Why Policymakers Can't Wait for Employment Statistics
Employment statistics are a lagging indicator. By the time redundancy figures appear in quarterly reports, the damage is already embedded: workers have left the labour force, skills have atrophied, and the institutional response is playing catch-up against a structural shift that started 18 months earlier.
The new Workforce Resilience Index is designed to give the Isle of Man's policymakers something they currently lack: a leading indicator of AI-driven labour market stress, updated weekly, with auto-generated insights and concrete intervention recommendations attached to every data point.
This isn't an academic exercise. The Isle of Man has 43,000 workers, no inter-regional migration safety valve, and an employer base concentrated in sectors sitting directly in AI's crosshairs. When displacement arrives, it arrives everywhere at once.
The Research That Demands Action
Two bodies of academic work converge on the same finding, and both carry direct implications for small-island economies.
The China Shock Lesson
David Autor, David Dorn, and Gordon Hanson documented that the surge in Chinese manufacturing imports from 2000-2012 caused damage persisting nearly two decades in affected US communities. The critical finding wasn't that trade liberalisation was harmful in aggregate — it wasn't. It was that geographically concentrated, occupationally specialised labour markets absorbed a disproportionate share of the costs while receiving almost none of the benefits.
Workers in affected communities didn't retrain. They didn't migrate. They didn't find equivalent work. They left the labour force permanently, and the communities around them suffered cascading effects: reduced tax revenue, declining public services, social deterioration.
The Isle of Man is, by definition, the ultimate concentrated market. Every worker lives within 20 miles of every other worker. There is no commuting to an unaffected region. The financial services sector alone represents a substantial share of high-skilled employment. This is precisely the structural profile that the China Shock literature identifies as most vulnerable.
The Adoption Speed Finding
Eduardo Levy Yeyati's formal model adds the critical timing dimension. His key finding: the difference between 3% and 15% permanent labour force exit depends not on how much gets automated, but on how fast firms adopt. Social welfare is maximised at a significantly lower adoption speed than markets naturally choose.
His kappa ratio — displacement rate divided by reallocation rate — provides the mechanism. When kappa stays below 0.36, retraining systems absorb displacement comfortably. Between 0.36 and 0.65, the system is under strain but recoverable. Above 0.65, permanent exits accelerate and the damage compounds.
The current IoM kappa of 0.37 has just crossed the safe threshold into the warning zone. This is exactly the moment when intervention is most effective — and most commonly missed.
What We Built — And Why Each Section Matters for Policy
The Composite Index (0-100)
Each occupation group receives a score from four equally weighted dimensions:
- Automation Exposure (inverted) — what share of tasks can current AI fully automate? Uses Anthropic's task-level scoring, not occupation-level estimates.
- Occupational Diversity — Shannon entropy of sub-occupations. The China Shock lesson: concentrated groups break harder.
- Skills Transferability — augmentation share. Workers whose skills remain valuable in AI-augmented roles have more options.
- Salary Buffer — financial runway during retraining. An empirical factor from the China Shock literature, not a value judgement.
Policy relevance: The composite score identifies which occupation groups need intervention now versus which can absorb change autonomously. A group scoring 35 needs active retraining programmes. A group scoring 70 needs monitoring but not intervention.
Auto-Generated Insights
The dashboard now generates findings directly from the data — not editorial opinion. Each insight carries a severity level (critical/warning/info), affected worker counts, and a specific recommendation.
Policy relevance: Instead of reading charts and inferring conclusions, policymakers see actionable findings: "Elementary occupations (8,200 workers) have composite resilience below 40 — immediate retraining pathway design needed" or "Professional services employer concentration flagged — top 3 employers hold 67% of vacancies."
These insights update automatically every time the pipeline runs. No analyst bottleneck. No quarterly report cycle.
