The Numbers on Job Retraining Show How Loud AI Disruption Is About to Be

TRANSMISSION RECEIVED · PLANET ECON-3 · COORDS [-0.32, -0.44]

Retraining is the most popular policy answer to AI-driven labor disruption: teach displaced workers new skills and they land new jobs. Anthropic’s Economic Research team, with independent researcher David Roodman, did what few have bothered to do — read the evidence carefully. Drawing on 56 randomized US studies combined in a new meta-analysis, plus experiments from Europe, the verdict is sobering: the programs work, but only a little.

The average program moves the needle a fraction of an inch

For each person offered a training slot, employment rises two to three percentage points and earnings by roughly $1,000 a year — against a cost of about$13,000. Count the added tax revenue and reduced benefit payments and the government recovers more than half of what it spends; the programs roughly break even. That sounds fine until you set it next to the scale of disruption AI could bring.

A small set of sector programs reverses the picture: they partner with employers in a high-demand industry and place people directly into real jobs, and their gains are several times larger. That is the good news — and the bad news. Attempts to scale or replicate them have repeatedly failed. The authors conclude that if AI displaces workers at scale, existing retraining programs would likely fall short. Their central recommendation is not to promise retraining as a balm, but to invest now in demonstrating, evaluating, and scaling the few programs that actually work — and to measure results rigorously rather than assume good intentions produce outcomes.

Why marketers should care

This sits on Anthropic’s Economic Index, which tracks how AI is actually used across occupations and industries — a live map of which roles get augmented, not which get headlines. For agencies the report is a lesson in where real effect lives in the economy: what works is employer-partnered and tied to a specific real output and a real job, not a generic intervention. The same logic governs how marketing deploys AI: a generic assistant rolled out to everyone is the average retraining program (modest, diffuse); a program aimed at one business outcome, instrumented and measured, is the sector program (larger effect, but it must be built for it, not bolted on). When clients ask what AI does to their workforce or their agency roster, citing the actual magnitudes — not the panic, not the hype — is a competitive advantage.

How to use it

• Treat Anthropic’s Economic Index as a live map of which occupations actually get augmented, to see where demand and talent are moving.
• Demand effect sizes, not anecdotes: the report found several-fold differences between truly working programs and average ones; apply that skepticism to AI-spend claims.
• If you upskill your own team with AI, partner it to a real output and measure the outcome — generic training fails at scale, and so do unfocused rollouts.

// END OF LOG