Insights from category creators and the investors who believe in them.

How Enterprises Are Adopting AI—and What Founders Should Know

March 11, 2024

Building enterprise-grade software is never easy, and particularly in the fast-changing world of AI, the path to enterprise-readiness is murkier than ever. Founders are adjusting on the fly to navigate adoption roadmaps, security policies, and compliance requirements of their large prospective customers. 

In a recent roundtable discussion for founders in GGV Capital U.S.’s portfolio, I brought together Or Hiltch (VP, Chief Data and AI Architect at JLL) and Raji Subramanian (Chief Technology Officer at Opendoor*)—two of the most forward-thinking enterprise executives we know—and Gil Geron, CEO and Co-founder at Orca Security,* whose new AI addition has quickly become their fastest adopted feature. 

Here are five ways that these enterprise leaders are already adopting AI—and how startups can fit into the picture:

Evaluate AI use cases with an outcome-driven approach

For leaders looking to weave AI into their companies, Opendoor’s Raji Subramanian suggests starting with dozens of broad use cases that span several business areas over the next five to seven years. Consider halving those use cases into what you believe your teams can impact in the next couple of years, then brainstorm hypotheses around what that impact could be. This outcome-driven approach can give “very good insights into where we want to invest today versus the future,” Raji says. 

Whatever you do, don’t “peanut butter,” Raji says. Instead, “identify your top three, five, or six programs that you're going to really put a lot of muscle behind and generate the big-bang value.” 

When done well, this approach “can raise the ceiling of work—and the quality of the work that [AI engineers are] doing—for more higher-value work.” For startups, understanding their customers’ priorities is critical to ensuring they’re applying AI in ways that will materially impact their stakeholders’ success. 

The bottom line: The best enterprises are disciplined in prioritizing where AI can impact their business. Understanding those priorities can help startups scale quickly.

View GenAI as a game changer for democratizing data access

As models progress, “where we could use some help from generative AI is in the adoption of those tools,” JLL’s Or Hiltch says. “[It can be] really a game changer when it comes to user experience … All of a sudden, you don't have to be an expert user when you want to just extract data.” 

Democratization can make a big impact on platforms like Orca’s Cloud Security product that historically required users to have specialized knowledge. “While we want our platform to be very easy, very intuitive, there are still tasks that require you to know where to click, how to click, what to do,” Orca Security’s Gil Geron says. “Leveraging GenAI as a bridge—‘How do I produce reports? How do I fix a certain security issue? Where do I have a certain misconfiguration in my cloud?’—really bridges that language barrier.”

The bottom line: AI can help ensure fast and broad adoption of otherwise complex products.

Consider security and privacy concerns from Day 1

With increasing concerns around security and privacy, here are three ways to centralize AI tools and put guardrails in place:

Decide if you need to build your own version of ChatGPT: Rather than allow JLL’s thousands of eager developers to build individually, Or Hiltch describes launching JLL GPT. “Our challenge was actually to say, ‘Hey, guys, wait a second. Let us deliver this platform capability so that you could consume those foundation models via one central place that has all of that security, privacy, and performance baked into it so we don't end up building dozens of solutions separately.’ That’s one thing that has been key to our success with integrating AI.”

​​Limit the risk of exposing private data: At Orca Security, “we've basically narrowed down the options of what you can actually do,” Gil says. “We’ve made it a toggle-like approach where we can sanitize what's being sent to the third party without the risk of private data being exposed or with low risk of data being exposed … We even gave customers a choice which vendor they would like to use when we are leveraging a third-party engine. Combining all of that and wrapping it into the product allowed a lot of confidence in the way they are using it—and it’s one of the fastest adopted features in the platform as of today.”

Create a responsible AI framework: When buying AI products, Raji describes an evolving security policy that evaluates vendors’ data governance policies. If a vendor is “sending a lot of information out to OpenAI, you have to go deeper in understanding not just the security policies but how they're handling data governance … So it goes one level deeper not just in terms of how they're securing the systems from a technology perspective but also how they're dealing with the data.”

The bottom line: Security and privacy are top blockers for AI adoption. Startups need to understand their customers’ policies and build in compliance with them from the beginning.

Build a framework for build vs. buy decisions

Raji recommends taking the time to “create an organization that can very rapidly make decisions on what to buy.” 

Here are Raji’s three tips to keep in mind when deciding which AI products to buy—and which to build:

  • Be wary of “vendor proliferation”: Create a clear model of how you attach vendors to capabilities so you don’t wind up managing 10 vendors around the same capability.
  • Decide where AI sits within the technology org: “Either technologists need to wear multiple hats and play that business bridge in terms of value creation, or you need to be able to bring people across the organization who can.”  
  • Be patient when assessing value: “You're going to get a lot of false positives, and you're going to get a lot of false negatives during the journey of experimentation … Just because AI didn’t give you a positive result on Month Three, you don't necessarily ignore it—or because it gives you a fabulous result in Month One doesn't mean that it's going to give you long-term value. So I would separate out interpretability and deterministic nature of the AI models and how you use that to deliver value.”

The bottom line: To swiftly navigate sales cycles, startups should aim to understand how enterprises are delineating between what they should build vs. buy, and ensure they can position themselves in alignment with those frameworks. 

Always prepare for what’s next

Given that it has only been a year since ChatGPT’s release, what fundamental shifts should founders anticipate when building?

Holistic platforms are the future: “I’m pretty certain that—let's say one or two years from now—the ability to bring your own data into AI will be a lot more managed and streamlined compared to today,” Or says. “That's probably the biggest technical challenge today because supposedly there are textbook solutions, but none of them actually work when it comes to a company if your data is complicated enough. I think all of the AI model providers and the obvious big ones like Microsoft, Google, Amazon, and so forth—they will probably move to a more streamlined approach of connecting your data into their services.” 

“Let's see how much more innovation comes out of AI versus humans,” Raji adds. “That'll be an interesting race to watch.” 

The changing landscape could spell trouble for startups that are overly reliant on one model provider. Orca’s flexible approach, which allows customers to choose their LLM, is critical to “future-proofing” products. 

The bottom line: Remain flexible in your AI strategy. As we’ve seen, a lot can change in a year.

*Represents a company in GGV Capital U.S.’s portfolio