Early Action Interventions
Priority-ranked interventions derived directly from the resilience analysis, each with:
- Priority level — immediate / short-term / medium-term
- Target groups — which SOC categories need this intervention
- Estimated workers affected — scale of the response needed
- KPI metric — how to measure whether the intervention is working
Policy relevance: This is the bridge between analysis and action. A policymaker can take the "immediate" actions list to a Treasury briefing with worker counts and measurable outcomes already attached. No translation layer needed between the data team and the policy team.
Scenario Modelling
Three 10-year projections using logistic displacement curves calibrated to the current kappa:
- Baseline — current adoption speed continues
- Accelerated — kappa doubles (faster AI adoption, e.g. if a major employer restructures)
- Managed Transition — kappa halves (active retraining intervention achieves its target)
Each scenario shows projected workers displaced, income impact, reallocation rate, and time to equilibrium.
Policy relevance: This is the fiscal argument. When Treasury asks "what's the cost of inaction versus intervention?", the scenario comparison provides the answer in pounds and worker-years. The managed scenario typically shows 40-60% reduction in cumulative displacement compared to baseline — that's the return on investment for retraining infrastructure.
Risk Matrix
A scatter plot of every SOC major group on two axes: automation exposure risk (x) versus transition readiness (y). Bubble size shows worker count.
Policy relevance: The top-right quadrant — high exposure, low readiness — is the emergency zone. Groups here need immediate attention. The bottom-left — low exposure, high readiness — can be safely deprioritised. This visual alone can structure an entire policy prioritisation discussion in under 60 seconds.
Sector Transition Pathways
For at-risk groups (composite < 50), the dashboard identifies the most viable lateral moves based on skills overlap, salary proximity, and feasibility.
Policy relevance: Retraining programmes fail when they're designed in the abstract. "Retrain admin workers" is not a policy — "retrain admin workers for associate professional roles, 62% skills overlap, +£4k salary uplift, 3,200 eligible workers" is. These pathways give programme designers specific from-to targets with realistic expectations.
Adoption Speed Gauge
The weekly kappa tracking provides the only real-time signal of whether the island's AI transition is within safe parameters.
Policy relevance: This is the smoke alarm. A kappa crossing 0.36 (which just happened) should trigger the first-tier response: accelerate existing retraining programmes, increase Job Centre engagement with high-risk employers, commission a skills audit. A kappa approaching 0.65 should trigger second-tier: emergency retraining funding, employer consultation on adoption pacing, social safety net review.
The Policy Window Is Open Now
The critical insight from both the China Shock and Levy Yeyati literatures is that intervention timing determines outcomes more than intervention scale. Modest retraining programmes deployed before displacement peaks outperform massive emergency responses deployed after.
The Isle of Man's kappa has just crossed 0.36. The displacement-to-reallocation ratio is elevated but manageable. The occupation groups most at risk are identifiable. The transition pathways are mappable. The fiscal cost of inaction is calculable.
This is the window. The dashboard provides the evidence base. The question for policymakers is whether to use it.
Data Sources
The index combines four data systems updating at different cadences:
- Census 2021 (static) — 43,000+ working-age respondents classified by SOC4 occupation
- Anthropic Economic Index (periodic) — task-level automation and augmentation scores
- ONS ASHE 2025 (annual) — UK median salary benchmarks by occupation
- IoM Job Centre (daily) — live vacancy data with AI-classified automation risk
Explore the full dashboard at /data/workforce-resilience, or dig into occupation-level data on the Census AI Exposure Treemap.
References: Autor, Dorn & Hanson (2013), "The China Syndrome: Local Labor Market Effects of Import Competition in the United States", American Economic Review; Levy Yeyati & Sosa (2024), "The Adoption Speed Dilemma: Technology Transitions and Labour Market Absorption", CID Working Paper; Anthropic (2025), "The Anthropic Economic Index"; Felten, Raj & Seamans (2023), "Occupational, Industry, and Geographic Exposure to Artificial Intelligence", Strategic Management Journal.