When a new MIT study lands with the claim that 95% of enterprise generative AI pilots aren’t delivering measurable ROI, it’s worth paying attention. The report, The GenAI Divide: State of AI in Business 2025, is based on over 300 corporate AI implementations, 350 employee surveys, and 150 industry leader interviews.

The headline isn’t that AI doesn’t work. It’s that most companies are struggling to apply it where it matters.

Three Fault Lines Exposed

  1. The Learning Gap AI pilots often fail not because the models are weak, but because they don’t learn into the enterprise. Tools that forget context or don’t integrate into workflows create a gap between capability and utility. Employees may experiment with ChatGPT or Claude, but those gains rarely transfer into the company’s official systems.

  2. Misaligned Budgets The report found more than half of enterprise AI budgets went into sales and marketing pilots. These are visible, easy to launch, but rarely deliver sustained ROI. The real payback sits in back-office automation: invoice matching, contract processing, compliance checks. Cutting external agency spend and streamlining operations showed the clearest returns.

  3. Brittle Tools Many enterprise pilots relied on rigid, single-purpose tools. They looked good in demos but cracked under the complexity of live enterprise workflows. Flexibility, memory, and integration—not novelty—are what separate the 5% of successes from the 95% of stalled projects.

The ROI Picture

Where pilots did succeed, the value showed up less in flashy revenue growth and more in cost avoidance:

  • Reduced agency/BPO costs (some cases cut by 30%).
  • Cycle time reductions in finance and operations.
  • Lower error and exception rates in compliance workflows.

The study also found internal builds failed about twice as often as external solutions. And while official deployments lagged, “shadow AI” use by employees on personal accounts was widespread.

Why This Matters

The lesson is clear: the problem isn’t generative AI technology itself. It’s how organizations are choosing to apply it, where they’re placing their bets, and whether they’re willing to re-engineer processes to let AI actually take root.

For independent consultants and SMEs—the audience Imbila speaks to—this is a crucial insight. Large enterprises can afford pilot theatre. Smaller, faster players can’t. The opportunity lies in closing the learning gap quickly, aligning budgets with real operational leverage, and choosing tools that adapt rather than break.

Sources

  • MIT, The GenAI Divide: State of AI in Business 2025 PDF link
  • Fortune: “Human skills are at a premium again now that big companies are backpedaling on error-prone AI”
  • McKinsey, The State of AI: 2024 Global Survey