Smart chatbots and the broader trend of “generative AI” are dominating the headlines today—and they’re also topics on the minds of IT buyers and data experts at large companies.

At Battery Ventures’ Executive Innovation Retreat, held earlier this month in Carmel Valley, Calif., executives from more than 30 Fortune 500 and Global 2000 organizations discussed a host of new data/AI applications they are leveraging to run their companies more efficiently and serve customers better.

At the same time, many agreed that generative AI is still in the very early stages of adoption in large enterprises. What’s more, it’s often still difficult to assign ROI from some futuristic data initiatives, or to run new applications on sprawling IT infrastructure that is more old-fashioned (read: not running in the cloud) than many might expect. “If data is an asset, then I want to know what the ‘return on asset’ is,” one participant noted.

Still, overall, when it comes to AI, “there’s a lot coming, and it’s important to lean into it,” said one executive of a large, national insurance company who spoke on a panel at the event.

Many enterprises are still trying to get a handle on the foundational elements of generative AI and have “trust in the data,” which is leading to rapid adoption and use of technologies like data labeling, data observability and data governance. Some participants agreed it will take years for those foundational systems to harden before next-generation AI can truly proliferate within organizations.

Yet even small applications of AI can be impactful. Using data to more smartly target the sales of Medicare Advantage policies to seniors helped boost sales of that product by a few percent for one company that attended, which was impactful, according to one executive. Another spoke about using data/AI to pair customer-service agents from certain regions of the country with similar customers; the company found that some customers from the Northeast, for example, might not respond as well over the phone to an agent with a thick Texas accent.

Other executives spoke about successfully using AI internally, such as pairing avatars with chatbots to help employees navigate their benefits.

But for many, “AI is a real buzzword, but it’s hard to implement,” one participant noted. Regulatory roadblocks are one issue: In healthcare, many battles likely loom over exactly who owns health records, for example, and how deeply cloud vendors and companies can mine them for information to use commercially later on.

At the end of one panel, Battery General Partner Dharmesh Thakker asked the group if there was anything they wouldn’t trust AI to do. It was actually tough to come up with examples. One participant noted that a new company called DoNotPay offered to use AI to help defendants fight traffic tickets in court. It’s being challenged by lawyers who worry AI could put them out of business, according to the company. The company already claims to have used AI to help settle two million customer-service disputes and legal issues.

The Battery retreat also included short pitch sessions by a handful of Battery portfolio companies and discussions on non-data topics, including how foster positive DevOps cultures and get the most productivity out of software-development teams, which are often decentralized and difficult to manage today. Many participants were intrigued by a discussion of the so-called “DORA” metrics, four benchmarks for measuring DevOps performance. (These metrics were developed by Google’s DevOps Research and Assessment team after a multi-year study.)

The four standards—which some participants agreed could be a good way to engage both developers and business leaders inside companies—are deployment frequency; change lead time; change failure rate; and mean time to recovery.