Narrow AI deployments are beating broad rollouts
The week’s clearest wins came from small, specific use cases. The broad rollouts mostly produced costs, supervision time and work people can’t defend.
Return on AI: Weekly Digest – July 20, 2026
Seventeen stories split between where AI is underdelivering, where it is paying off, and how adoption and risk are shifting, including new findings on supervision time, output quality and liability coverage for AI agents.
Our take
Forced, broad adoption is showing up as cost without measurable gain. Workers now spend about seven hours a week supervising AI tools, and most AI users admit to shipping work they can’t fully defend. Those are real costs, and few companies count them.
The wins were narrow and specific. A distributor added $700,000 in revenue by applying AI to part of its pricing. Dollar Shave Club produced a top-performing campaign for $400. Wells Fargo drew thousands of adviser prompts at launch by targeting one defined workflow. Each started with a single job to be done and a way to measure it.
What to do about it
- Start with one recurring workflow and one measurable outcome, then expand from what works.
- Count supervision and rework time as part of the cost of every AI tool.
- Before deploying AI agents, check whether your liability coverage still applies. Many older policies don’t.