The number that should stop the spend

MIT's Project NANDA studied enterprise generative-AI pilots across dozens of executive interviews, a survey of 153 leaders, and 300 public deployments. The finding: about 95% produced no measurable return to the P&L. Only roughly 5% drove rapid revenue impact. McKinsey's 2025 State of AI, surveying nearly 2,000 respondents, reproduces the same shape — around 80% of organizations use generative AI, yet only about 5.5% report meaningful EBIT impact. Two independent studies, one conclusion: adoption is nearly universal, results are rare.

It's not the model — it's the wrapper

The instinct is to blame the tools. The data says otherwise: the winners and losers use the same models. What separates the 5% is everything around the model — the judgment to point it at a problem worth solving, the process to wire it into a real workflow instead of a demo, and the discipline to measure output that actually moves revenue. NANDA's own thesis is that the failures are integration and approach problems, not capability problems. The AI is a commodity. The scarce input is the judgment that makes it pay.

Buy the judgment, not just the subscription

For a $1M–$10M ARR company, the lesson is direct. Stacking AI tools onto an undiagnosed funnel buys you faster production of the wrong thing. The teams getting return put senior judgment first, use AI as a scale layer beneath it, and grade every play against pipeline and payback. That is exactly the operator-led model: one experienced operator deciding what to build and what to ignore, with an agent fleet handling the volume — so strategy ships as execution, not as another dashboard nobody acts on.

How gRO solves it

  • Judgment first. Fifteen years of operator judgment decides what to build and what to ignore — the scarce 5%.
  • AI as the scale layer. An agent fleet handles production volume, so strategy ships as execution instead of slideware.
  • Graded on revenue. Every play is measured against pipeline and payback, not activity or output count.

FAQ

Does this mean AI doesn't work for marketing?

No — it means AI without judgment and process doesn't produce return. The ~5% that captured value used the same models as everyone else; the difference was integration, workflow and measurement. AI is a scale layer, not a strategy.

Is the 95% figure about marketing specifically?

No. The MIT figure covers enterprise generative-AI pilots broadly, not marketing alone. It's cited here as evidence that adoption without judgment rarely pays — a pattern that holds in marketing as much as anywhere.

Sources cited in this analysis

  • The GenAI Divide: State of AI in Business 2025 — MIT Project NANDA (Aug 2025)
  • The State of AI 2025 — McKinsey (QuantumBlack), n=1,933
  • 2025 Career Outlook & B2B Content Benchmarks — Content Marketing Institute