“Make new mistakes. Make glorious, amazing mistakes. Make mistakes nobody’s ever made before.”  -Neil Gaiman

The old adage taught to most of us when we’re very young—if at first you don’t succeed, try, try again—is up against some stiff competition. One of the most powerful cognitive effects in humans is the negativity bias: the proven condition in humans that negative events have a greater effect on a person’s psychological state than positive things. People hate losing more than they love winning, which leads to a natural tendency to avoid risk and subsequent failure.

But risk-taking—and failing—is necessary for success, especially in the tech space, and especially as Generative AI makes its way into virtually every piece of technology we use. Experts at the recent Data Universe in New York City addressed the notion of trying innovative and risky things, perhaps failing, and then evolving toward success while leveraging data and AI.

New approaches for existing problems

According to Benn Stancil, founder and CTO of business intelligence technology provider Mode, the current environment around GenAI bears a resemblance to the dot-com boom and bust years of the early 2000s. In fact, he says, it’s almost inevitable that similar—and necessary—mistakes will be made before we see products that find mass appeal. Spectacular failures are coming, and we should welcome them.

Stancil points to the grocery ordering app Instacart as an example to consider. In 2001, the economy was going through a rough patch as internet darlings were regularly—and very publicly—going out of business. One such company was Web Van. It had received $440 million from VC firms, had reached a valuation of $10 billion and was hailed as a new way to buy groceries online. Within two years of its peak, it was out of business. But, Stancil notes, if there was no Web Van, there would be no Instacart. Why did the idea for a company that imploded 20 years ago reemerge a decade later in a far more successful way?

Simple: lessons were learned, conditions changed and novel approaches were developed. And that process needs to play out again and again for each advancement.

“Successful companies are inspired by old ideas, but approach them very differently,” he said. “Netflix isn’t Blockbuster on the Internet, even though it started there. It became streaming, which wasn’t even a concept when the company started.”

Stancil notes the conversation about AI in 2024 is very similar to the one we were having about the internet in 1998. Approaches to integrating AI into products are following a familiar path as well—one that will need adjustment.

“Chat GPT comes out and is hailed as the most important advancement in decades and just about every company reacts to it by saying, ‘let’s take our product and put a chat bot on it,’” he laughed. “AI is probably going to be a big and important thing but we’re in the put-a-chat bot on it phase, not the phase where we’re delivering really valuable things.”

Businesses will need to experiment, fail, iterate, and evolve. Some won’t make it, but Stancil, and many others who spoke on the Data Universe stages, agree that those most prepared organizationally to accept the process of fail, learn, improve, are the ones best suited to succeeding in the era of Gen AI.

‘We need to be provocateurs’

While Stancil used the Data Universe platform to talk about how experimentation and failure will factor into successful AI products, Tony Mazzarella, director of Enterprise Data Enablement & Governance at Voya Financial, spoke about why it matters to an organization’s culture and people.

“We need to be provocateurs,” Mazzarella says. “We're heading into a very interesting time in business and in technology. And we need to challenge the status quo.”

Figuring out a technology that some believe will have the same impact as the Industrial Revolution will likely cause some collateral damage. From a culture perspective, he notes, organizations and employees have to be prepared for failure.

“Eighty-five percent of Gen AI initiatives will fail,” he flatly states. “It's going to happen.”

Robin Sutara, Field CTO for Databricks agreed that a culture of experimentation from leadership through the rank-and-file will be necessary to progress through this early phase of GenAI development. She warns businesses to resist chasing “the cool factor” and reminds them that Google—one of the most successful companies in history—experiences a 95 percent failure rate on its experiments. Something leadership, importantly, encourages.

“Make sure that you're thinking about experimentation versus actually showing immediate value for the company. There must be some level of experimentation.,” Sutara says. “It's just moving so quickly that the company has to be willing to have some risk appetite to allow you to experiment within the company or within the organization. You have to make sure that you're preparing your leadership and your board that there will be some level of failure in this Gen AI world as you figure out what works within your organization.”

As we’re building AI products, Stancil recommended we take a long view. The lessons of the dot-com boom were not internalized for years after the bust.

“There is no shortcut to this,” he said. “You can build an early version of something very quickly, and it’s tempting to release those things. But putting out a product that only works 80 percent of the time is very frustrating for people to use. Building a functional product often takes years of hard work, experimentation and failure.”