The return on AI is going to firms that stopped piloting
This week’s evidence points one way: companies running AI as a production system are compounding margin gains, while the rest are still testing.
Return on AI: Weekly Digest – July 13, 2026
Fifteen stories on where operational AI is producing measurable returns, from government tax recovery to enterprise cost reduction, plus sustained coverage of the legal sector and new constraints on data center capacity.
Our take
The gap between AI leaders and everyone else is no longer a forecast. McKinsey’s data shows leaders have already grown EBITDA by 20% while most organizations remain in pilot mode. The operational stories explain why: document review that took two weeks now closes in 48 hours, legal teams are absorbing more work without adding headcount, and AT&T reports that routing each task to the right model cut its AI costs by 90%.
The common thread is that these firms made deployment decisions. A pilot that runs indefinitely costs money and produces a slide. A production system changes how work gets done and shows up in the P&L.
What to do about it
- Give every pilot a decision date: scale it, change it or stop it.
- Route work to the model that fits the task instead of defaulting to the most capable, and most expensive, one.
- If your plans depend on new AI capacity, build in time. Infrastructure constraints are stretching data center timelines by years.