Unemployment insurance is a state system that spends most of its life underused. Claims run at a low, predictable rate, staffing is set against that rate, and the technology is judged adequate because it handles what arrives.
Then something happens to the labour market and the volume rises by an order of magnitude in a fortnight. That is the only condition the system exists for, and it is the condition it is not provisioned for.
The surge is the design case and nobody funds it
The economics are what make this durable rather than incidental. A state legislature funds administration out of a budget that competes with roads and schools, in years when the caseload is low and nobody is complaining. Capacity that sits idle for eight years and matters enormously in the ninth is the hardest thing in government to appropriate for, which is why it consistently is not.
So the surge finds a system with no headroom. Phone lines fail first, then the web front end, then the adjudication queue — because a claim that needs a human decision cannot be resolved by adding servers, and the humans were sized for the ordinary week too.
The technology story is real and it is second, not first. A great deal of this runs on decades-old software that few people are left to maintain, which is the trillion-dollar rewrite nobody can postpone in its most consequential form. Modernisation projects are large, frequently overrun, and have a poor record — and a modernised system provisioned for ordinary volume fails the same way, faster.
The staffing half is the one that gets least attention and is harder to fix. Adjudicators need training measured in months, so a state cannot hire into a surge; it can only have hired before one. That is the same problem as agencies unable to hire fast enough to spend what Congress appropriated — the constraint is not money at the moment of need, it is the lead time on competence.
What states with better records have in common is unglamorous: a standing surge roster of trained former staff who can be recalled, load-tested systems tested at multiples of normal volume rather than at normal volume, and a claims process that can be simplified deliberately under stress rather than improvised.
The cost of failure is not administrative. A benefit that arrives eleven weeks late has already failed the household it was for, whatever the eventual payment says.
