Copilot was supposed to shrink the calendar. The calendars of every executive I coach got longer.
The sponsors I talk to all report the same drift. They expected the dashboard to show meetings falling and reclaimed hours returning. It shows the opposite. More meetings. More email. More time spent with other humans. They squint and ask what broke.
Nothing broke.
The slide that bent the wrong way
A COO at an industrial-services firm in Rotterdam, two weeks ago, nine months into a staged Copilot rollout. Her six-month review deck opens with the "meetings reclaimed" chart. The line goes up. She slides it across the table and asks whether we should pause the program.
I ask what else is in the deck. Adoption, high. Sentiment, positive. Help desk tickets, flat. One chart is embarrassing her, and she wants to kill a program that is otherwise working.
I tell her the chart is right. The rollout did something real. It just was never what the business case promised.
What Bocconi just measured
Giuseppe Soda and his colleagues at Bocconi pulled twelve months of Microsoft Workplace Analytics data. One business unit of a large consumer-products multinational, mid-staged Copilot rollout. 11,868 knowledge workers. 1,992 adopters. 9,876 never-adopters. Two independent identification strategies: a Callaway-Sant'Anna staggered difference-in-differences, and a two-way fixed-effects IV with cohort-level exposure as the instrument. Kleibergen-Paap F of 80.43, well past the Stock-Yogo 10 percent threshold of 16.38.
Adopters' total collaboration time went up about 0.9 hours per month (p<0.001). Generated collaboration hours up 1.4 (p=0.004). Out-degree, the number of distinct colleagues each user interacted with, up by 2.179 (p=0.037). Local clustering down 0.009 (p=0.009). The strong-tie share went up by 0.011 (p=0.016) at the same time.
"Deepening reliance on already trusted colleagues."
The user is reaching further out on the periphery AND leaning harder on a smaller, trusted inner core. Both signs. Simultaneously. Invisible to anyone watching only the meetings line.
What the ROI slide missed
The business case asked whether Copilot saved meetings. The question that actually mattered was whether Copilot changed who the user talks to, and how often. If your six-month review deck does not include out-degree, local clustering and strong-tie share, you are measuring a rollout you cannot see.
The second effect is the one nobody priced. Broadening the periphery widens reach. Concentrating the core speeds execution inside it. Both signs are real, and both change the shape of the organization under the rollout. The trusted core gets tighter, which quietly slows the integration of weak-tie information from outside it.
The COO is now watching three new metrics. The "meetings reclaimed" chart is still on slide seven. We agreed to retitle it.