GenAI suffers from data overload, so companies should focus on smaller, specific goals

“There is no AI without data, there is no AI without unstructured data, and there is no AI without unstructured data at scale,” said Chet Kapoor, chairman and CEO of data management company DataStax.

Kapoor was kicking off a conversation at TechCrunch Disrupt 2024 about “new data pipelines” in the context of modern AI applications, where he was joined by Vanessa Larco, partner at VC firm NEA; and George Fraser, CEO of data integration platform Fivetran. While the chat covered multiple bases, such as the importance of data quality and the role of real-time data in generative AI, one of the big takeaways was the importance of prioritizing product-market fit over scale in what really is still the early days of AI. The advice for companies looking to jump into the dizzying world of generative AI is straightforward — don’t be overly ambitious at first, and focus on practical, incremental progress. The reason? We’re really still figuring it all out.

“The most important thing for generative AI is that it all comes down to the people,” Kapoor said. “The SWAT teams that actually go off and build the first few projects — they are not reading a manual; they are writing the manual for how to do generative AI apps.”

While it’s true that data and AI go hand in hand, it’s easy to become overwhelmed by the sheer amount of data a company may have, some of it possibly sensitive and subject to strict protections, and maybe even stored across myriad locations. Larco, who works with (and sits on the board of) numerous startups across the B2C and B2B spectrum, suggested a simple-but-pragmatic approach to unlocking true value in these early days.

“Work backwards for what you’re trying to accomplish — what are you trying to solve for, and what is the data that you need?” Larco said. “Find that data, wherever it resides, and then use it for this purpose.”

This is in contrast to trying to splash generative AI across the whole company from the get-go, throwing all their data at the large language model (LLM) and hoping that it spits out the right thing at the end. That, according to Larco, will likely create an inaccurate, expensive mess. “Start small,” she said. “What we’re seeing is companies starting small, with internal applications, with very specific goals, and then finding the data that matches what they’re trying to accomplish.”

Fraser, who has led “data movement” platform Fivetran since its inception 12 years ago, amassing big-name customers such as OpenAI and Salesforce en route, suggested that companies should focus narrowly on real issues they’re facing in the now.

“Only solve the problems you have today; that’s the mantra,” Fraser said. “The costs in innovation are always 99% in things you built that didn’t work out, not in things that worked out that you wish you had planned for scale ahead of time. Even though those are the problems we always think about in retrospect, those are not the 99% of the cost you bear.”

So much like the early days of the web and, more recently, the smartphone revolution, early applications and use cases for generative AI have shown glimpses of a powerful new AI-enabled future. But so far, they haven’t necessarily been game-changing.

“I call this the Angry Birds era of generative AI,” Kapoor said. “It’s not completely changing my life, no one’s doing my laundry yet. This year, every enterprise that I work with is putting something into production — small, internal, but putting it into production because they’re actually working out the kinks, on how to form the teams to go and make this happen. Next year is what I call the year of transformation, when people will start doing apps that actually start changing the trajectory of the company that they work for.”