The pilot worked. Why hasn’t the profit moved?

Treat the research as a question, not a verdict.

The preliminary July 2025 Project NANDA report, The GenAI Divide, described a wide gap between enterprise GenAI experimentation and measurable profit-and-loss impact. Its widely repeated 95% headline should be read with its limitations: the report uses interview and survey evidence, category samples vary, and success definitions differ. [1] It is not a universal probability that your project will fail, or a current audit of every company’s AI returns. Its useful challenge is simpler: what evidence connects your implementation to value?

Write the bridge from task to economics.

A faster proposal process may matter because it reduces paid overtime, releases capacity for additional qualified bids, or improves response speed enough to affect wins. Each route needs different evidence. If the same team serves the same customers at the same cost, hours saved may be a productivity gain without becoming cash savings. Distinguish capacity released, capacity used, and financial benefit realized. The business owner should be able to describe that bridge before a larger rollout is funded.

Count the work around the model.

The relevant cost includes integration, data preparation, usage, human review, training, support, and the cost of correcting failures. Compare a bounded workflow against its baseline using a consistent quality standard. If a campaign changes at the same time as pricing and sales coverage, a revenue increase cannot simply be assigned to AI. Use a credible comparison where possible and state the limits of attribution. A modest, defensible result is a stronger basis for investment than a large claim assembled from unrelated measures.

Make stopping a legitimate outcome.

Agree what evidence would justify scaling, revising, or stopping. A pilot should answer a decision, not establish a permanent subscription by default. Our suggested review brings together the workflow owner, finance, and technology leadership to assess adoption, output quality, full cost, and economic effect. This is Robuzt’s advisory approach, not a conclusion proven by the NANDA study. The final question is whether this is the best next use of capital, compared with simpler automation, process repair, or another growth priority.