Buried inside Gallup's State of the Global Workplace 2026 is a single comparison that says more about AI adoption than the last twelve consulting decks combined. Employees whose manager actively supports AI use are 8.7 times as likely to strongly agree that AI has transformed how work gets done in their organisation. The same group is 7.4 times as likely to say AI gives them more opportunities to do what they do best every day.
For context: across organisations that have implemented AI, only 12% of employees strongly agree that AI has actually transformed how the work gets done. That headline number gets quoted as evidence the AI revolution is overhyped. Gallup's contribution is to show that the 12% is not evenly distributed. It is concentrated among the teams whose manager is in the conversation. Where the manager has stepped out of it, the number collapses.
Why The Multiplier Is A Manager Number
Gallup's wider 2026 dataset is consistent on this point. Manager engagement has fallen from 31% in 2022 to 22% in 2025, a nine-point drop in three years, and the cliff in employee engagement tracks the cliff in manager engagement almost step for step. The same pattern now shows up in AI. Where the manager is engaged and openly using AI in front of the team, the team views AI as transformative at an 8.7x rate. Where the manager is disengaged, AI tends to land as another tool, another login, another reason to feel behind.
The supporting evidence is multi-sourced. Microsoft's Work Trend Index 2026 found only 26% of AI users say leadership is aligned on AI. Deloitte's 2026 Global Human Capital Trends found 80% of workers suspect colleagues use AI to appear more productive, and 34% say the culture itself inhibits AI goals. Three different research houses, three different methodologies, the same finding from different angles. AI does not roll itself out, managers do, and most are not yet equipped to.
What Most Organisations Are Trying
The default response to a low AI-transformation number is more enablement. New prompt libraries. A revised acceptable-use policy. A Copilot champions network. A town hall featuring a five-minute demo from the CTO.
None of those things meet the manager where the conversation actually happens, which is in the one-to-one when the team member says "I tried using ChatGPT for this and it gave me nonsense", or in the standup when someone asks whether they are supposed to disclose that the deck was AI-drafted, or in the appraisal when the manager has to decide whether the employee who is faster because of AI deserves more credit than the employee who is faster without it. These are unrehearsed, real-time, ethically loaded conversations. They are run on the manager's instinct, not their policy knowledge.
That is the gap the Gallup 8.7x speaks to. The teams where AI is landing as transformation are not the teams with the best policies. They are the teams whose manager has practised these conversations enough that the right instinct is the first one.
What Actually Moves The Number
Two findings from the behavioural-science literature explain why most AI-enablement programmes underperform, and what works instead.
The first is the retention curve. Roediger and Karpicke (2006), in Psychological Science, established that being tested on material lifts long-term retention by roughly 50% compared with re-reading the same content. Most manager AI training is the re-reading kind. A webinar, a deck, a policy PDF. By the time the corridor conversation happens, very little of what was taught is accessible, and the manager defaults to whatever instinct they had before the training.
The second is the first-reaction effect. Edmondson's 1999 work in Administrative Science Quarterly on psychological safety identifies the manager's first reaction as the strongest single predictor of whether the team will surface problems early. First reactions are motor behaviours. They are run before the conscious mind catches up. Applied to AI: whether the manager's first reaction to "I used Claude to draft this" is curiosity or suspicion determines whether the next person on the team admits the same thing, or hides it. The 80% of Deloitte workers suspecting their colleagues use AI to look more productive is, at root, a first-reaction problem.
In Sidestream's own academic behaviour-change work, building on research from UCL, Cambridge and Bocconi, participants who learned a communication skill through immersive role-play scored roughly 20% higher on observed behaviour than those who learned the same content through video or slide-show training. Self-rated confidence did not predict observed performance. A Dunning-Kruger pattern we designed out of subsequent studies by replacing self-reports with behavioural measurement.
What We Do About It
Our manager workshops and immersive simulations can be tuned for exactly the AI-shaped conversations that the average enablement programme skips. Small groups of managers work through realistic scenarios with professional actors, the team member who is overusing AI, the team member who is refusing it, the colleague returning to a role that has just been partly automated. The manager has to run the conversation live, with feedback, more than once. By the third rehearsal the new instinct is starting to feel automatic rather than effortful. That is the only point at which the 8.7x multiplier becomes available to your organisation.
The 8.7x multiplier is not a feature of the tool. It is a feature of the manager who has practised the conversation often enough to mean it.
The organisations that get AI ROI in the next two years will not be the ones with the biggest licence pool. They will be the ones whose middle layer has rehearsed the awkward conversations until the instinct is right. Read our piece on the 26% leadership alignment gap, or book a call to look at what manager rehearsal would look like in your organisation.
Book a free 30-minute diagnostic call → or read about our research-backed approach.