Companies are rushing to use generative AI, and here’s where the tricky part comes in.

• A majority of U.S. companies are adopting generative AI tools, according to a new Bain survey.
• Also, most people are not sure where the value of these tools comes from.
• Lack of in-house expertise is the biggest obstacle to widespread enterprise adoption.

Companies are adopting generative AI at breakneck speed. This raises a tricky question: getting a return on these massive investments.

That’s the main conclusion of a new survey of the business world by consulting firm Bain & Company.

The survey surveyed 200 U.S. companies with revenues of more than $5 million. Half were high technology companies, with the rest spanning the retail/consumer goods, manufacturing, healthcare and financial services sectors. Key findings include:

• 85% of companies rank AI adoption among their top five priorities. 12% rank AI technology as their top priority. Only 1% say it’s not a priority.
• Language generation and software coding are two of the most common AI applications in businesses of all kinds.
• Companies reported spending an average of $5 million per year on generative AI.
• One-fifth of those surveyed said they spend more than $50 million per year on generative AI.
• Companies with over $5 billion in revenue had an average annual spend on generative AI of $13.1 million.
• Companies with annual revenues of less than $500 million spend an average of $1.6 million per year on the technology.

The big question: ROI

Most companies told Bain that generative AI is meeting or exceeding their expectations, but also said the business case for large-scale AI investment is unclear.

That’s a key question that AI giants like Nvidia, Microsoft, OpenAI, Google and others have yet to answer. These tech giants are betting that AI will be used on a mass scale in the future, but for the boom to continue, customers must see value in these new services.

Despite the rapid adoption of generative AI, only 11% of companies surveyed by Bain had a clear vision for how they would use generative AI and how it would add value.

It could be a matter of vision or talent, or it could be a problem with the tools themselves. While most respondents reported that the AI ​​technologies they tried met their expectations, a small minority felt that the tools fell short of their expectations.

Top concerns

Bain’s research found that concerns about the quality of generative AI’s output, combined with data privacy and security concerns, are issues preventing companies from adopting the technology more quickly.

In-house expertise is also in short supply, with concerns about expertise increasing, while concerns about performance and security are declining, compared to a similar Bain survey conducted last year.

Gene Rapoport, who leads the generative AI initiative at Bain’s private equity division, says CEOs need to be more aggressive in adopting AI tools, in part because while most companies expect generative AI to increase revenue and make existing employees more productive and efficient, far fewer have a full understanding of how this will happen.

Cost reduction

Those in the field building AI tools may be less concerned, citing the power of the underlying technology and the history of past innovations.

“There’s a natural cycle of investing in new technology and hoping for a return,” said Oren Etzioni, an AI investor and professor emeritus at the University of Washington in Seattle. “I’m very optimistic about what the return on investment will be.”

He stressed that ROI can and must be improved in two ways: tangible contributions to the bottom line are one, but cost reduction is another.

“Computer scientists are a field that is very good at reducing costs. Even now, 19 months after we started, we have demonstrated that the cost per query has come down significantly, and training has become more efficient,” Etzioni said.

Nvidia CEO Jensen Huang reiterated this urgency during a keynote at Computex in Taiwan earlier this month, denouncing “computing inflation,” where rising computing costs outweigh improvements in AI model performance.

“Obviously, this situation cannot continue,” he said.

Any tips or insights you’d like to share? Please contact Senior Reporter Emma Cosgrove at email address Or use our secure messaging app, Signal: 443-333-9088