Artificial intelligence was expected to speed up business processes, cut costs and raise productivity, but some companies are reassessing how they use the technology. Incorrect results, rising operating costs and unmet expectations are increasing scrutiny of broad AI adoption. According to the World Economic Forum’s report on the future of work, published last year, about 92 million jobs worldwide are expected to disappear or change fundamentally by 2030 because of AI and related automation processes.
Over the same period, about 170 million new employment opportunities are expected to emerge. That outlook has fuelled high expectations among companies, which want AI to accelerate workflows and reduce costs. Yet some businesses that have invested substantial time and money in deploying the technology as widely as possible are taking a closer look at its actual benefits, continuously rising costs and associated risks.
International consulting company Accenture is one example. According to company statements, it invested a total of $3 billion to develop its data and AI activities. More than $1 billion a year is allocated to employees’ professional training, with a significant share spent on dedicated AI training. It is also reported that employees who do not regularly use the company’s AI tools cannot be promoted to management positions.
This approach is said to encourage employees to use costly AI tools even for relatively simple tasks in order to meet usage requirements. Examples include creating presentations from text documents, summarising reports and carrying out simple programming tasks. Anne-Sophie Mayer, Professor of Digital Work at Ludwig Maximilian University of Munich, said: “A few years ago, these applications were relatively inexpensive. That was because, in a sense, what you paid with was your data.”
According to Mayer, the rapid development of AI has substantially increased licence costs for highly specialised models. Using these tools often requires the purchase of costly processing units known as tokens. After costs rose at Accenture, management reportedly decided to apply the brakes: only $29 per month will be allocated for AI applications per employee, while spending above that amount will require prior approval from company management.
Costs are not the only concern. AI-generated material can appear professional at first glance, but detailed review may show that it is superficial, inaccurate or lacking in meaningful new information. Rather than saving time, this can require employees to check information, correct errors and make outputs usable. The source describes this type of low-quality AI content, which creates additional workload, with the English term “Workslop”.
Ryan Roslansky, a senior executive at Microsoft, raised the same issue on his public LinkedIn account: “AI slop isn’t just on social media. It shows up in our day-to-day work too.” He said he received a document that was clearly prepared with AI, had his own AI summarise it, and ultimately found nothing that offered a new idea. “It’s a vicious cycle,” Roslansky said.
Mayer says using AI tools for basic summarising and routine work that does not require creativity makes sense. But she stresses that people must retain the decisive role in work requiring interpretation and assessment. Eric So of the Massachusetts Institute of Technology warns: “Outsourcing the work of thinking leads to the atrophy of cognitive abilities.” Mayer also underlines that AI makes mistakes and that responsibility for the final result always remains with people.
