AI Tools Failed To Give Returns To 95% Organisations, Says MIT Report

While many tech companies are fast switching to artificial intelligence-powered modules and also scaling down the workforce as part of cost-cutting, a study by the Massachusetts Institute of Technology (MIT) has claimed that majority of the projects have not made any impact so far. It said that over 95 per cent of the organisations that implemented AI systems were getting zero return on the investment.

There were reports of many tech giants investing heavily in the AI wing, expecting a boost in their profits, but the MIT report presents a completely different picture.

As part of the study, 300 AI deployments were surveyed and the researchers spoke to approximately 350 employees. The report stated that only a select few AI tools like ChatGPT and Copilot were found to be the most adopted models. Only five per cent of the integrated AI pilots are making profits in millions, while the vast majority remain stuck without any measurable profits.

In a report titled, The GenAI Divide: State of AI in Business 2025, it was stated that despite $30-40 billion in enterprise investment into GenAI, over 95 per cent of the organisations are getting zero return.” Further, it stated that over 80 per cent of the organisations have explored or piloted them, but only 40 per cent report deployment. “These tools primarily enhance individual productivity, not profit and los P&L performance. Meanwhile, enterprise-grade systems, custom or vendor-sold, are being quietly rejected,” the report stated.

The MIT study found that the failure on the investment was not due to the fault of the AI models but that the modules were hard to adapt with the pre-existing workflows in a company. Another major hurdle in AI-powered projects is the learning gap in their workforce but companies are blaming the performance of the AI models. It is worth remembering that Taco Bell Chief Digital and Technology officer Dane Mathews the fast-food franchise was slowing down AI rollout at drive-through restaurants as technology proved counterintuitive. He admitted that human workforce is better to take orders when restaurants get busy. At that time, he recommended that voice AI be used to handle the situation.

Meanwhile, recently Apple released a study claiming that new-age AI modles might not be as smart as they have been made out to be. The tech giant, in a study titled “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity”, said that some models designed based on reasoning like Claude, DeepSeek-R1, failed to reason.  These models are good at memorizing patterns but when the questions are altered or if complexity is increased, they fall apart.

This means that AI is failing to deliver the desired results when the patterns break or become complex.