Feature
If you've been following the seemingly non-stop coverage of the AI insurgency in the workplace, you'd think workers everywhere are free of low-level work. The reality? Over three-quarters, 77%, say AI has decreased productivity and increased workload. While generative technologies were supposed to free them from workplace drudgery, they've done the opposite.
Don't believe it? Look at the numbers. They tell a story that Silicon Valley's AI evangelists would prefer you ignore. A report by S&P Global found that 46% of companies surveyed said that no single enterprise objective had seen a "strong positive impact" from their generative AI investments.
"Considering the investments that organizations have made into generative AI, and the clear opportunity costs, the performance of these applications is concerning," wrote the authors of the report.
GenAI Returns, By the Numbers
S&P Global isn't alone with these kinds of findings. Gartner, in its July Hypecycle for Artificial Intelligence, found that despite an average spend of $1.9 million on GenAI initiatives in 2024, less than 30% of AI leaders reported their CEOs were happy with return on investment. Not only that, it also determined that AI has entered the trough of disillusionment where it "becomes apparent that the initial enthusiasm was misplaced when many limitations become clear or issues arise with technology. This leads to a decrease in enthusiasm as people question its viability and potential applications."
This is already ringing true. According to the Economist, 42% of companies are abandoning most of their generative-AI pilot projects. That's a 17% increase from 2024. Not only that, but a study by the Rand Corporation concluded that by some estimates, more than 80% of AI projects fail — twice the rate of failure for information technology projects that do not involve AI.
"Those numbers surprise me a little bit," Holger Mueller, principal analyst and VP Constellation Research, Inc. told Reworked. He noted that many of the failures are the result of "vendors who are not able to make AI work." Dion Hinchcliffe, vice president of the CIO practice at the Futurum Group was less surprised by the AI failures. "The 2030s will be (about) those who survive today's POC [proof of concept] abandonment rates because they're treating them as tuition, not failures," Hinchcliffe said.
Cost Savings Is the Wrong Short-Term Goal
Sometimes tuition comes at a high price. Consider that in May of 2022 Swedish fintech company Klarna laid off around 700 employees. Not long after, the company's CEO Sebastian Siemiatkowski told CBS News, "I am of the opinion that AI can already do all of the jobs that we, as humans, do." In August of 2024 he claimed that Klarna's AI assistant was performing the work of 700 employees, reducing the average resolution time from 11 minutes to just two.
But Klarna's customers weren't happy with the service they were receiving and complained so loudly that the company was practically forced to hire human customer service agents to replace AI. Siemiatkowski admitted that the previous AI approach led to "lower quality."
The lesson learned? Hinchcliffe told Reworked that the CIOs he speaks with are "mostly looking at cost savings right now." But that might be the wrong goal, at least in the short term. "Total transformation of their business is on top in the next 3-5 years," he said.
Weighing the Risks of GenAI Use – or Lack of Use
While Klarna's efforts with AI versus human customer service seem enthusiastic, broadcasting the company's wins and losses to the world might seem ill-advisable to some. But then again, sometimes customers do it for you. Several years ago McDonald's partnered with IBM to develop a generative AI powered drive-thru ordering system at about 100 of its drive-up kiosks. Some of the results went viral. One account published on TikTok shows two people laughing as hundreds of chicken nuggets are added to their order. In another, published by the New York Post, bacon topped a customer's ice cream cone.
Though McDonald's and IBM worked together to eliminate such problems, the hamburger maker later shut down the project. "After a thoughtful review, McDonald's has decided to end our current partnership with IBM on AOT [Automated Order Taking] and the technology will be shut off in all restaurants currently testing it no later than July 26, 2024," the company stated.
The lesson learned? "Don't use AI in critical parts of the business without measures to verify results before they are used," said Hinchcliffe (speaking in general, not specifically about McDonald's and IBM project).
While it's easy to understand why AI projects at Klarna and McDonald's were shut down, Mueller told Reworked that others fail because "the technology is moving so fast and projects are moving too slow, the underlying technology needs to get upgraded and changed."
Mueller went on to explain that many first-generation projects were tied to a certain LLM vendor and when the LLM vendor's tools were no longer supported, the company didn't have the budget to make the necessary change. "They need a lot of help obviously and the system integrators are just gearing up," said Mueller.
The bottom line: Enterprises that develop AI-based solutions need to account for risk from the start. However that option is preferable to, "losing experience using AI and being outpaced by competitors who are already investing in institutional AI literacy and operational muscle," said Hinchcliffe.
3 Imperatives for Early AI Adoption
Hinchcliffe shared three rules of thumb to build early AI momentum:
- Focus ruthlessly on fit-for-purpose use cases — Target narrow, high-impact workflows where value is most easily accessed and where GenAI complements, not replaces, human expertise. Don't exclusively chase moonshots.
- Build AI fluency across leadership — Make sure executives have a good understanding of what AI can and cannot do today. Unrealistic expectations kill more POCs than technical limits. The CEOs Hinchcliffe talks to realize total transformation of their business is the goal in the next 3-5 years. Help them map out that path in more detail.
- Engineer trust into the stack — Invest early in AI governance frameworks, data quality pipelines, and model observability to reduce failure modes. Buy down risk with experience. Don't use AI in critical parts of the business without measures to verify results before they are used.
Conference
Sep
16
HR Tech Conference Las Vegas 2025
Register
Conference
Oct
27
Gartner HR Symposium/Xpo Orlando 2025
Register
The mantra: Scale the organization's readiness before scaling the models.
Related Articles:
- Big Tech Bets Billions on GenAI, But Adoption Is Slow — Companies like Microsoft, Google and Salesforce are betting the house on generative AI, yet adoption rates lag far behind the investment. Here's why.
- What It Takes for GenAI Pilots to Scale — Alan Pelz-Sharpe, Rebecca Hinds and Craig Durr join Three Dots to discuss why individual productivity gains with GenAI haven't scaled across the organization.
- How Businesses Are Turning Generative AI Into Measurable Value — Successful GenAI rollouts share at least one thing in common: the initiative starts by identifying and aligning with a core business goal.
