ROI from AI is still a work in progress

Everyone’s talking about the potential of AI—but what about the cost of putting it into practice?

FWildly fluctuating token usage (and spending) is one strong indicator of the AI space’s chaos. For enterprise IT executives, pricing and cost controls are becoming more important as the capabilities of AI are varying widely, Skan AI CEO and co-founder Avinash Misra told IT Brew.

“The world is still learning to have the most optimal, well-architected approach towards delivering value with AI in those scaled workflows and complex business processes,” Misra said. “The translation from personal productivity to enterprise productivity in complex business processes is an art, a science that is being learned by enterprises.”

Cashflow questions. AI investment remains high but the ROI is still debatable. Using tokens to measure AI usage has made the disconnect clear, Retool CEO David Hsu told the WSJ Technology Council Summit on September 16.

“We believe that about 90% of the tokens you’re spending are probably negative ROI,” Hsu said. “It’s just that last 10% of tokens that are really driving a lot of that ROI.”

The implementation side has seen some wild swings in token use. For many organizations, it’s a question of how to balance promise and reality; AI demos often don’t reflect the technology’s actual use once integrated into workflows—and a recent report from Tempo AIshows that the disconnect is having an impact on how AI is actually being deployed.

Numbers game. Tempo found that while 91% of surveyed tech leaders are piloting or using AI in the project delivery process, actual ROI is hard to gauge, with 42% reporting they don’t have a clear throughline on spend.

Vic Chynoweth, Tempo CEO, told IT Brew that enterprises are between a rock and a hard place because they have to deploy the technology to keep up with their peers, even if the path to revenue is unclear.

“Now, how do we govern and make sense of it so we deploy it in the places that we’re actually getting returns? Ideally, optimized returns, but that’s where people are stuck, how do they measure it?” Chynoweth said. “Most of the measurement tools out there are code-created; AI is commoditized code creation, so that doesn’t really tell you anything.”

Let me at ’em. Solutions exist, notably the potential subsidization of AI, to allow enterprises to move forward while the industry works out profitability.

As costs go up, usage goes down. That’s a problem for organizations looking to maximize the output from AI-assisted workflows. As IT Brew has reported, localized AI solutions and open-source models are seeing increased attention from enterprises trying to find efficient solutions.

Subsidization is another path forward, and the one Misra believes executives across industries will continue to advocate for.

“There’s a massive appetite to bring in AI, and that appetite will make sure that the current level of subsidy, that level of investment, will continue at least for the near future,” Misra said. “Every large organization realizes that we are in a state of rewrite and a new set of competition and everything is up for grabs.”