Measuring What Works: AI-Displacement Reskilling under WIOA Title I

State and local workforce boards do not need a new law to begin funding AI-displacement reskilling. The authority already exists under Title I of the Workforce Innovation and Opportunity Act, and the six Section 116 primary indicators of performance already define what “working” means for any program a board runs. The only missing layer is a measurement design that connects an occupation-level AI exposure baseline to those exact six indicators. This paper supplies that design.

It is written for the people who carry the accountability: Workforce Development Board directors, WIOA program and performance managers, Rapid Response coordinators, and state policy staff who negotiate performance levels and face sanctions for two consecutive program years of failure under Section 116(f). The argument is not that AI changes what a board is responsible for. It is that AI displacement is now a routable, measurable category of dislocation that fits squarely inside the system boards already operate, and that the way to make it defensible is to design every measurement decision backward from the indicators a board already reports.

JobRoute’s role in this is narrow and deliberate. It supplies the measurement engine: a reproducible, version-locked, source-traceable layer that reconciles public sources per occupation. It does not publish a competing headline displacement statistic, and it does not make eligibility determinations, which remain the board’s under state policy. For the public-sector view of how that engine is meant to be used, see JobRoute for public sector.

Why are episodic AI studies and the 2013 Frey-Osborne 47 percent figure insufficient for program design?

The labor-market signal is real and large. The problem for a workforce board is that the signal arrives in two forms that cannot run a program. The first is the episodic study: a major report from a research institution, published once, describing the aggregate. The second is a single dated probability that has migrated into consumer “will AI take my job” calculators. Neither can tell a board which dislocated worker to route where, and neither maps to any Section 116 indicator.

Start with the scale, because it justifies treating AI displacement as a planning category at all. The World Economic Forum’s Future of Jobs Report 2025 projects that structural labor-market shifts will create 170 million new jobs and displace 92 million by 2030, a net gain of 78 million. Over the same 2025 to 2030 window, the same report finds that 39 percent of workers’ existing skill sets are expected to be transformed or become outdated, and that 59 percent of workers will need reskilling or upskilling by 2030, of whom 11 percent are unlikely to receive it. These are 2030-horizon figures, and a net positive aggregate does not mean a smooth transition for any individual worker or any local labor market.

The dated-probability problem is concrete. The widely cited 47 percent figure comes from Carl Benedikt Frey and Michael Osborne’s 2013 Oxford Martin School working paper, which estimated that about 47 percent of total US employment was at risk of computerisation across 702 occupations. That estimate has aged into shorthand, but it is a single occupation-agnostic probability from 2013. It cannot tell a board which of its dislocated workers face high task-level exposure, which face mostly augmentation, or where adjacent lower-exposure roles with real local demand sit. It maps to no performance indicator.

More recent, task-level evidence shows why a headline percentage is the wrong unit. The Anthropic Economic Index, which maps Claude usage against the US Department of Labor O*NET task database, found in its January 2026 report that for the November 2025 data window, roughly 52 percent of Claude.ai conversations were augmentative and 45 percent were automative. The point for a board is not the exact split, which the index itself shows moving release to release. The point is that exposure is task-level and mixed, not a whole-occupation death sentence. That is precisely why a per-occupation, task-aware baseline beats a single number, and why the baseline should be a reconciled, version-locked reference rather than a moving headline. For how that reconciliation is constructed, see the methodology behind the score.

The public-system stakes show up most sharply in reemployment data. The Bureau of Labor Statistics, in its Worker Displacement 2021-2023 release, found that 65.7 percent of long-tenured displaced workers were reemployed when surveyed in January 2024. That headline hides the variance the board has to catch: among prime-age workers ages 25 to 54 the rate was 74.5 percent, but it fell to 55.3 percent for those ages 55 to 64. Reemployment is uneven across exactly the kind of population an AI-displacement cohort is likely to include, and that unevenness is what a measurement design has to surface rather than average away.

Is AI-displacement reskilling fundable under existing WIOA Title I authorities?

Yes. WIOA Title I authorizes three formula programs, Adult, Dislocated Worker, and Youth, delivered through the American Job Center one-stop system and administered by the Department of Labor’s Employment and Training Administration (Congressional Research Service R44252, updated September 26, 2022). Two of these streams, plus the Rapid Response reserve, directly support AI-displacement reskilling today.

The Dislocated Worker program is the primary stream for workers losing roles to AI-driven restructuring. It funds career services and training for individuals who meet the dislocated-worker definition, which the next section addresses in detail.

The Adult program carries broader eligibility, subject to priority-of-service rules, and can reach incumbent and lower-wage workers whose tasks are being automated before a formal layoff. This matters because AI exposure often appears first as task erosion inside a still-existing job, not as a clean termination event.

