How corporate giants get their A.I. projects to the finish line

Whether at semiconductor giant Intel or retailer Gap Inc., companies across disparate industries are increasingly working to incorporate artificial intelligence into their operations. But one of the potential pain points is how to scale an A.I. project from the idea and gestation period to a full-fledged integration in the business.

To help prevent hitches, Sundari Mitra, chief incubation officer, corporate vice president, and general manager of the emerging growth and incubation (EGI) group at Intel, believes that “if we start doing everything too big of a scale right from the get go, it doesn’t work,” she told the audience at Fortune‘s Brainstorm A.I. conference in Boston on Tuesday. So with A.I. projects, “the strategy we’re using is: go small, go really fast, but connect them up.”

At Intel, Mitra explained, leaders are able to get those ideas from incubation to integration by allowing engineers to have assignments, approved by their manager, to work on projects for typically 12 weeks at a time—funded by Intel’s incubation team. After that, they “figure out how can we roll it into the right business unit so that it’s going to be successful,” she said.

Ray Bajaj, the senior vice president and chief technology officer of Cardinal Health, noted that “for us, we don’t look at [it] as big or small—we look at the strategic focus areas of the organization” and how A.I. can be leveraged to support those. At Cardinal Health, that includes areas like managing the company’s supply chain (a theme that came up often at the conference) and improving patient outcomes with medication adherence and chronic-care management.

“The biggest stumbling block” in scaling up from proof-of-concept to integration, according to Vimal Kohli, the head of data science and analytics at Gap, is a lack of clarity around the reason for a project. “What’s the purpose?”

In that sense, thinking about projects fitting into a bigger goal is also something Kohli believes is important. “We’re trying to get away from initiatives that are episodic and individual. We’re trying to go after larger outcomes” to grow the business, and “the ideas fit into those long term goals,” he said. At Gap, that could include projects like working to reduce return rates from online shopping by addressing issues with sizing charts and fit.

Whether on a big or small project, Tony Kreager, the vice president of data engineering and data science at FedEx Dataworks, has a very particular method for getting something across the finish line. “Our framework for delivery is six days, six weeks, and six months,” he said. “Within six days, we normally have within our experimentation environment some form of output somebody can touch and feel.” Once the relevant business unit has been able to examine the idea, he said, “they’re immediately giving us feedback within six days, and it’s ugly, I’m not gonna lie.”

But within six weeks, Kreager continued, “we’re deciding whether we’re going to actually build out the hardened infrastructure to do it at scale,” and within six months, they’ve built out product teams and are working to deploy it in the business.

Key to Kreager’s strategy is to give things a chance—at least for a little while. He said he and his colleagues almost never say “no” to an idea “until after the six-day period that proves it’s showing no promises, or the cost to do it will be too high,” he said. “We don’t kill innovation at all until six days of trying it.”