Deloitte sees enterprises adopting AI without revenue lift • The Register

Making money isn’t everything … at least not when it comes to AI. Research from professional services firm Deloitte shows that, for most companies, adopting AI tools hasn’t helped the bottom line at all. But researchers still sing the technology’s praises.

According to Deloitte’s “State of AI in the Enterprise” report [PDF], 74 percent of organizations want their AI initiatives to grow revenue, but only 20 percent have seen that happen.

The consultancy’s findings echo a recent PwC business leader survey that found only 12 percent of CEOs saw both lower costs and higher revenue as a result of AI investments.

Deloitte’s explanation of the current state of affairs is that money isn’t everything.

“[S]uccess with AI isn’t just about boosting efficiency or even growing revenue,” the report says. “It’s about achieving strategic differentiation and a lasting competitive edge in the marketplace.”

Corporate AI investment hasn’t been entirely futile. Among the 3,235 business and IT leaders from around the globe who participated in Deloitte’s survey, 25 percent said AI is having a transformative effect on their organizations, up from 12 percent a year ago.

Asked about what benefits AI is actually providing today, 66 percent said it’s improving productivity and efficiency. How that works when only 20 percent report revenue growth is left unanswered. We note that a study published last year by non-profit METR found AI coding tools made developers less productive, despite expectations to the contrary.

Even without a compelling financial reason, workforce access to AI tools is expanding. Under 60 percent of workers now have access to IT-sanctioned AI tools, up from 40 percent a year ago. But among these AI-enabled workers, fewer than 60 percent use their AI tools as part of their daily workflow.

“This suggests that while access is widening, enterprise AI remains underutilized, and its productivity and innovation potential are still largely untapped,” the report speculates. 

That said, more AI pilot projects look likely to move into production. Currently, 25 percent of organizations say they’ve shifted 40 percent or more of their AI experiments into live use. That number is expected to reach 54 percent of organizations within the next three to six months.

The deployment of AI within companies appears likely to affect jobs as Deloitte sees it. 

“Within a year, more than a third of surveyed companies (36 percent) expect at least 10 percent of their jobs to be fully automated,” the report says. “The majority of surveyed companies (82 percent) expect at least 10 percent of their jobs to be fully automated when looking out three years.”

This expectation, however, hasn’t been accompanied by much organizational change. About 84 percent of respondents said they have not redesigned roles based on AI capabilities.

The people filling those positions also remain unconvinced about AI technology. Among non-technical workers, just 13 percent are highly enthusiastic about AI and trying to use it, according to the report. While 55 percent are open to the technology, 21 percent would prefer to avoid it and four percent actively distrust it.

Ultimately, companies committed to AI need to convince their employees that the technology has benefits beyond automating jobs away.

“The organizations succeeding with AI aren’t just investing in automation and algorithms, they’re investing in their people,” said Jim Rowan, US head of AI at Deloitte, in a statement. “As AI continues to spark new ways of working, this dual focus – advancing both the capabilities of their talent and AI tools – empowers teams to embrace reimagined business models and sets the foundation for competitive advantage.”

Another concern is “sovereign AI,” meaning that companies control their AI software and data in accordance with local laws and regulations and aren’t dependent on foreign vendors or infrastructure. Eighty-three percent of respondent companies say sovereign AI is at least moderately important to them, and 43 percent say it’s very important or extremely important.

With regard to agents – AI models given access to tools – usage is modest at the moment, but is expected to rise. Twenty-three percent of companies report using agents at least moderately today, and two years from now, that figure is projected to reach 74 percent.

The slow uptake may prove beneficial because just 21 percent of companies report having a mature governance model in place for autonomous agents.

Ali Sarrafi, CEO and co-founder of Kovant, an enterprise agent platform, told The Register in an interview that the problem with the way people use AI is that they see it as a form of fancy workflow automation. 

“There are studies out there to show that personal productivity is not actually going that far if you do that,” he said. “People start using it. But as soon as they get bored by it, they go back to how they did things before.”

The big change, he said, where companies start to see revenue results, comes from giving AI agents a job as if they were a coworker and running those agents automatically.

“We’re working with this big large manufacturing company,” Sarrafi explained. “They have about 7,000 suppliers. And every single time they needed to restock something, they had to coordinate with so many suppliers. It’s actually the most boring job ever for everyone. But then they deploy this agent worker or a team of agent workers that basically monitors the stock levels. As soon as it goes below the forecast requirements level, it sends a preliminary email to the supplier saying, ‘Can you tell us if you can supply this and what price?'”

The result is a summary report sent to Microsoft Teams that a company planner has to review and approve. If allowed to do so, Sarrafi said, the agent then sends the purchase order and follows up with the supplier until the goods reach the warehouse.

“So all that manual, annoying work, all of a sudden, it’s actually saving about 95 percent of that,” he said.

With regard to Deloitte’s report, he said the consultancy’s emphasis on governance can be addressed in product design – designing AI workflows with care.

“They make governance a big deal, but actually you need to have a nimble model of governance,” he said. “It’s the same way as when you hire people, you create governance around them. The information classification needs to be dealt with in the beginning. If you’re actually just opening up the entire world into the agent, then of course, it’s a statistical model – it might create problems. So it is more of a design problem in my mind than a need for massive governance architecture.”

Sarrafi also said that Deloitte’s findings about worker AI hesitancy can be attributed in part to unwieldy enterprise tools. “Most of the enterprise AI tools that are built, applications that are built for employees, they actually are not on par with what they expect in terms of user experience,” he said, adding that people don’t want to switch between multiple tools. 

Successful implementations, he said, tend to allow people to interact with agents through familiar tools like Microsoft Teams or Slack.

“I’m not going to name products,” Sarrafi said, “but most of the enterprise AI tools right now are about a year or two behind the consumer ones in user experience.” ®