Workforce Trust
Whether the people asked to work alongside a system believe it will not be used against them.
Maturity
capability 4 / 5
deployment 3 / 5
absorption 2 / 5 ← sets the score
Bound by trust · trend flat.
Basis, source, vintage
Low and not improving. Deployment surveys consistently show adoption running ahead of any settled answer to the displacement question, and exposure estimates concentrate on task categories where that question is sharpest.
Source: ILO, Generative AI and Jobs (E-054); ILO Working Paper 140 (2025); US Census BTOS (2024–26) (2023–2026)
Where you have already met this: A hospital buys an AI triage tool. It works. Six months later most nurses are quietly running their own checks alongside it, because nobody answered what happens to the person whose judgement it replaces. The tool is deployed and the work has not changed.
If this moved: If this went from 2 to 4, the automation megatrends would steepen sharply — because every other enabler under them already scores 4 or 5. It is the shallowest link in the chain, so it is the only one where movement changes the gradient at all. That is unusual and worth noticing: most enablers, improved, change nothing.
What to watch for: Deployments that visibly remove the worst part of a job before they remove any headcount. Retraining with a named destination rather than a budget line. Worker representation in the deployment decision rather than in its aftermath.
Not to be confused with: Trust in the technology. People are usually fine with the tool; what they distrust is the intent behind installing it. That is a contract question, not a communications one.
What it is
Whether the people expected to work alongside a new system believe it will not be used against them. It covers trust in the tool's output and trust in the employer's intent, and the second usually decides the first.
Common misconception
That it is a communications problem. It is a contract problem: what happens to the person whose task the system now performs. Where that question has a credible answer, trust follows; where it does not, no amount of internal messaging substitutes.
How it works
- A system is deployed into work someone already does — Not into empty space. There is an existing task, an existing person doing it, and an existing understanding of what that person is for. The system arrives into that, not into a vacuum.
- The person infers intent from the deployment, not from the announcement — They read the signal: was this bought to make my work better, or to make me cheaper? The inference is made from who was consulted, what happened to the last team this was done to, and whether anyone answered the redundancy question out loud.
- Discretionary effort is withheld or given — This is the mechanism, and it is invisible in every deployment metric. Compliance is mandatory; effort is not. A trusted tool gets edge cases reported, prompts refined, colleagues trained. A distrusted one gets minimum compliance and a parallel manual process nobody mentions.
- The parallel process becomes permanent — Staff run their own checks alongside the system. Output is unchanged, cost has gone up, and the deployment is recorded internally as a success because the licences are active.
State of play
- US Census BTOS shows firm AI adoption rising continuously since 2023, while measured productivity effects remain small — the gap between deployment and effect is the observable.
- ILO exposure work finds clerical work the most affected category globally, and women's employment 2–3× more exposed than men's in high-income countries (ILO WP140, 2025: 9.6% vs 3.5%).
- No large employer has published a binding commitment on what happens to workers whose tasks are automated. The absence is itself the datum.
- Trust in employers tracks separately from trust in technology, and moves slower — OECD Trust Survey rounds 2021, 2023, 2024.
Why it matters
It determines whether a deployed capability is actually used, worked around, or quietly sabotaged. A tool that staff believe is a redundancy exercise gets minimum compliance and no discretionary effort — and discretionary effort is where most of the value in a deployment sits. This is the enabler that most often turns out to be binding for automation megatrends, which is why the constraint on the AI ramp is rarely compute.
- It is the binding constraint on cognitive-work automation. Every technological enabler under that megatrend scores 4 or 5; this scores 2, and a chain takes its minimum.
- It explains the productivity paradox without invoking measurement error: the capability is deployed and the work has not been redesigned, because redesigning it requires the cooperation of people who have not been told what happens to them.
- It is the cheapest enabler to move and the least funded. No ribbon is cut at workforce trust.
What holds it back
- No credible answer to the displacement question — Retraining budgets without a named destination read as severance with extra steps. The question is contractual, and it is answered with communications.
- Deployment decisions made without the people affected — Consultation after procurement is not consultation. By then the tool, the vendor and the business case are fixed.
- Measurement that cannot see the failure — Licences issued, logins, prompts per user — all rise in a distrusted deployment. Nothing in the standard dashboard detects a parallel manual process.
Terms
- Discretionary effort
- The work someone does beyond what is required and observable. It is where most of the value in a deployment sits, and it cannot be mandated.
- Shadow process
- An unofficial manual workflow run alongside an official automated one. Usually invisible to management and to any metric.
- Task exposure
- The share of a job's tasks a system could perform. Distinct from job loss — most exposure produces changed work, not removed work.
- Absorption
- In this model, whether people and institutions actually take a capability up. One of the three readiness dimensions, and usually the shallowest.
Steepens
- Automation of Cognitive Work — large · ○ shared candidate constraint (with 4 others)
- Hyper-automation of Industrial Production — large · ○ shared candidate constraint (with 2 others)
- The Application Gap — large · ○ shared candidate constraint (with 2 others)
- Precision Everything — moderate · ○ shared candidate constraint (with 2 others)
- The Care Economy — moderate
What would change its gradient
- A credible, funded answer to what happens to the displaced person — retraining with a destination, not retraining as a gesture
- Worker representation in deployment decisions rather than in their aftermath
- Deployments that visibly remove the worst part of a job before they remove any part of the headcount
Shares megatrends with
- Managerial Capability
- AI & Machine Learning
- Automation
- Digital Transformation
- Digitalization
- Robotics
In the book
Why the Future Arrives Unevenly — Ch. 3 — Why No Technology Is a Megatrend · Ch. 5 — The Optimisation of Everything · Ch. 15 — Society Out of Sync · Ch. 16 — An Economy Without Capital · Ch. 18 — The Doom Loop
Sources
- Low and not improving. Deployment surveys consistently show adoption running ahead of any settled answer to the displacement question, and exposure estimates concentrate on task categories where that question is sharpest. Source: ILO, Generative AI and Jobs (E-054); ILO Working Paper 140 (2025); US Census BTOS (2024–26) (2023–2026).
- Generative AI and Jobs: A Global Analysis of Potential Effects on Job Quantity and Quality — ILO, 2023
- Business Trends and Outlook Survey — US Census Bureau, 2026