AI center of excellence: A new engine for driving business transformation

When Shell first launched its AI Center of Excellence in 2013, it wasn’t even known as AI, but as predictive analytics, and it was tiny.

“It was just me,” says Dan Jeavons, who now leads what’s known as Shell’s Data Science Center of Excellence, which includes 180 full-time data scientists and engineers.

“They expanded it somewhat,” he says.

Originally, the purpose of the CoE was to support business units working on analytics-related projects.

“We had a whole bunch of business-line projects spinning up all over the place,” he says. “And we had some leaders who recognized that we needed to coordinate what we were doing. The original intent was to facilitate what was going on, but to allow the businesses to do the work themselves.”

But AI requires deep technical skills, and a business unit’s generalist data scientist may not have sufficient expertise in niche topics such as deep learning or machine vision or natural language processing to make the best use of it. In addition, allowing business units to do their own thing resulted in people taking different approaches to the same problems.

“There’s a lot of need to create common standards, to create common platforms to access data, to develop assurance processes,” Jeavons says.

As a result, the role of Shell’s CoE expanded to include more people with deep, specialized skills and to set standards and policies.

Today, Shell uses Microsoft Azure for most of its AI projects, with some work on Amazon’s AWS, Jeavons says. The company also has close partnerships with data analytics companies Databricks and C3, with various partners on top of that providing more specialized technology capabilities. That has helped accelerate the development of the company’s AI solutions, Jeavons says.

For organizations looking to achieve AI success at scale, establishing such a centralized approach can be key. According to Deloitte’s latest State of AI in the Enterprise survey, released in July, seasoned adopters of AI are much more likely to take a centralized approach to AI technology and vendor selection.

“Without a center of excellence, you have bespoke investments across eight to ten business lines, everyone pinging IT from different areas, less efficient investment,” says Dave Kuder, principal at Deloitte Consulting, adding that AI CoEs can help companies go from single project prototypes and proofs of concept to deploying AI at scale.

“Disparate activities taking place over the last several years now need to be industrialized, hardened, and operationalized,” he says. “The AI center of excellence plays a big role, and is a good stepping stone to coordinating some of these activities.”

As Shell’s experience shows, the development of in-house expertise and common platforms and standards is the first phase of establishing an AI CoE lifecycle. Here we take a look at how AI CoEs are impacting businesses today.

AI at scale

QTS Data Centers established its QTS Innovation Lab two years ago to help accelerate the data center provider’s AI journey.

“Our business was growing, and we had some inefficiencies in our analog processes,” says Brent Bensten, CTO of products at QTS, which operates 26 data centers globally.

For example, an engineer used to walk around QTS facilities, which include million-square-foot buildings, to personally visit all the equipment, Bensten says. “An engineer would go to our air conditioning panels, for example, take the readings, manually write them on a clipboard, then go back and check ‘all okay.’ You didn’t have the underlying data points, just the ‘all okay.'”

So the lab’s first goal was to digitize this process to increase business efficiency. Today, all those measurements are gathered automatically, in detail, providing the company with a complete digital footprint of its operations. This data can now be used to predict power requirements, optimize maintenance, and reduce the company’s carbon footprint.

“All of that is now a completely digital experience managed and operated by AI,” he says. The lab, which spends about 90 percent of its time on AI projects, has also configured its systems to integrate with third-party platforms, such as ServiceNow and Salesforce. “We centralized governance and that allowed us to roll out the digital experience across the broader QTS portfolio.”

And the benefits haven’t just been internal. The same predictive tools help benefit the company’s customers.

“Using AI, using machine learning, neural concepts, we can predict things that are going to happen, outages, failures, power consumption,” says Bensten, including the ability to predict customers’ power utilization up to 60 days out. As a result, QTS customers can be more fluid with their environment, reduce power costs, and get ahead of problems, he says.

“And they can reduce their sustainability footprint,” he adds. “They can submit themselves for sustainability credits because we provide the data to them.”

For Ernst & Young, launching an AI CoE in 2016 helped accelerate AI adoption at scale, says Jonathan DeGange, Ernst & Young’s associated director of machine learning at its AI CoE.

“We felt there was a bit of a problem with the siloing effect, where you have different groups that are not talking to each other,” he says. In addition, Ernst & Young was looking to tackle massive projects, like stopping money laundering, that required coordination across business units and geographies.

“There’s a force-multiplying effect of bringing together expertise in a particular domain,” says Carl Case, principal in the financial services office at Ernst & Young, who looks to the AI CoE for help with AI projects. “And we’re looking at tackling some very meaty, large problems — global financial crime, complex tax law and regulation, addressing the future of work and the impact of digital transformation.”

Before the formation of a global AI CoE, he says, his team was interacting with a smaller US AI team. But once the global center opened, Case’s team could start thinking big. “We’ve seen drug rings, human trafficking rings, and stopped them,” says DeGrange.

For example, using graph-based networking approaches and anomaly detection, the AI system can identify suspicious patterns of behavior, and not just individual transactions.

“You’re now looking at how the network is behaving as a whole,” he says. “Are new relationships being created in a way that is anomalous? For example, the way a criminal ring opens accounts and transacts looks very different from a person starting a legitimate new business. There’s a distinct difference to the patterns that can be detected with AI.”

