Bad news for companies swelling with pride over their new artificial intelligence project: The odds of pocketing any business value from one AI pilot are slim. They’re going to have to rapidly cycle through lots of projects to improve odds. It’s possible — companies just have to stop thinking of AI as magic and start plying the right data tools.

Companies that play the numbers game are getting tangible gains from AI, according to Rob Thomas (pictured), general manager of IBM Analytics at IBM Corp. “Don’t talk to me about the one or two AI projects you’re doing. I’m thinking, like, hundreds.”

Some companies’ idea of an AI project is nine months of hand-wringing, planning and staging before going live. This is not the kind of AI project that Thomas is referring to.

“These are not one-year-long projects with hundreds of people,” he said. Instead, companies should let spontaneous ideas for small projects tumble out into the world. Fail fast, try again, and repeat.

“You have to be comfortable that probably half of your experiments are going to fail,” Thomas said.

Thomas spoke with Peter Burris (@plburris), host of theCUBE, SiliconANGLE Media’s mobile livestreaming studio, during the IBM Innovation Day event in Yorktown Heights, New York. They discussed the formula for AI success and IBM’s recent data-virtualization announcement. (* Disclosure below.)

Virtualization breaks data-migration chokehold

Rapid iteration requires solid groundwork and de-siloed, governed data that’s ready for action. “Once you have a single instantiation of your data, it becomes very easy to train models,” Thomas said.

IBM Cloud Private for Data is one environment — a single console for all data. It recently introduced a feature for data virtualization that makes bringing together data from distant corners of the world for easier, according to Thomas.

The idea is: Leave data where it is — be it a data center in the U.S., a data center in India, an automobile, etc. The technology virtualizes data and federates it. It eliminates the bane of many AI projects — how to move scattered data to one place. That’s too costly and time-consuming; leave it where it is and virtualize it, Thomas explained.

With virtualized data, companies are freer to rapidly experiment. “The goal is how do you increase your win rate? And do you learn from the ones that work and from the ones that don’t work so that you can apply those?” Thomas concluded.

Watch the complete video interview below, and be sure to check out more of SiliconANGLE’s and theCUBE’s coverage of the IBM Innovation Day event. (* Disclosure: TheCUBE is a paid media partner for the IBM Innovation Day event. Neither IBM, the event sponsor, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE