Nature of the Opportunity/Problem

"Making the World a Better Place"¹¹

Understanding the anatomy of a decision and its opportunity costs

ML makes predictions and it scales. But the value it provides is dubious outside the context of decisions.

The seemingly simple concept of supply and demand has driven a rapid shift in decision making dynamics. In complex dynamic systems, small changes to inputs (such as price) cause dramatic changes in output. Decision making is no different. The proliferation of output from “prediction machines” dramatically increased the value of decision making based on these predictions as an economic complement.^ix

Despite the massive availability of data and rapid advancements in the efficacy of machine learning, the era of big data has not represented a panacea for complex decision making. If anything, these developments make the environment even more complex.

¹¹Mo Lo So Bro

Figure 3 From raw data to impactful decisions. Source: gapingvoid

Complexity creates an opportunity, but only for organizations that have the savvy and nuance to navigate it. History shows that innovative tech doesn’t always lead to innovative results if not leveraged properly. Computers didn’t increase productivity as expected. The service sector invested an additional 116% per worker on technology, but output grew less than 2.2%. The Federal Reserve estimated technology drove only 0.2% of real business growth.^x

Peter G. Sassone studied the effect of computers on productivity at five major corporations between 1985 and 1991. He found that computers shifted admin work to knowledge workers. When companies integrated computers into their workflow it disrupted knowledge work because it lacked framework.^xi

Drucker made it clear: Knowledge work requires autonomy, but workflow still needs structure.

It is essential to build a framework for AI to prevent similar failures, but only 20% of companies use a formal decision-making process.^xii DI should be used to free domain experts to work in their specialty and develop a formal process to interface their “decision flow.”^xiii

National Grid, a multinational gas and utility company, reduced its damage rate by 65% from 2006 to 2014 in their downstate New York operation — however progress stagnated. In 2016 they partnered with AI services company Urbint.

Urbint creates DI models for utility companies and other infrastructure operators that minimize risk and maximize reliability of their clients’ assets. Head of Product Benjamin Berry said, “What’s exciting about the combination of machine learning and decision intelligence is that often the treatment and/or ordering benefit curves are nonlinear. For example, treating 20% of the assets might produce 80% of the benefit. The question is, which assets fall within that 20%?”^xiv

Using DI, Urbint developed models that predicted National Grid’s highest risk excavations. National Grid Gas Damage Prevention Program Manager Robert Terjesen said, “The higher the risk, the greater the potential for damage on that particular site. This allows us to be more targeted with our limited resources. Urbint helps us know where to send our field personnel every morning.” The utility optimized use of their small team, and from 2016 to 2020 they reduced damages per thousand another 37%.¹²

Figure 4 National Grid: Damages per Thousand

¹²Urbint National Grid Case Study