Companies scale back AI use as costs, poor outputs and risks mount

Companies that once rushed to adopt artificial intelligence are increasingly pausing and recalibrating as costs, disappointing outputs and unintended side effects pile up. Global forecasts that predict large labour-market shifts have driven big expectations from AI — faster processes, lower costs and higher productivity — but several recent corporate experiences suggest adoption is not yet the straightforward efficiency gain many hoped for.

Accenture's course correction and rising licence bills

One prominent example comes from Accenture. The consulting firm says it invested a total of $3 billion in building its data and AI capabilities and spent more than $1 billion a year on upskilling staff, including specialised AI training. That heavy investment was accompanied by an internal policy that promoted employees who regularly used AI-based tools, a pressure that in practice pushed staff to apply costly AI services even for routine tasks such as turning documents into slides or summarising reports.

As Ana Sofi Mayer, a professor of digital work at Ludwig Maximilian University in Munich, put it: "A few years ago the applications were relatively cheap because, so to speak, you paid with data." Mayer says the economics have changed as more specialised models and licences become available, and token-based usage can drive bills up quickly. Accenture has responded by sharply limiting discretionary AI spending: going forward each employee will have a $29 monthly allowance for AI applications, and expenditures above that will require managerial approval.

Beyond licence fees, companies report a second cost: time. AI tools often produce outputs that resemble professional work at first glance but, on closer inspection, are superficial, inaccurate or add no new insight. Rather than saving time, these outputs create extra tasks for staff who must fact-check, correct and rework machine-generated content — a phenomenon some observers have nicknamed "workslop."

"The AI-generated rubbish does not live only on social media. It exists in everyday work," Ryan Roslanski, a Microsoft vice president, wrote on LinkedIn. "Someone sent me a document that was obviously generated by AI. I read it and realised everything in it had been produced that way, then I asked my AI tool to summarise it. In the end I didn't find anything that brought a new idea... It's a vicious circle." Experts also warn that overreliance on AI can hollow out skills. "Delegating thinking to others leads to cognitive decline," said Eric So of the Massachusetts Institute of Technology (MIT).

Mayer stresses the human role: "That is really a big danger. AI constantly creates the impression that the human is right. As a result the human is not encouraged to question things. The tools get better and people naturally become more comfortable." She adds: "We simply must realise that AI makes mistakes. And responsibility for the final result always remains with the human."

Other organisational risks have surfaced as well. Some employees, worried that feeding accurate examples into internal AI systems will make them redundant, have deliberately entered incorrect data. High-profile reports about Meta — which let go roughly 8,000 workers after secretly recording employees' day-to-day work to train an internal AI system — have amplified employee scepticism about how data will be used.

Not all markets have embraced AI quickly. Research from the Institute for Employment Research (IAB) shows that two-thirds of companies in Germany had not used AI tools by the end of 2025, often citing a lack of clear use cases, missing expertise, incompatible IT systems or data-protection hurdles. Mayer recommends companies satisfy four preconditions before trying to scale AI: define the problem and the area of application, secure high-quality data, adapt processes (digital and analogue), and ensure employees are involved in development so tools fit real needs.

The wider context remains unchanged: the World Economic Forum's Future of Jobs report warned that by 2030 roughly 92 million jobs could disappear or be fundamentally altered because of AI and automation, while about 170 million new roles may emerge. For many organisations the immediate challenge is balancing those long-term structural shifts against the short-term practical and financial headaches of using AI today.

Photo: press material from the event