BOSTON -- CIOs looking for a digital transformation case study will find an ongoing master class at GE. With its Predix platform, which collects and analyzes sensor data from industrial assets such as MRI machines and has been generally available for 18 months, the company is attempting to position itself as the backbone of the industrial internet of things.
But the transformation efforts have been slow to produce results. The company's earnings are lagging behind industrial competitors -- making shareholders uneasy and ultimately leading to the recent departure of CEO Jeff Immelt, digital advocate and proponent of the two-year-old GE Digital.
That was the backdrop for a recent media day event at the company's temporary headquarters in Boston. Three representatives from GE Digital -- Mark Bernardo, vice president of professional services; Mike Varney, senior director of product management; and Jeff Erhardt, vice president of intelligent systems -- provided an informal presentation on GE's Predix platform, the critical role of data and domain expertise for machine learning, and what the future of GE's young business unit might look like.
Predix platform is key
Immelt was replaced last month by John Flannery, a GE veteran who most recently worked with the company's healthcare division. One of Flannery's early tasks as CEO is performing a deep dive into each of GE's businesses. He plans to complete his audit later this year and present recommendations to investors.
What Flannery's investigation will mean for the future of the company is yet to be seen. But the representatives from GE Digital said they've seen no change in strategy to date and that Immelt's vision to create the platform for the industrial IoT will likely continue.
In fact, Bernardo, a GE employee for more than 10 years, described reports that GE Digital will need to step up revenue production in 2018 as "normal GE behavior" and not a deviation from strategy.
CIO Jim Fowler expounds on GE's digital thread and digital twin technologies.
"Our platform, our application investments, our investment in machine learning, our investment in our talent, the reason why domain expertise is important to us is because we need it in order to generate the outcomes our customers need, and to generate the growth and productivity that we need as a business," he said. "We are as dependent on this strategy as any of our customers."
With the mention of machine learning, Bernardo is referring, in part, to GE Digital's 2016 acquisition of Wise.io, a startup out of Berkeley, Calif., that specialized in predicting customer behavior. That may seem like a far cry from industrial assets, but Erhardt, CEO at Wise.io at the time of acquisition, said the key to solving hard problems like predicting customer or machine behavior hinges on a common, underlying data platform that provides a foundation for application development.
"That's what Salesforce.com has done," Erhardt said. GE's Predix platform is built on the same basic model. Erhardt said Wise.io observed from dealings with customers that a data platform is necessary to successfully scale a company based around machine learning, and that it was one of the reasons why being acquired by GE made sense for the startup.
Data is the new oil pipeline
For Wise.io's part, its job is to make GE applications intelligent. Doing so generally requires computational power and machine learning algorithms -- both of which have become commoditized at this point -- as well as the increasingly valuable data and domain expertise, according to Erhardt.
"[Data and domain expertise] are at the forefront of both research and how you apply these intelligent techniques, as well as where you can create value," he said.
He used GE's intelligent pipeline integrity services products, which rely on the same basic imaging technology packaged in the healthcare business's products, as an example. "We stick [them] in an oil pipeline and we use [them] to look for defects and weaknesses indicative of that pipeline potentially blowing up," Erhardt said.
But the technology captures so much data -- Erhardt said roughly a terabyte of images -- that it can take highly trained experts months to sort out. The machine learning technology, which he defines as "the ability for computers to mimic human decision-making around a data-driven work flow," relies on past data and decisions to flag problematic areas at super-human speeds.
"The purpose and the idea behind this is to clean up the noise and allow the people to focus on the highest risk, [the] most uncertain areas," Erhardt said.
The technology doesn't replace human decision-making outright. Erhardt said his team is spending a good chunk of its time striking the right balance between automation, augmentation and deference. In the latter case, the system defers to domain experts, who may have decades of experience working with complex industrial assets. Domain experts also help GE's managed service customers prioritize anomalies surfaced by machine learning technology.
Keeping a human in the loop, in other words, is essential. "What's really important here -- and this is different than the consumer space -- the cost of being wrong can be very, very high," Erhardt said.
It's another reason why machine learning algorithms have to be well-trained, which requires enormous amounts of data. Instead of relying on data generated by a single pipeline integrity product or even a single customer, the Predix platform enables the company to collect and aggregate data across its customer base -- and even across its businesses -- in a single location. This gives the machine learning tech plenty of training data to learn with and potentially gives GE Digital the raw material to create new revenue streams.
"We're looking for commonality across these very powerful business cases that exist within our business. What it then gives us the ability to do is to create these derivative products," Erhardt said. He cited Google's 2015 acquisition of Waze, an application that helps users avoid traffic jams by using geolocation driver data, as an example of how companies are using data generated by one application to help power other applications. Waze remains a stand-alone application, but the data shared by drivers is now used for city planning purposes.
"The way that we approach this is if you get the core product right -- if you can entice your customers to contribute back more data -- you not only make that good but you create opportunities you didn't know about before," Erhardt said. "That's what we're working on."
Machine learning is full of untapped opportunities
Tom Davenport's take on automation vs. augmentation
GE CIO is not the chief digital officer -- and for good reason