Research opportunity · Organizational absorption: Why micro gains do not become macro outcomes
Does deploying AI to whole teams change the work that individual licences did not?
Question
In real organizations, when generative AI is deployed to whole teams or units, with and without a redesign of the team's workflow, do coordination-dependent work (meetings, handoffs, jointly written documents) and the unit's measured output change, compared with units where only some individuals have the tool?
Causal mechanism not isolatedFunded team projectNeeds a budget (medium or higher)Partly addressed
Why it matters
The largest randomized trial of an integrated assistant gave licences to individuals across 66 firms and found less time on email but no detectable change in meeting time or in the quantity or composition of tasks. Interview studies in Chinese newsrooms, German firms and a large software company describe the same pattern, with AI used as a personal assistant and team processes unchanged. If deployment to whole units, with a redesigned workflow, moves output where individual licences did not, the organizational-bottleneck position (P4) is right and the remedy is organizational. If it does not, the low-ceiling position (P2) gains ground.
The debate lists this among what we still do not know: Does deploying AI to whole teams, with redesigned workflows, change coordination-dependent work where individual provision did not?
Evidence that motivates it
- In a six-month randomized trial of an integrated AI assistant across 66 firms and 7,137 knowledge workers, workers who used it spent about two fewer hours a week on email, but the authors detect no shift in the quantity or composition of workers' tasks from individual-level provision, and no change in time spent in meetings.
Individual-level randomization across 66 firms: users spent about two fewer hours a week on email, with no detectable change in meeting time or task composition; the authors suggest that coordinated work may change only when colleagues adopt together. - In 27 interviews in regional Chinese newsrooms, journalists used generative AI frequently in their own tasks, but its use stayed disconnected from team workflows, hindered by structural barriers and a reluctance to share practices, and no formal mechanism coordinated individual use.
Journalists in regional Chinese newsrooms used generative AI often in their own tasks, but its use stayed disconnected from team workflows and no formal mechanism coordinated it. - In a case study of three German organizations, generative AI was used mainly as a personal assistant rather than changing team processes, the effort of validating its output partly offset efficiency gains, and policies written in response to regulation without regard to actual use led to gaps between policy and practice and to shadow IT.
In three German organizations generative AI was used mainly as a personal assistant rather than to change team processes. - In a preregistered field experiment with 791 professionals at Procter & Gamble, individuals working with AI matched the performance of teams working without it on product-innovation challenges, and AI helped bridge the gap between R&D and commercial staff.
On product-innovation challenges at Procter & Gamble, individuals with AI matched teams without it; this manipulates team composition on a bounded task, not ongoing deployment to existing units.
A possible design
Cluster-randomized trial inside one or more organizations with many comparable units (branches, case teams, product squads). Randomize units to three arms: no change; team-wide licences; and team-wide licences plus a facilitated redesign of the workflow (agreed uses, shared prompts, changed handoffs and review steps). Add a partial-saturation arm in which only some members of a unit get licences, to estimate spillovers. Measure coordination from collaboration telemetry (meeting time, co-authored documents, handoffs) and output from the unit's own records (cases closed, releases, sales), with a pre-period and at least six months of follow-up.
Data needed
An employer willing to randomize by unit; consent to collaboration telemetry; unit-level output measures that exist before the trial; enough units (tens, not a handful) for cluster-level statistical power.
Feasibility
Funded team project. Medium; licences, facilitation of the redesign and data engineering for telemetry, and access depends on an employer partner. Vendors and employers already run such trials on their own products, so an independent team needs a pre-registered analysis plan and data access agreed before results are known. Contamination between units that share an office must be measured, not assumed away.
Has this been done?
Partly done already: related work covers some of this question (first-pass screen of 15 October 2026; not yet checked in depth).
