GitHub and Genpact execs talk measuring AI ROI

At many companies, AI is moving out of the experimental phase.

“I think most CFOs are now viewing this as a real investment,” Alfred Sanders, CXO forums program leader at consulting and outsourcing firm Genpact, told CFO Brew. “It’s no longer ‘Are we going to implement AI?’ It’s more like, ‘What is the actual business outcome that we’re going to get out of this?”

But tracking the ROI of AI can be challenging. Companies and even departments within them are at different levels of AI maturity. Organizations may use a bewildering variety of AI applications, some of which have different pricing models. Complicating things is the fact that AI costs have risen, pressuring CFOs to quantify the value of their organizations’ investments in the technology.

When it comes to tracking AI ROI, “I can’t say that there is one golden ticket,” Richard Paik, CFO of GitHub, told CFO Brew. Paik came to GitHub, one of the largest code hosting platforms in the world, after more than 20 years at Microsoft. He and Sanders shared with us their advice for determining AI ROI.

Tokens ≠ ROI: To begin with, Paik has advice on what not to do: “Never measure the tokens,” he said. The concept of “tokenmaxxing”, he jokes, definitely wasn’t developed by a finance person. “We look at that and are like, ‘There’s no way I’m gonna measure productivity based on how much people spend on tokens,” he said.

Similarly, reduced labor costs are no longer a primary consideration when gauging AI ROI, Sanders said. (He’s heard more mentions of “stable headcount” recently, though, whereby companies aim to use AI efficiency gains to avoid adding new personnel.)

Instead, he said, “the ROI that’s being measured now is really on margins, on growth, on speed, on risk reduction, and converting all of that into some sort of business value.”

Paik advocates a project-based approach to measuring ROI. “The outcome is going to be, ‘What are you doing with those tokens?,’” he said. “And that’s getting measured in the products [staff] create,” like tools that can be used internally or externally, or changes they make to processes.

Get techy. To do that, though, takes “a strong partnership between finance and engineering to understand what are those things you can measure [and] the timeframe you can measure it,” Paik said.

Since joining GitHub in July 2025, Paik’s had a crash course in “the constraints of GPUs and CPUs and storage,” he said. “I really need to understand inference and throughput and all of the costing of how much goes into a GPU and [how] the efficiencies work,” he said.

The education goes both ways. ROI discussions are “where finance adds a lot of value” to engineering, Paik said. “Finance needs to step in and say, ‘Explain to me the product you created’ but also ‘What are the unit economics of that?’”

Soft ROI. Some value gains from AI will be “soft” and more difficult to quantify, Sanders and Paik agreed. “It’s very difficult to say ‘We’re going to expend X and then this is our ROI from a dollar perspective,’ Sanders noted. “That was always a challenge, even with any software.”

But soft ROI still matters, Paik said. For instance, giving employees AI tools can prove a recruiting advantage, and can even increase staff’s well-being by automating redundant work. “You’re optimizing a process, you’re improving work-life conditions and satisfaction with the role,” he said. “They’re learning and doing something new. These might be soft ROI measurements, but they’re tangible.”

The token economy. With the price of AI escalating, CFOs must also pay more attention to the cost side of the equation. Tokens are becoming a resource that finance teams need to steward, Paik said.

“The best CFOs right now are thinking about resource redeployment,” he said. “Who gets the tokens, and how do I allocate those tokens toward the best ROI that I expect, or the needs of the company?” He thinks of tokens “a lot like dollars in the bank,” he said. “It’s a budget.”

At a company like GitHub, coders can start to view them as a form of compensation, Paik said. “What excites a coder or developer is being able to create amazing products, and they can only do this if they have the resources.”. Tokens are part of “how we’re recruiting those top talents and how we’re retaining them.”