I used to tell execs the AI rollout would reshape their teams.

I was half right. The reshape only happens in one half of the teams: the ones that formed after the tool landed. The teams that pre-existed the tool absorb it and keep working the way they always did.

The dashboard that explained nothing

Last month I sat with a COO at a mid-market B2B SaaS firm. Six months into a company-wide Copilot rollout. He opened the vendor's dashboard. 78 percent monthly active users. 42 minutes saved per user per week. Satisfaction at 7.3.

He went quiet. Then: "If this is true, where did the 42 minutes go? Nothing has reorganized."

He had expected the tool to reshape how his pods worked. It had not. The pods were still shaped the way they were in May. The hours came back as more work of the same kind.

I asked him which pods had formed in the last year and which had been together three years or more. He had not sorted the data that way. Nobody does.

What the research just put numbers on

Stefano Bianchini and Georgiana-Ioana Coroama at BETA Strasbourg ran an event-study on 117,211 PLOS ONE papers from 2018 to 2025. They used CRediT contribution statements to see who did what on each paper. The break is 2022, when ChatGPT shipped.

By 2025, participation in the cognitive and writing-heavy roles sits below pre-trend. Ideation down 2.6 points. Writing down 2.7. Execution down 0.9. Coordination down 1.4. Average tasks per author drop by 0.08. Pairwise task-overlap between co-authors drops by 0.018. All against a rising pre-trend.

Then they split the sample.

Established teams: no significant departure from pre-trend in any domain.

The reshape sits entirely in the newly formed teams. 102,228 of them. Writing 3 points below trend. Ideation 2.7 below. Execution 1 point below. Coordination 1.6 below. Established teams, 14,983 of them, keep working the way they always did.

One caveat the authors flag. The unadjusted raw means for established teams actually rise on Ideation (3.09 to 3.42) and Writing (3.59 to 4.18). The event-study controls carry the inertia story. Read it with that in mind.

What a COO should do with this

Your aggregate dashboard hides a Simpson's paradox. If the saved-hours number is modestly positive, the new teams are moving the needle and the old ones are doing nothing. Average them and the signal softens.

The lever is team formation. Everything else moves the surface.

If you are standing up a new pod, a new region, a new product line, form it around the tool and the specialization will follow. If you leave every existing team intact, expect absorption.

Reorg precedes tool benefit. The order matters.