Rapid Response is the proactive entry point. Of the state allotment for dislocated worker activities, the governor must reserve not more than 25 percent for dislocated worker Rapid Response activities (Congressional Research Service IF11530, 2020). Rapid Response is built for layoff events, and an AI-driven layoff is still a layoff. It is also the natural vehicle for pre-layoff engagement at employers who are signaling automation, which lets a board reach affected workers before separation rather than after.

Does AI-driven role elimination fit the statutory definition of a dislocated worker?

Yes, in the typical case. The statutory definition at 29 U.S.C. 3102(15) turns on the fact of dislocation and the prospect of return, not on the cause. A worker who has been terminated or laid off, or has received notice of termination or layoff, who meets the workforce-attachment test, and who is “unlikely to return to a previous industry or occupation,” qualifies regardless of whether AI was the trigger. The Congressional Research Service notes there is no requirement that the dislocation have a specific cause.

The categories the statute covers map cleanly onto AI-driven scenarios:

• Terminated or laid off and unlikely to return to a previous industry or occupation. This fits role consolidation, where a function is restructured around automated tooling and the prior role does not come back.
• Permanent closure of a plant or facility, or a substantial layoff. This fits an automation-driven layoff event, which is the ordinary Rapid Response trigger.
• The anticipated-closure provision for workers at a facility expected to close within 180 days. This fits proactive engagement when an employer signals an automation-driven downsizing.
• Self-employed workers, including farmers, ranchers, and fishermen, who are unemployed due to general economic conditions or natural disasters. This can fit independent workers whose niche is eroded by widely available automation.

One honest caveat. Eligibility is determined by the local board under state policy and on the facts of each case. An exposure baseline informs the routing decision and helps document the rationale. It does not override the eligibility determination, and it should never be presented as if it does.

What exactly are the six WIOA Section 116 primary indicators of performance?

These six indicators are the federal definition of “working.” Any AI-displacement initiative a board launches must be defensible in exactly these terms, because these are the terms the board reports and against which it negotiates and is sanctioned. The statutory list is at 29 U.S.C. 3141(b)(2)(A), and the regulatory detail for indicators five and six comes from Department of Labor guidance and the joint final rule.

Two indicators deserve regulatory detail because that detail shapes the measurement design.

Indicator six, Effectiveness in Serving Employers, was defined by the Department of Labor and Department of Education joint final rule (RIN 1205-AC01 and 1830-AA32, 89 FR 13814, Federal Register, effective March 25, 2024) as Retention with the Same Employer. It is the percentage of participants who exited in unsubsidized employment and were employed by the same employer in both the second and fourth quarters after exit. It is a statewide measure, reported by one core program on behalf of all six core programs in the state.

Indicator five, Measurable Skill Gains, is documented progress toward a credential or employment during a program year, captured through recognized types of evidence (Department of Labor Employment and Training Administration, WIOA Performance Indicators and Measures). These include an Educational Functioning Level gain, attainment of a secondary school diploma or recognized equivalent, a transcript or report card showing satisfactory progress, a documented training milestone via a progress report, and a skills progression documented by passage of an exam. The practical implication is that in-program progress must be captured during the program year, not inferred after exit.

All six levels are negotiated and adjusted, a point the sanctions section returns to.

How does an occupation-level exposure-and-routing approach map to each of the six indicators?

The central design principle is simple. Do not build a parallel vanity metric. Design every measurement decision backward from the six indicators the board already reports. Exposure-and-routing data is useful only to the extent that it improves performance on those indicators, and the contribution should be stated operationally, not as an invented statistical lift.

For indicators one and two, the design objective is re-entry that holds at four quarters, not just two. Routing toward lower-exposure adjacent roles with real local demand is what makes the fourth-quarter number hold. For indicator three, adjacent-role mapping should favor lateral or upward moves, with the honest acknowledgment that the fastest reemployment is not always the highest-earning placement. For indicator four, the training must be tied to a recognized credential that is attainable within one year after exit. For indicator five, training milestones and the five MSG documentation types should be aligned to durable-skill acquisition so progress is recorded during the program year. For indicator six, routing should be coordinated with employer-side demand, including Rapid Response employer engagement, so placements land with employers positioned to retain the worker through Q2 and Q4. The employer-facing complement to this, modeling the cost of redeployment and retraining, is covered in the enterprise cost model.

How would a board design a measurable AI-displacement program end to end?

The design is four steps, and reporting falls out of it as a byproduct rather than a retrofit.

Step 1: Set the baseline. Establish a per-occupation AI exposure baseline for the local labor market using a reconciled, version-locked engine across O*NET 30.2, ESCO v1.2.1, Lightcast Open Skills, the Anthropic Economic Index, the WEF Future of Jobs Report 2025, and the Bureau of Labor Statistics. The point of version-locking is that the baseline is a fixed, reproducible reference rather than a moving headline number, which is what makes it auditable later. See the methodology behind the score for how the reconciliation works.

