This site is not optimized for Internet Explorer 9 and lower. Please choose another browser or upgrade your existing browser in order get the best experience of this website.

Collaborative A.I. Success: Capitalizing On Insights Hiding In Plain Sight

July 02, 2023

Ben Nussbaum

The plane flying above graphs and human, who using graphs in radar

A.I. systems need great data to produce great insights. To do that requires looking for insights beyond the confines of internal databases, as one major branch of the U.S. military found out recently. The resulting work showed that over three-quarters of the delays for maintaining sensitive equipment – delays that could cost billions if not performed in a timely fashion – were preventable. Senior officers have since moved to optimize maintenance operations.

The work is an example of what Accenture researchers Paul Daugherty and Jim Wilson termed “collaborative intelligence” in a lengthy article published in 2018 in Harvard Business Review. Their findings were later collected in a book titled “Human + Machine: Reimagining Work in the Age of AI.” 

In their work, Daugherty and Wilson defined roles humans and machines would occupy in mutual achievement. Among them: amplifying, the idea that A.I. can lead us to better decisions, faster, by providing the right data to do so at precisely the right time. This short blog post is part of an ongoing series in which we highlight observed cases of machines amplifying humans, a process we believe is best captured by the term collaborative A.I. success.

Read on for the inside story of how military leaders discovered and subsequently eliminated roughly 20,000 hours of aircraft maintenance delay at just one base, simply by looking at the data differently.


The surprising structural limits of data science

Aircraft maintenance is a highly structured job. It has to be: just one jet engine can have 25,000 to 45,000 distinct parts. Military aircraft carrying sensitive equipment for important missions carry multiples of that. Caring for each system is a difficult enough task that  mechanical engineers are assigned different levels according to their system expertise.

For example, an engineer rated to maintain, repair, and replace fuel cells might not be rated to care for hydraulics or avionics. Putting the right personnel in the right job at the right moment with the right tools and parts is crucial for mission readiness.

But it’s also not that easy. Even modern equipment can still fail unexpectedly. If key personnel and parts are unavailable in those moments, aircraft that should be in the air would instead end up grounded.

Recognizing this, one base commander asked a team of data scientists to study years of raw maintenance data and look for patterns. Were there obvious signs of pending failure in squadron aircraft? Were there better ways to optimize the 35 technicians tracking their maintenance work in Excel? Surprisingly, the initial answer was no.

The data science team looked through thousands of columns and rows filled with orders, scheduled jobs, time and dollars spent … every structured detail of the base’s maintenance history. Still nothing emerged.

The numbers offered no key that might unlock a better way to work. Only when a new team analyzed the text fields where technicians described the work they’d done would there be a breakthrough.


Hidden patterns waiting to be found

Text fields allowed technicians to explain their choices, and that context proved crucial for challenging assumptions that had informed maintenance scheduling to that point.

Getting that insight would be a challenge. The new team used GraphGrid to build the sort of explainer system that Daugherty and Wilson describe in their work. At its root: a knowledge graph that would capture information filtered via Natural Language Processing (NLP).

Every Excel note was scoured for outliers, such as parts being used in repairs for which they hadn’t been formally approved. The team then added external data to the mix to study the potential effects of geography and weather. Whereas data scientists stopped at studying what technicians had done, GraphGrid helped senior officers understand why technicians were making different choices in their maintenance work.

Ultimately, graph + A.I. in the GraphGrid-powered explainer became skilled enough to predict maintenance cycles with at least 70% accuracy. That, in turn, eliminated roughly 20,000 hours of preventable delays. Mission readiness has improved dramatically as a result and leaders are now considering whether to apply the system at other bases.

Whether they choose to adopt it or not, the lessons of collaborative A.I. are sure to stick and the future of America’s military will be more data-driven and automated than it is today.


Get started now for free

Schedule a free solutioning session with a GraphGrid professional. Tell us about your highest impact data-intensive challenge or opportunity so we can align our platform and professionals to accelerate your graph-enhanced solution journey.

In the meantime, feel free to download GraphGrid at our Freemium license level and check out our step-by-step tutorials, video tutorials, or read more about GraphGrid in our use cases, blogs, and documentation.