Key takeout
- MIT's research reveals that 95% of generative AI pilots are unable to provide measurable return on investment.
- This creates what experts call “genai divide” between companies that stick to basic chatbots and companies that use smarter systems that actually learn.
- Organizations fail not because of model quality or infrastructure issues, but because most workplace AI tools forget everything after each conversation and fail to adapt to the specific needs of the company.
- Experts say the windows for improving products in this high-growth space, providing on-demand and competing are narrower.
Companies have invested billions in AI tools like ChatGPT, aiming to reduce costs and increase productivity. But it's not working.
A new MIT study found that 95% of companies using AI have not regained their money. Despite all the hype about AI changing everything, most companies are stuck with expensive tools that aren't available.
The question isn't the technology itself, but how companies use it. What is the problem and why so many AI projects fail?
Learning gap
Popular tools like ChatGpt and Copilot are celebrated to increase productivity. However, when it comes to boosting the profits of a company, it is not successful, primarily because it does not learn or adapt to the company's needs.
ChatGpt discovered by the report will not forget the context and learn or evolve. In other words, it cannot be used for mission-critical tasks that replace humans. “It's great for brainstorming and initial drafting, but I don't retain knowledge about client preferences or learn from previous editors,” the lawyer told the researchers. “It repeats the same mistakes and requires extensive contextual input in each session. High-stakes work requires a system that accumulates knowledge and improves over time.”
Tip
AI can save millions by replacing outsourced business services, reducing agency fees and automating compliance checks. However, companies often follow attractive projects that appear good in presentations but rarely rewarding, research suggests.
Go solo
Reports suggest that many companies are stripped away by spending millions of people on their own generated AI systems that appear to fit their internal systems. Enterprises are making better profits when partnering with mature, supported third-party vendors.
Only 5% of custom enterprise AI tools reached production, but “most failed due to fragile workflows. [a] Lack of contextual learning and inconsistency with day-to-day operations. This included chatbots, a popular investment area.
The success rate was 67%, and teaming up with a specialized third-party vendor proved to be a better play. But that also presented a challenge. Executives complained about being hit by proposals, saying many of the solutions are “fragile, violating actual workflows or are essentially “science projects.”
Companies invest heavily in custom products, while employees vote on keyboards. As the report points out, “Only 40% of 40% of the companies surveyed reported that they purchased official LLM subscriptions, but more than 90% of workers in the companies surveyed reported regular use of personal AI tools for work tasks.”
Companies misunderstand AI
When it comes to AI, MIT researchers have discovered that money often ends in places where it doesn't do the best.
According to the report, most companies have poured their AI budgets into sales and marketing. But real money slaves were a boring task no one could talk about. It involves automating documents, processing invoices, and handling daily management tasks.
That was not the only mistake. Companies also tried to do more at once, rather than fixing one problem very well. Instead of having people actually work decide what tools they need, they relied on special “innovation labs” and AI teams.
The report discovered a simple truth. “As a successful organization, rather than relying on centralized AI capabilities to identify use cases, has allowed budget holders and domain managers to represent problems, veterinary tools, and lead rollouts.” In other words, those closest to the job know what they need. It's not a far-off AI committee.
Conclusion
Most AI companies fail because the technology is not so refined that it is not so refined and used and nurtured in the best possible way. Although some missteps are understandable, not being able to adapt quickly can mean missing the boat. According to a study by MIT, innovation is moving rapidly, with windows closed for developing durable moats through “building adaptive agents from feedback, usage and results.”
