Overview

• A new MIT NANDA analysis reports only about 5% of AI pilot programs move beyond incubation into production.
• The study says most projects falter because of enterprise adoption and execution issues rather than deficiencies in model quality.
• Fortune-cited reporting notes resource misallocation, with more than half of some generative AI budgets flowing to marketing tools instead of automation tied to efficiency gains.
• The small share that scale follow common playbooks: define a problem with business upside, set KPIs linked to revenue or cost savings, design for user uptake, and secure cross‑functional champions.
• Following the paper’s release, U.S. tech stocks shed roughly $1 trillion over four days, as a disappointing GPT‑5 rollout and OpenAI’s reversion to older models like GPT‑4o—now available only to paying users—deepened skepticism.