:max_bytes(150000):strip_icc():format(jpeg)/GettyImages-2166183185-166fe9c73b4d4ea19d96f361dba1f332.jpg)
Key Takeaways
- MIT research reveals that 95% of generative AI pilots fail to deliver measurable return on investment.
- This is creating what experts call a “GenAI Divide” between companies stuck with basic chatbots and those using 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 a company’s specific needs.
- The window to improve offerings, deliver on demands, and compete in this high-growth space is narrowing, experts say.
Companies are investing billions in AI tools like ChatGPT, aiming to reduce costs and boost productivity. But it’s not working.
A new MIT study found that 95% of businesses using AI aren’t making their money back. Despite all the hype about AI changing everything, most companies are stuck with expensive tools that don’t deliver.
The problem isn’t the technology itself—it’s how companies are using it. Here’s what’s going wrong and why so many AI projects fail.
Learning Gap
Popular tools like ChatGPT and Copilot have been celebrated for improving productivity. However, when it comes to boosting corporate profits, they’re less successful, mainly because they don’t learn from or adapt well to a company’s needs.
ChatGPT, the report found, forgets context and doesn’t learn or evolve, meaning it can’t be used for mission-critical work that replaces humans. “It’s excellent for brainstorming and first drafts, but it doesn’t retain knowledge of client preferences or learn from previous edits,” a lawyer told the researchers. “It repeats the same mistakes and requires extensive context input for each session. For high-stakes work, I need 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. Yet companies are often chasing glamorous projects that sound good in presentations but rarely pay off, the study suggests.
Going Solo
The reports suggests many companies are getting ripped off by spending millions on proprietary generative AI systems supposedly adapted to their in-house systems. Enterprises, it says, fare better when partnering with mature, well-supported third-party vendors.
Only 5% of custom enterprise AI tools reached production, with the majority failing because of “brittle workflows, [a] lack of contextual learning, and misalignment with day-to-day operations.” That included chatbots, a popular area of investment, which often were said to fail because of a lack of memory and customization.
With a success rate of 67%, teaming up with specialized third-party vendors turned out to be a better play. However, that, too, presented challenges. Executives complained about being bombarded with proposals and noted that many of the solutions were “brittle, overengineered, or misaligned with actual workflows” and essentially “science projects.”
While companies are investing heavily in custom products, their employees vote with their keyboards. As the report notes, “While only 40% of companies say they purchased an official LLM subscription, workers from over 90% of the companies we surveyed reported regular use of personal AI tools for work tasks.”
Companies Are Misspending on AI
When it comes to AI, the MIT researchers found that the money often ends up where it doesn’t ‘t do the most good.
According to the report, most companies poured their AI budgets into sales and marketing. But the real money-savers were the boring tasks nobody talks about: automating paperwork, processing invoices, and handling routine administrative work.
That wasn’t the only mistake made. Companies also tried to do too many things at once instead of fixing one problem really well. They relied on special “innovation labs” and AI teams instead of letting the people who actually do the work decide what tools they need.
The report found a simple truth: “Rather than relying on a centralized AI function to identify use cases, successful organizations allowed budget holders and domain managers to surface problems, vet tools, and lead rollouts.” In other words, the people closest to the work know what they need—not some distant AI committee.
The Bottom Line
Most AI companies are failing because the technology is not yet sophisticated enough for widespread business use and isn’t being utilized and nurtured by them in the best possible way. Some missteps are understandable, but failing to adapt quickly could mean missing the boat. Innovation, according to the MIT study, is moving fast, and the window to develop a durable moat by “building adaptive agents that learn from feedback, usage, and outcomes” is closing.