Step 2: Identify and tag the cohort. Define and tag the AI-displacement Dislocated Worker cohort within the existing case-management system. The cohort must be separable for reporting, but it should not require a new data system. A flag inside the system the board already uses is enough.

Step 3: Route. Map each worker’s prior occupation to durable skills and to adjacent, lower-exposure roles with local demand, then route career services and training accordingly. The worker-facing and board-facing views should share one data graph, so connect the participant experience to the free AI Ready Score and to how the score works. A worker seeing the same baseline the board uses is a coherence and trust gain, not a duplication.

Step 4: Instrument outcomes. Wire outcome tracking to the six indicators from day one: Q2 and Q4 employment via wage records, median earnings in Q2, credential attainment within one year, MSG documentation types captured during the program year, and same-employer retention for indicator six. When the indicators are wired at design time, reporting is a byproduct of the program rather than a scramble at the close of the program year.

The same baseline supports proactive Rapid Response. A board can use occupation-level exposure to engage employers signaling automation before separations occur, rather than waiting for a post-layoff event, which is exactly the proactive use the Rapid Response reserve is designed to fund.

How do negotiated levels, the statistical adjustment model, and Section 116(f) sanctions shape a defensible design?

The accountability mechanics are the reason a measurement design has to be defensible, not merely plausible. States negotiate performance levels for each of the six indicators. Those levels are adjusted through a statistical adjustment model that accounts for participant characteristics and economic conditions, and performance is scored against the adjusted level, not a flat target.

Sanctions follow from failure. Under Section 116(f) and the Department of Labor’s Employment and Training Administration negotiations-and-sanctions guidance (TEGL 11-19, including Change 2 issued in January 2025), sanctions apply for failure to report or failure to meet adjusted levels of performance when the same failure occurs in two consecutive program years. Financial sanctions can reach a defined share of the Governor’s Reserve. A single off year is a warning. The second consecutive failure is what carries the financial consequence.

This is where cohort characterization becomes a design obligation rather than a nicety. The AI-displacement cohort may carry harder-to-serve characteristics. The BLS reemployment gap for workers ages 55 to 64, at 55.3 percent against 74.5 percent for prime-age workers, is a concrete example of a population that performs differently through no fault of the program serving it. A board must be able to characterize its cohort accurately so the negotiated levels and the statistical adjustment model reflect reality, rather than penalizing the board for taking on a tougher population.

A measurement layer that publishes no competing headline statistic is exactly what a board needs here. When a board must defend its outcomes against negotiated Section 116 levels, it is far better served by a reproducible engine that reconciles public sources than by a vendor forecast it cannot reproduce or audit.

Methodology, data, privacy, and the honest limits of exposure measurement

Stated plainly, the measurement approach and its limits are as follows.

Reproducibility and source-traceability. The engine reconciles O*NET 30.2, ESCO v1.2.1, Lightcast Open Skills, the Anthropic Economic Index, the WEF Future of Jobs Report 2025, and BLS data. Every input is attributable, and every run is version-locked and reproducible. That property maps directly to audit, to federal reporting, and to defending negotiated levels under the statistical adjustment model. Reproducibility is the compliance asset, not a technical flourish.

Privacy by design. Use occupation-level exposure for accountability and program design, not individual AI scoring of named participants. Keep wage-record and case-management data inside existing WIOA confidentiality and data-sharing frameworks. The accountability question a board needs to answer is about occupations and routing, which does not require scoring named people.

Honest limitations. Exposure is a probability-weighted, task-level signal about an occupation, not a prediction about any individual. The augmentation-versus-automation balance shifts over time; the Anthropic index moved several percentage points between its August 2025 and November 2025 windows. Local labor demand can diverge from national exposure. And the credential and earnings timing windows constrain what any single program year can show.

A pre-launch checklist for boards:

• Confirm the cohort definition and tagging inside the existing case-management system.
• Confirm each indicator’s data source and measurement timing.
• Confirm that the negotiations conversation reflects the cohort’s actual characteristics so the adjustment model is fair.
• Confirm the privacy and data-sharing posture against existing WIOA frameworks.
• Confirm version-locking of the exposure baseline so it can be reproduced for audit.
• State the limitations publicly, before launch, rather than after a challenge.

Exposure is the start of a plan, not the end of a career. For a public workforce board, it is also the start of a measurable, defensible program, one that can be funded under authorities that already exist and reported against indicators a board already owns. For the public-sector view of the engine and how it fits a board’s accountability obligations, see JobRoute for public sector.