One working paper reportedly randomizes a generative-AI assistant at the team level inside a real firm (42 teams of 316 employees) and measures collaboration-network centrality and project performance, but not meeting time, co-authoring or handoffs. No cluster-randomized trial was found that deploys AI to whole units in a real organization and measures coordination-dependent work from telemetry, and none that crosses team-wide provision with a workflow-redesign arm. The largest trial randomizes individuals; team studies found are bounded workshop tasks or laboratory experiments; a registered two-by-two trial with freelance pairs has no results yet.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 15 October 2026 (review of PR #77): verdicts, query counts and 13 abstract-read labels checked; label slips corrected. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 5 of the 6 nearest works below were judged from an abstract, summary or search listing only, and 1 has details that could not be confirmed; each is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search (7 queries) and arXiv and NBER abstract pages on the date shown; SSRN, the AEA registry and Microsoft Research were not searched directly, so unpublished cluster-randomized vendor deployments may be missing. The two INSEAD papers and the registered trial were seen only through search listings. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Dillon, Jaffe, Immorlica & Stanton, "Shifting Work Patterns with Generative AI," arXiv:2504.11436 (v4) / NBER Working Paper 33795 (2025)
Full text read for the LD-03 debate (LD03-SL-C06). Randomized at the individual level across 66 firms and 7,137 workers; less email time, no detectable change in meeting time or task composition. Three of the four authors are at Microsoft Research, whose product was tested. - "The Impact of Generative AI Adoption on Organizational Networks: Evidence From A Field Experiment," INSEAD Working Paper (about 2025–2026; authors not confirmed) Judged from an abstract, summary or search listing onlySome details not confirmed
Judged from search listings; not read directly. Reportedly 42 teams of 316 employees at a European technology-services firm randomized to an assistant customized with the organization's knowledge or to business as usual; after three months treated employees were more central in collaboration and knowledge networks. The nearest design, but no meeting, co-authoring or handoff measures were reported in the listings. - Dell'Acqua, Ayoubi, Lifshitz, Sadun, Mollick, Mollick, Han, Goldman, Nair, Taub & Lakhani, "The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise," Organization Science 37(4): 1217–1242 (2026); NBER Working Paper 33641 Judged from an abstract, summary or search listing only
Abstract read for the LD-03 debate (LD03-SL-C16). 791 Procter & Gamble professionals randomized to AI or no AI, alone or in pairs, on product-innovation challenges. Team-composition effects on a bounded task, not deployment to existing units over time. - Li, Zhou & Mikel-Hong, "Generative AI Enhances Team Performance and Reduces Need for Traditional Teams," arXiv:2405.17924 (2024) Judged from an abstract, summary or search listing only
Abstract read. Laboratory experiment with 435 participants in 122 teams: AI-augmented teams outperformed, and AI used centrally by a few members beat distributed use. A laboratory, not an organization. - Kim, Kim & Koning, "Mapping AI into Production: A Field Experiment on Firm Performance," INSEAD Working Paper (March 2026) Judged from an abstract, summary or search listing only
Judged from search listings; not read directly. Reportedly 515 startups randomized at the firm level to information on how other firms reorganized production around AI, with the same tools for all; treated firms found more use cases. A redesign arm at the firm level, with no team-coordination measures. - De Stefano & Oseguera, "Generative AI and the Restructuring of Workplace Collaboration," AEA RCT Registry trial 18958 (registered June 2026) Judged from an abstract, summary or search listing only
Judged from search listings; not read directly. A registered two-by-two design (individual or team, with or without AI) run with freelance pairs on Upwork. No results found.
Search queries used
cluster randomized trial generative AI teams organization coordination Copilot team-level deployment"cybernetic teammate" Dell'Acqua Procter Gamble generative AI teams field experimentfield experiment generative AI randomized at team level knowledge workers team output meetings collaboration 2026generative AI peer adoption spillovers coworkers team share of colleagues using AI productivity firm natural experimentINSEAD working paper randomized 42 teams generative AI assistant collaboration networks European technology services firmgenerative AI rollout workflow redesign whole department field experiment firm process redesign output randomized units"Mapping AI into Production" field experiment firm performance
Provenance
RO-LD03-01 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-03 debate on 2026-10-15. Checked by an independent AI reviewer with fresh context (not a person), 15 October 2026 (review of PR #77): verdicts, query counts and 13 abstract-read labels checked; label slips corrected. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.