If you are trying to figure out how to use generative AI in your office and still haven’t cracked it yet, rest easy. You are not alone.American companies have poured an -estimated $40 billion into generative AI, but 95% of those efforts are producing no measurable return, according to a new MIT report, The GenAI Divide: State of AI in Business 2025.
Why it matters: The findings underscore a widening gap: a small group of organizations are using AI to generate millions in value, while most remain stuck with stalled pilots, skeptical users, and little impact on the bottom line.
By the numbers:
- 95% of enterprise AI pilots fail to scale.
- 5% of organizations have crossed the “GenAI Divide,” embedding tools that adapt to workflows and deliver P&L impact.
- 70% of AI budgets go to sales and marketing tools, even though the highest ROI is in back-office automation like accounts payable, document processing, and risk checks.
- 90% of employees say they use personal tools like ChatGPT at work, compared to only 40% of companies that provide official subscriptions.
Driving the failure rate: The core issue is what MIT researchers call the “learning gap.” Most AI systems don’t retain feedback, adapt to context, or improve over time. That makes them great for brainstorming or summaries, but unreliable for mission-critical workflows.
Executives echoed this frustration: “We’ve seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects,” one CIO told researchers.
Where AI is working: The 5% of companies finding success share common traits:
- Narrow scope: Starting with one pain point, such as contract review or call summarization, then scaling.
- External partnerships: Purchased tools succeed twice as often as internally built systems.
- Learning systems: The most effective tools retain memory, adapt, and improve over time.
Between the lines: The report highlights a booming “shadow AI economy” inside companies. Employees are relying on personal ChatGPT or Claude accounts to automate their work, even when official AI projects stall. In many cases, this underground use is delivering more ROI than formal enterprise deployments.
What’s next: The next frontier is agentic AI, which are tools that can learn, remember, and act autonomously. MIT researchers warn the window is closing: within the next 18 months, large enterprises will lock in vendor relationships that will be hard to unwind.
“Once we’ve invested time in training a system to understand our workflows, the switching costs become prohibitive,” said a CIO at a $5 billion financial services firm.
