Artificial intelligence is often heralded as the disruptor that could affect every facet of the enterprise -- from technology infrastructure to business models.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
While the playbook on how to build an enterprise AI practice is still being dreamed up by the Googles and the Amazons of the world, it's clear that CIOs and their IT departments will play -- and are playing -- a vital role as technology educators, business collaborators and AI project enablers.
Equinix Inc. CIO Milind Waglé, for one, is leading the charge by finding artificial intelligence projects that introduce new, never-before-seen efficiencies to the data center company. And Ben Clark, chief architect at Wayfair Inc., is empowering the e-tailer's data science and analytics teams by making data, and the tools to explore and analyze that data, as accessible as possible.
While the two IT executives represent a diversity in artificial intelligence projects that is as sprawling and nuanced as the term itself, they also provide a good illustration of how CIOs are preparing a fertile landscape for artificial intelligence projects to take root and thrive.
AI and efficiency
Equinix's Waglé, who leads the Redwood City, Calif., company's global IT team, is a level-headed optimist when it comes to artificial intelligence. He sees promise in the technology but knows that the road ahead will require technological savvy to separate hype from reality and wade through marketing speak.
This is one of the reasons he believes CIOs are well-positioned to take on the role of artificial intelligence pioneer in the enterprise. Another? "Given that CIOs and their organizations touch every system and every process in the company, I think they should harness the knowledge in their own teams to identify these use cases and then proactively bring them to the table," he said.
Waglé is introducing artificial intelligence projects to the enterprise in three areas. The first is to modernize business processes with robotic process automation (RPA). Sometimes referred to as the "dumb" end of artificial intelligence, RPA is used to automate repetitive, rules-based tasks that enable systems to talk to each other and are performed by humans. A second initiative is to closely monitor artificial intelligence capabilities being built out by its technology partners and to find potential applications for the tools at Equinix.
Who should lead the charge?
Waglé tapped his enterprise architecture team to take on AI for a few reasons: The team is embedded with the business and, so, has a unique perspective on the goals and challenges it faces; all emerging technology is incubated with this team; and the team has a complete view of the enterprise architecture footprint. Plus, he said, "I don't want my apps team and my infra team, who are consumed with delivering against business priorities, to be in contention for some of these incubation ideas that I want to follow through on."
A third, perhaps, least mature artificial intelligence project is a chatbot for employees called Eva, or Equinix virtual assistant. The user interface was built in house; the meat of the technology was built on Google Cloud and uses Google's natural language processing libraries and cognitive APIs.
It was designed to answer basic questions such as how to procure new office equipment or where one of the company's data centers is located and how to get there by combining different sources of search-intensive information under one interface. "If you look at our intranet, it's full of content, but it's not easily searchable or the results are not presented in a context-sensitive sort of manner," Waglé said.
One of Eva's key targets is new employees. "We felt we could make an impact to onboarding by presenting this bot where an employee could ask anything that's on their mind and then specify the intent and get a response back," he said. Rather than searching the intranet for a list of steps on how to file a ServiceNow ticket to get a new Skype headset, they can just ask Eva -- either in the office or via the mobile version of the chatbot.
Eva is a work in progress, and Waglé said employee adoption -- currently, the key metric for the project's success -- as well as the feedback on Eva's accuracy is rising. The feedback loop with employees enables continuous improvement, a valuable element for the AI project because mapping the intent of a question to the right response has been harder than anticipated. "We were expecting more self-learning to happen, to be very honest," he said. "But we're finding that we have to train the bot with much more manual effort than we would like to."
Still, when asked what advice he could give CIOs on how to build an AI practice of their own, he suggested that a "conversational information presentation system" could be a quick and easy win that that would help CIOs find their footing. "Then you can evolve into more transactional processing through these AI interfaces," he said.
Automate, augment, disrupt
Waglé's examples of artificial intelligence projects are internally focused to either automate processes or augment human intelligence with AI. According to Brent Leland, co-founder and partner at Cimphoni Consulting LLC and former CIO at Trek Bicycle Corp., automation and augmentation are two ready-made opportunities for artificial intelligence projects in the enterprise.
Automation can improve operational efficiencies with technologies such as RPA and robotics. While it takes time -- Leland said about a year -- to determine exactly what processes should be handed over to a bot, the ROI could be considerable. "A single bot can do the work of one to five workers and an annual cost at scale for less than $1,000," he said.
Augmentation can improve the performance and productivity of a company's existing resources. "It's basically a force multiplier," Leland said. In this category, Leland includes decision-making recommendation engines as well as virtual agents and conversational platforms. He also includes auto-generated content creation, such as sports stories written by robots, as well as augmented reality, which he believes will transform training and anything owner-manual related.
But the third category of enterprise AI has the potential to be the most transformative: Leland referred to it as disruption. "This is really about remaking your business models and creating new opportunities to drive market share," he said.
These opportunities are complex, involving disruptive technologies such as the internet of things and computer vision, as well disruptive techniques such as unbundling and crowdsourcing. Together, they create a potential to disrupt established players in the market.
AI and disruption
Wayfair, a Boston home goods company built on digital technology, is one such disruptor. The e-commerce business was founded in 2002; less than 10 years later, it started building out a formal data science program.
"We didn't even call it that then. We called it customer recommendations," said Wayfair's Clark, who initiated the data science program for the company. It's another way of saying that predictive models, recommendation engines and data are at the heart of how Wayfair conducts business -- and so are more advanced techniques like artificial intelligence and machine learning.
Indeed, being a disruptor is as much about cutting-edge technologies like artificial intelligence as it is about culture. Despite having gone through an initial public offering in 2014, Wayfair relies on lean startup principles including speed to market and experimentation, which has led to innovative services such as "search with photo" that uses image recognition technology to match a customer's uploaded photo to Wayfair products.
Clark, who left the data science team to "people who are better at math than me," described his role as chief architect for the company as central to Wayfair maintaining its nimble, startup mentality. One of his responsibilities is to empower the data science and analytics teams who are running artificial intelligence projects. He does this in two ways. The first is making sure they never have to wait for data. "If they have some theory that they want to explore, and if they have to wait for someone to move a large data set out of one data store and to another so that they can get access to it, that slows us down and that's sort of the kiss of death," he said.
Clark maintains three central data platforms for the company's data scientists. SQL Server is used for online transaction processing. "The typical problem with a setup like that is that you can't put everything in one place so that you can join it all if you need it," he said. That's what Vertica, a massively parallel processing column-store database, and Hadoop, a parallel processing framework, are for.
"When I say Hadoop, I don't just mean Hadoop," Clark said. "I mean all of the other stuff that typically comes with it. So, Hadoop plus Kakfa plus platforms that are able to give data scientists access to a combination of static data and streaming data and to look at those two things together."
The second is making sure "a good, modern version" of the tools needed to explore the data are accessible to data scientists and artificial intelligence practitioners. Wayfair data scientists tend to use a combination of RStudio, an open source development environment for the programming language R, and Jupyter Notebook, an interactive development environment that cuts across dozens of programming languages.
Clark said the work by data scientists and artificial intelligence practitioners is "a strategic capability." He meets with department heads on a regular basis to ensure the infrastructure is meeting the needs of the data explorers. "We really try to feed it as much as it has an appetite for," he said.