships. The patterns that emerge in the time-series data allow correlations to be observed and the patterns of user responses allow inferences about the possible significance—and thus meaning—of the data and how it changes over time.

This is a brand new approach, but one that has been in beta since mid 2016. My observations are preliminary, but I will be fascinated to see how and how fast this evolves. Is it really possible that a chunk of analytics and BI may proceed without reliance on modeling and metadata?