Although major tech companies like OpenAI, Google, and Anthropic are still painting a picture of a trillion-dollar market, the actual figures from enterprise users are far less optimistic.
Three years after the ChatGPT boom, the wave of massive investment in generative artificial intelligence (GenAI) is facing a harsh reality.
According to the latest surveys, the majority of businesses are struggling to profit from AI investments, as the technology reveals many limitations such as being "too polite," inconsistent, and prone to fabricating information.
Although major tech companies like OpenAI, Google, and Anthropic are still painting a picture of a trillion-dollar market, the actual figures from enterprise users are far less optimistic.
A Q2 2024 survey by Forrester Research of 1,576 business leaders revealed that only 15% of respondents reported improved profit margins thanks to AI in the past year.
A survey by consulting firm BCG of 1,250 executives also indicated that 5% see value in this technology.
Brian Hopkins, an analyst at Forrester, noted that tech companies have fabricated the narrative that things will change quickly, but people don't change that fast.
Forrester predicts that by 2026, companies will postpone approximately 25% of their AI spending plans.
Besides cost, the quality of AI output is a major hurdle. A prominent issue is the tendency of large language models to "flatter" users, as they always try to please them rather than provide accurate advice.
Eric LeVine, CEO of the wine collecting app CellarTracker, said they spent six weeks testing and debugging just to "teach" their AI chatbot to give honest reviews.
Initially, the AI was too polite and didn't dare to criticize, even when the data showed that users wouldn't like that type of wine.
More serious is the issue of AI fabricating information. Jeremy Nielsen, CEO of Cando Rail & Terminals (North America), said the company spent $300,000 developing an AI product to summarize railway safety rules.
However, the project failed because the AI sometimes forgot the rules, and other times created its own regulations that were not in the 100-page legal document.
Given the limitations of AI, many businesses are returning to a model that relies on human staff, especially in customer service.
Klarna, the Swedish payment company, once claimed its AI chatbot could do the work of 700 full-time employees. However, by 2025, CEO Sebastian Siemiathowski admitted that they would have to readjust their strategy because customers still prefer to speak to a real person when facing complex issues.
Similarly, the American telecommunications giant Verizon is also refocusing on the role of call center employees. Ivan Berg, head of AI at Verizon, said that empathy is a key factor preventing AI from completely replacing humans.
According to him, about 40% of consumers still prefer the idea of speaking to a human and feel frustrated when they can't reach a call center operator. Researchers call this contradiction in current AI capabilities a "puzzling paradox."
AI can solve complex problems or write code quickly, but it fails at simple tasks like scheduling work or understanding geographical context.
At investment firm Prosus, an internal AI tool was unable to understand the concept of "last week" or differentiate between areas in Berlin when analyzing delivery data, forcing the company to incur significant costs to standardize input data.
To address this situation, AI companies like OpenAI and Anthropic are shifting their strategies towards a "hands-on" model.
They established teams of engineers who worked directly with clients to build specialized tools instead of generic models, acknowledging that effective AI application required far more intensive training and collaboration than initially expected.