Transfer of learning

Once an AI CoE has established in-house expertise and common platforms, its next stage is to share best practices across the enterprise.

“AI doesn’t respect organizational boundaries,” Shell’s Jeavons says. Predictive maintenance, for example, is applicable to almost every business in Shell, and having established an AI CoE, a technology developed in one area can then be deployed in many different places.

“Machine vision is another example,”Jeavons says. “We’ve been developing use cases in retail, but the enabling capabilities behind those use cases are applicable to inspection in our manufacturing sites and to look at issues with corrosion.”

General Electric has also launched an AI CoE to help leverage AI across different business units. Known for its use of Six Sigma and Lean methodologies to improve manufacturing efficiency, GE is now using AI to bring the same kinds of process improvements to other areas of the organization. The company, which already had a center of excellence for AI in its global research center focusing on creating digital twins of the machinery that GE builds, now has a new CoE focused on using AI to drive digital transformation throughout the company, based at GE Digital, says Colin Parris, senior vice president and CTO at GE Digital.

Because business units typically focus on immediate needs, they don’t usually have the kind of deep AI expertise required for major transformative projects, Parris says.

“We do the research on the hypothesis to make sure it works, bring the top talent together, and then take it across multiple parts of the GE business,” he says.

Right now the center is working on what’s causing warranty costs to rise.

“It’s rising because we have certain parts that are being damaged and we’re not replacing them fast enough,” he says. “If we can figure out earlier that the parts had damage on them and replace them when they’re still category three, it could cost us $5,000 and would take a week. But if we wait until it’s category five, it could cost us $500,000 and take a month.”

AI-powered image recognition is being used to help speed up inspections by helping human experts focus on the most problematic areas, such as damage on blades. And AI is being used to schedule repairs to make optimal use of staff while also doing the maintenance earlier, when it is less expensive.

“We’ve now deployed this as an experiment in the field in South America,” he says. “We’ll see the results by the end of the year, and if the warranty costs go down and the experiment makes sense, we’ll deploy it in other areas.”

Company-wide transformation

By establishing an AI CoE and infusing best practices across the company, organizations are better positioned to uncover new insights from AI pilots and projects that can dramatically transform how the company operates.

At Shell, it started with sensors.

“We provide monitoring services, in the form of IoT sensors, and we’re able to monitor what’s going on in real time and give the customer additional services,” says Jeavons.

For example, Shell knows that water is getting into an engine before the customer does. As a result of insights such as these, Shell is becoming more than just a supplier to its customers, but a partner — a shift that is creating new business models powered by the transformation enabled by AI, including new energy businesses, such as the management of digital chargers.

In fact, Shell as a whole is on its way to becoming an AI-powered company, Jeavons says. “We have a program, Shell to AI, an integrated change program, which is designed to embed AI into every part of our business.”

And it’s not just manufacturing, or maintenance, or research and development. “I fundamentally believe that the way that software is going to be developed, powered by cloud computing and AI, is going to transform every part of our business in the coming years. Just like the internet, this technology is going to become pervasive, and we need to get ready for this,” he says.

As part of this process, the CoE helps coordinate a grassroots project of people at the company interested in AI, which has grown from 30 people in 2013 to 4,000 today.

“We are sharing what’s going on at Shell with AI and how it can be applied,” Jeavons says. “We also partnered with Udacity to develop training for citizen data scientists to help them upskill, and are investing in common platforms to enable people to deploy things at scale.”

Shell is not alone in seeing the transformative potential of AI. According to a survey by MIT Sloan and the Boston Consulting Group conducted late last year, 90 percent of companies see AI as representing a business opportunity for their company.

But most companies are still in the early stages, as fewer than 40 percent report any business gains from AI over the past three years, according to the survey. Those that are successful, the report shows, are able to unify their AI initiatives with larger business transformation efforts.

One way that this can happen is that a company has an AI project that results in insights that lead to new product lines, or even a complete rethinking of the company’s business model. That can happen relatively early, says Ken Seier, chief architect for data and AI at Insight, a Tempe-based technology consulting firm.

Having an AI CoE in place can help ensure that learning can be communicated to the rest of the company, and to senior executives, thus increasing the chances of driving real change.

For example, Seier says, he has worked with a large aerospace company that was using AI to improve maintenance. Its business model was selling equipment and maintenance contracts. But with a better understanding of the health of the equipment, they were able to move to a subscription-style service.

“Now they’re taking more ownership of the customers’ success because they have a better view of the customers’ business than the customers do,” he says. “Customer satisfaction goes up, the revenue model gets more predictable and smoother, and their overall costs go down.”

With big changes come big risks, however, and with AI, that means more than just garden-variety business risks.

“If we dramatically change the way we do business, we need to make sure we do it in a way that’s responsible to our workforce, our customers, our stakeholders and general population,” he says.

“AI has the same or greater level of potential disruption as nuclear power,” says Sounil Yu, CISO-in-residence at YL Ventures, and former chief security scientist at Bank of America.

The negative ramifications have to be understood across an organization, and AI centers of excellence have a role to play here as well, he says. When the AI risks are evaluated in silos, it creates the potential of leaving out critical stakeholders.

“And we run the danger of leveraging AI for activities that it is unsuited for,” he says. “This creates potential liabilities.”