HALDEMAN Notes · 01 ·
AI activity is easy to count. Changed work is harder.
Logins measure access. Changed work predicts results.
Every AI update counts something. Licenses. Active users. Training hours. All real. All access.
The number that predicts a result is rarer: how many workflows now run differently, with an owner.
The evidence agrees. BCG found employees at companies that redesigned their workflows, or built new products and business models, were 24 points likelier to report measurable business improvement: 67%, against 43% at companies that only deployed AI.
Changed work counts when quality holds. If saved time comes back as rework, the effort moved. It didn’t shrink.
Constructed example
At Orrin & Vale, weekly use reaches 41 percent of seats, yet no AI-assisted output in Content has a designed check. Two published errors in Q3 trace to unchecked drafts, and the use number alone misses them.
So: before your next renewal, ask for the second number, with a quality line beside it. Which workflows changed? Who owns each one? What check shows the output is still good? Then set the date you decide to keep, fix, or stop.
Sources checked October 2026
BCG, AI at Work: Why Strategy Matters More Than Tools, 2026. Read the original