AI: Expensive and Inefficient? Corporations Scale Back Usage

The global job market faces significant shifts. A World Economic Forum report on the Future of Work projects that by 2030, around 92 million jobs worldwide will be eliminated or drastically altered by artificial intelligence and automation.

Simultaneously, an estimated 170 million new jobs are expected to emerge. This fuels high expectations for AI among many companies, promising faster processes, reduced costs, and increased efficiency.

However, some businesses are now reversing course. After investing considerable time and money to implement AI broadly, the focus is shifting to its actual utility, ongoing expenses, and inherent risks.

Highly Specialized AI for Simple Tasks

Accenture, a global consulting firm, provides an extreme example. The company reports investing three billion US dollars specifically in expanding its data and AI capabilities. Over a billion dollars annually was reportedly allocated to employee training, with a substantial portion dedicated to specialized AI programs.

Accenture allegedly pursued an aggressive AI strategy: only employees who regularly used internal AI tools were promoted to leadership positions.

This intense pressure reportedly led employees to use costly tools for relatively simple tasks to meet usage requirements. Examples include generating presentations from text documents, summarizing reports, or performing basic programming.

"A few years ago, applications were relatively cheap because the currency, so to speak, was data," explains Anne-Sophie Mayer, Professor of Digital Work at Ludwig Maximilian University in Munich.

"That has fundamentally changed. AI is getting better, causing license costs for highly specialized models to rise sharply."

Using these AI tools often requires purchasing expensive tokens, as was the case at Accenture. Costs there "exploded," prompting the company to pull back.

Moving forward, employees will have a monthly AI budget of just $29. Any expenses beyond this require approval from company leadership.

More Effort Instead of Efficiency

The use of costly AI tools for simple tasks wasn't the only cost factor. Often, AI-generated content appeared professional at first glance but was superficial, erroneous, or yielded few new insights upon closer inspection.

Instead of saving time, significant extra effort was generated. Employees had to meticulously review AI-generated information, correct errors, and bring results into usable form. This phenomenon, observed in many companies, has become known as 'AI sloppiness' or 'work-slop'.

Ryan Roslansky, a senior vice president at Microsoft, recently lamented this on his public LinkedIn profile.

"AI junk isn't just on social media. It's in our daily work too. Someone sent me a document that was clearly AI-generated. I read it, realized it was all AI, had my AI summarize it, and in the end, found nothing that offered new ideas... It's a vicious cycle."

AI expert Anne-Sophie Mayer refers to this as a problematic "homogenization of content." For simple summaries or routine tasks not requiring creativity, AI tools can be useful, the professor states.

"But when it comes to recommendations for action or interpretation, that still needs to be done by humans to generate added value and uncover new insights, ensuring I don't lose my competitive edge."

Does AI Make Employees Dumber?

Labor market researchers see another long-term problem: the loss of expertise. "Outsourcing thinking leads to cognitive decay," warns Eric So of the Massachusetts Institute of Technology (MIT).

If employees rely too heavily on AI, their own specialized skills could diminish.

"This is indeed a great danger," agrees Anne-Sophie Mayer, "because AI also always suggests it's right. This doesn't encourage questioning things. Of course, AI tools are getting better. And then, as a human, you naturally become very comfortable."

It becomes more problematic when erroneous information is not checked, is incorporated into internal processes, and spreads further. "You just have to realize: AI makes mistakes. And the responsibility for the output always remains with the human."

AI: A Motivation Killer?

Another issue for many companies is the often inadequate involvement of employees in AI implementation. "If I introduce a company-specific AI tool, I depend on my employees helping to train that AI tool," explains Anne-Sophie Mayer.

This also breeds many fears, "for example, employees thinking: 'If I help train these AI tools, I might become obsolete in a few years'."

Mayer recounts witnessing employees at some companies deliberately inputting incorrect information, fearing the AI tool might make them redundant.

Recently, reports about Facebook's parent company Meta caused a stir. The company laid off around 8,000 employees after secretly recording their daily work for internal AI training.

Hesitant Companies

In Germany, many companies still shy away from using AI. According to a survey by the Institute for Employment Research (IAB), two-thirds of German companies had forgone the use of AI tools by the end of 2025.

Companies often cited a lack of suitable applications. However, some also pointed to a shortage of expertise, incompatible IT systems, or data privacy hurdles as reasons.

Anne-Sophie Mayer understands this reluctance. She believes companies should meet four conditions for successful AI implementation.

"First: An application area must be defined: Where do I have a problem, and what could the solution look like? The second point is data quality. If I have a poor analog process, I will also have a poor digital process. If I don't possess certain data, don't prepare it properly, and then feed AI into it, it won't help."

Third, the expert states, it's still important to maintain a certain openness to the new technology – "finding the balance between the potential of AI without chasing every trend and using it blindly."

"And fourth: Employee empowerment. Encouraging them to find good application areas themselves, where they also recognize the added value of AI."