Companies are rushing to invest in AI — but less than 20% of AI investments are resulting in the transformations that AI promises. VB Transform 2022 brought together business leaders from Intel, Wayfair, Red Hat and Aible to discuss how they’re beating the odds to actually harness the full value of AI.

“The word ‘transformative’ is the catchphrase there,” said Arun K. Subramaniyan, vice president cloud and AI, strategy and execution at Intel. “Twenty percent of the investments are actually reaping the benefits they were supposed to when you sold the project. And then whether they’re getting you the business outcomes at the level you wanted for that investment is really the question.”

Companies are beginning to walk rather than crawl; now it’s a question of how quickly they can get to the running phase, and then sustain that level of transformation. But transformation and business outcomes can take months, said Fiona Tan, CTO of Wayfair.

As a tech-enabled company in the digital space, focused on the home goods category, they’ve found the secret is focusing on practical applications of AI that tackle urgent business use cases. They’re also selective in terms of where they’re applying the AI and ML work that they do. But transformation takes time, she noted, because AI and ML capabilities are quite different than traditional software algorithms, which offer instant results.

“With a lot of AI and ML-based models, it will take a while. It’s very iterative,” she explained. “To that point, when you’ll see transformational change, we don’t usually see that in the first X number of days or weeks. That usually does take time for us. With us, customers are coming in. We’re learning from them. We’re adapting.”

Experience, iteration and adaptation are key for Arijit Sengupta, founder and CEO of Aible. Sengupta said he went through more than a thousand AI projects with his previous company, BeyondCore, which built technology for smart data discovery — and then wrote a book called AI Is a Waste of Moneyafter most of those AI projects failed. But he partnered with Intel to start Aible, an enterprise AI solution that guarantees impact in one month.

“When we started, nobody knew how you would get to value in 30 days. It was just rational to say that large companies can’t do this,” he said. “The good thing was I had done it more than a thousand times myself. My team had done about 4,000 AI projects. We knew where the bodies were buried. We could do it right the second time.”

It does depend on the individual enterprise more than anything else, said Bill Wright, head of AI/ML and intelligent edge, global industries and accounts, at Red Hat.

“I’ve spoken with some customers that have phenomenal development capabilities,” he said. “They’ve gone through all the DevOps and MLOps steps to make everything very efficient. There’s so much more under the covers.”

But some data scientists don’t realize all the work that goes into those production environments, how much can go right and can go wrong. Enterprises are at so many different stages of the journey toward understanding where their challenges lie, and how to tackle them. Success comes not only from iteration, but understanding the customer.

“It’s always about talking to the customer, understanding what their pain is, understanding what they’re going through,” said Wright. “All the technical advances I’ve ever experienced have been through customer conversations. I think that’s been the biggest lesson.”

Moving outside the AI/ML comfort zone

To hit the point of true digital transformation requires tackling bigger challenges, where the risks might be larger. For Wayfair, the most urgent problems to initially be solved were marketing and customer acquisition. They were able to automate and take some measured risks around bidding, which also deepened a lot of their customer strategy.

“As we got more and more experience, we took that and it morphed into, how do we understand the customer better?” Tan said. “It became the beginning of building up our customer graph. Expanding our AI and ML journey.”

They did a similar thing on the product side, mining product information from suppliers to augment and enrich data the company already has. Combining the customer graph that arose from customer acquisition and marketing efforts with their product graph allows the company to offer the best possible experience to customers in every search and shopping experience. And each step in the journey builds on the one before it, enriching current capabilities and opening up opportunities to use AI and ML in other areas.

“We sell big things that are hard to move and expensive to move. How can I use AI and ML for optimizing my supply chain — offer up a capability where ideally I serve you the most relevant green couch based on what you’re looking for, but I also want to make sure I can serve you one that’s at the fulfillment center closest to you, so there’s the least possibility of damage,” Tan explained. “That’s the culmination of pulling together all these disparate components to be able to offer up a solution.”

Often the issue slowing down AI transformation is too little sponsorship from leadership, Sengupta said, and too-large expectations.

“We figured out that if you go to [the leadership team] and say, ‘What kind of AI do you want?’, they want a flying car from Back to the Future,” he said. “The data may be able to give them a really fast boat or a medium speed car or a really slow plane. But when you start from the data and you can show them interesting patterns in the data and engage them early, they’re not asking for something crazy. Then you can give it to them.”

If you take the risk points, solve them early in the project, and iterate very fast, you can get to a good result, he added.

“Remember the difference,” Sengupta said. “I’m not saying you can do any AI project in 30 days. I’m saying you can have significant success from AI in 30 days. The two are very different. An iPad can’t do what a supercomputer does, but an iPad creates a lot of value.”

When winnowing down the pain points and business use cases to get to the right AI projects, where you are in your AI journey matters a lot, Subramaniyan said.

“But where the world is, the world of AI, in terms of the spectrum of development also matters,” he said. “We’ve all heard about how fast the world of AI is moving. We can actually take advantage of that rather than being intimidated by it.”

The amount of investment required to actually build a large model can be daunting, but once the models have been built, or you find them open source, it’s about taking advantage of that so you can leapfrog, he said.

“As business leaders, that’s something you can think about rather than thinking about the large investment,” he said. “In some ways it helps you to be a little late, because now you can learn the mistakes made by everyone else, and also leapfrog ahead of them. You don’t necessarily have to think about your business as being small or large, or competing with the large AI powerhouses. We’re taking that and making sure we can democratize across the board. That’s what Intel is working on, both from a hardware standpoint, but more important from a software standpoint. AI is a software problem first. Hardware is an enabler for that.”

Watch the full, in-depth discussion and catch up on all Transform sessions by registering for a free virtual pass right here.