A new Wharton study finds two thirds of businesses are actually seeing a positive ROI from generative AI. Is it a watershed moment?

The Gist

• ROI reality check. MIT’s gloomy “95% failure” stat may not tell the full story — Wharton’s new study shows most firms seeing measurable returns from generative AI.
• Definitions matter. Wharton’s broader lens — including productivity and retention — paints a more optimistic picture than MIT’s profit-only criteria.
• Winners and laggards. Smaller companies and senior leaders are reporting stronger AI payoffs, while large enterprises and middle managers lag behind.

For months now, AI watchers have been citing an MIT study that found 95% of businesses weren’t seeing an ROI from their generative AI spend. The study’s popularity seemed to confirm what many close to the technology felt to be true: It was difficult to implement, often unreliable and disruptive enough to upend business processes and make things a mess.

These concerns may all be legitimate, but a new study from the University of Pennsylvania’s Wharton School complicates the narrative. The study found that 74% of businesses that measure the ROI from their generative AI efforts are already seeing a positive return, and more expect to see a positive ROI within the next two or three years.

“These results are more positive than I was expecting,” Wharton professor Stefano Puntoni, a study author, told me in a phone interview. “People are finding this technology is powerful. So it’s not just hype.”

Table of Contents

• Why the Wharton Study Tells a Different Story
• Generative AI ROI Varies By Industry and Role
• Methodology and the Bigger Picture

Why the Wharton Study Tells a Different Story

The Wharton study’s more positive findings stem, in part, from the broader set of use cases it considered. Where the MIT study looked at whether generative AI pilots made it into production and delivered a profit, Wharton’s incorporated a range of uses, including individual productivity, employee retention and operational efficiency.

“Our definition is very different from the MIT report,” said Puntoni. “Theirs is much more stringent.”

The Wharton study doesn’t immediately refute the MIT report, but it does suggest that the tech industry’s rollout of generative AI won’t fit into a simple story. It’s yet another data point among many about a nascent technology’s viability. And even with its more positive results, it shows anything but a smooth and uniform rollout.

Related Article: The Three Faces of Generative AI

Generative AI ROI Varies By Industry and Role

The Wharton report, for instance, shows jagged adoption of the technology, with some industries showing much greater enthusiasm for generative AI than others. Eighty-eight percent of respondents in tech and telecom said they were seeing at least a moderately positive ROI from generative AI technology, for instance, and 83% in banking and finance said the same. In retail, only 54% of respondents said they were seeing a positive ROI so far.

There were also divides by seniority. Eighty-one percent of VPs and above said they were seeing a positive ROI from the technology, while only 69% of those in manager or director jobs said the same. Middle managers, the Wharton study found, were twice as likely than VPs and above to say it’s too early to tell whether there’s a generative AI ROI or that they were still in pilot phase.

“Across several questions, several indicators, we often see more positive reactions to these Gen AI programs among very senior leaders than middle managers,” said Puntoni.

Smaller companies also tended to fare better with generative AI than massive companies. Only 57% of those with more than $2 billion in annual revenue said they saw a positive ROI so far. Box CEO Aaron Levie said the discrepancy was related to smaller companies’ ability to adapt faster.

“The amount of output you get from AI agents will be directly correlated to how much you change (or reset) your workflow,” Levie said. “Smaller companies definitionally can change their processes more easily or start from the ground up. This is clearly harder in big companies.”

Learning Opportunities

Methodology and the Bigger Picture

The Wharton study is self reported, an important caveat, and includes responses from more than 800 people in U.S.-based companies with more than 1,000 employees and $50 million in revenue. It was done in partnership with the GBK Collective, a company that does some AI advising. Wharton paid for the study. GBK Collective partner Jeremy Korst, an author on the study, did not respond to an interview request.

For the generative AI boom to persist, the technology’s end users will need to find a clear return on their spending. Without that, all the infrastructure investment in the world won’t matter. And even then, the generative AI financial experiment’s payoff is far from guaranteed.

“The numbers are so large,” Puntoni said. “I’m not in a position to say whether that makes any sense.”

Learn how you can join our contributor community.

About the Author

Alex Kantrowitz is a writer, author, journalist and on-air contributor for MSNBC. He has written for a number of publications, including The New Yorker, The New York Times, CMSWire and Wired, among others, where he covers the likes of Amazon, Apple, Facebook, Google, and Microsoft. Kantrowitz is the author of “Always Day One: How the Tech Titans Plan to Stay on Top Forever,” and founder of Big Technology. Kantrowitz began his career as a staff writer for BuzzFeed News and later worked as a senior technology reporter for BuzzFeed. Kantrowitz is a graduate of Cornell University, where he earned a Bachelor of Science degree in Industrial and Labor Relations. He currently resides in San Francisco, California. Connect with Alex Kantrowitz: