The public-sector talent marketplaces that will fail

State and local governments are investing millions into artificial intelligence-powered workforce platforms, talent marketplaces and career navigation tools. Many of them will quietly fail within three years, not because the technology is broken, but because the systems behind them were never designed to maintain trust over time.

A workforce platform can look successful before it has helped anyone. It can look great and function well and be delivered on budget and on time. Then it collides with real people and real data. A worker uploads a resume and the analysis is wrong, the job recommendations are weak after a long onboarding, a posting is a dead end, a training program is closed, or the cognitive load is so high the worker never begins.

The carrot that encourages people to contribute rots quietly. From the staff side, it is hard to notice. But workforce platforms carry a different risk profile than commerce platforms because a poor experience does not only repel a user. It can widen a labor gap by deepening the discouragement that already defines a job hunt. When metrics depend on a worker’s time and hope, a weak platform can lengthen the distance to a good job.

None of this looks like a dramatic failure, but the promise weakens in the one place workforce technology has to be strongest: trust. And the network effects that drive adoption often evaporate quickly.

Workforce technology must assert its own criteria for success. We are not selling shoes or matching couples, but we may be somewhere in between. In a recent analysis, we describe this as a lifecycle problem: public-interest technology is resourced for delivery, not for the stewardship that figures out why it is not working. Fortunately, it is less about increasing spending, but rescheduling it.

Across the country, state workforce agencies are experimenting with AI-powered job matching, skills-based hiring tools and integrated workforce portals. Governments are also facing rising concerns around ghost job listings, outdated training data and whether automated recommendations reinforce inequities instead of reducing them.

A Governance Gap

Most often, a platform degrades within three years. 

Modern talent marketplaces may include learning and employment records, credential registries, skills taxonomies, employer portals, case manager tools, and AI-enabled matching. Each piece introduces a governance question. Who validates a credential? Who updates a program record? Who owns the data after launch? Which staff can correct errors without waiting on a developer? What happens when a worker flags something wrong?

Requests for proposals define features but cannot be expected to predict operating responsibilities and how they’ll shift. The result is a platform that satisfies the contract but lacks the routines, roles and budget to stay trustworthy. Job postings and training records decay. It does not have to break to become unreliable, it just has to fail being cared for. In other words, distribute as little of the budget for launch and as much as you can for what’s after.

Missing Behavioral Design

Workforce platforms often assume better information leads to action. After examining product trends across four million workers, thousands of providers and employers, and hundreds of resources, behavioral friction is the problem to solve.

Displaced workers are often overwhelmed, discouraged and under time pressure. Workforce platforms that require long onboarding processes, multiple uploads or confusing navigation lose users quickly. 

A platform has to earn and keep trust and salve a wound along the way. That shows up in choices specific to workforce technology. Can a micro commitment be made? Can cognitive load be lessened? Can social proof surface sooner?

Hidden Costs and Bad Data

Workforce data and software decays faster than most budgets account for. Job descriptions, providers, public datasets, credentials and software versions are dynamic and many critical experiences break a lot. Even with clean data and great software updates, it’s an uphill trust battle as ghost jobs abound and bots account for 37% of internet traffic. 

If governance is not structured for this, trust drops, then usage drops. The system loses the behavioral signals and then the partner engagement it needs. Weak data lowers trust, lower trust lowers adoption, lower adoption makes good data harder to gather. 

AI can support monitoring and all manner of building efficiency, but it cannot replace responsible stewardship, which is increasingly complicated.

A Deficit of Product-Led Growth

Product-led growth means people get value on day one, adoption spreads through the product itself, and participation makes the system increasingly useful. Many workforce tools are quietly relying on staff heroics and manual work. 

When malfunctioning, this looks like a direct relationship between staff capacity and user outcomes. The loudest signal of this is when the work reaches a fever pitch and triggers the dreaded (needed) rebuild.

For public-sector technology, we are accustomed to depreciating technology, but platforms can appreciate: both the reality of its user behaviors, and in its value as an asset. They can also be just as affordable, well maintained, more trusted, and more useful on day 2,000 than on day 20. A starting test: What is your cost per successful outcome, year over year? The goal is that it drops.

Start at the Minimally Ethical Product

Aside from budgeting away from launch and towards the life of the product, how else might we start a crisp, impactful, sustainable build? The rest of the tech world operates on the MVP: “the smallest build that delivers value.” 

For workforce technology, we find that question is insufficient. A product can be viable for a board while still creating friction for a worker. It can point people to dead ends, fake jobs, expired programs, bad recommendations, overpromises, biased information, or just quietly decaying functionality.

If we elect a minimally ethical product, we can ask a different question: What valuable thing can we responsibly ask people to rely on without creating harm through confusion, inaccuracy, bias, dead ends, or neglect?

That may mean starting with one sector or region. A job board with spam fails the test. So does an AI chatbot with bias, which is far more common than expected. Can staff see which records are going stale? Can a case manager share a link they trust? Can a worker understand what is and is not promised?

Governments should evaluate workforce technology less like a one-time software purchase and more like long-term infrastructure. The key question is not whether a platform launches successfully, but whether it is appreciating in trust and value three years later. 

The next generation of workforce technology will not succeed because of bigger feature lists or better AI demos. It will succeed if workers trust it, staff can maintain it and governments budget for stewardship instead of just launch. In workforce technology, reliability is the product.

Leah Lykins is co-founder of WhereWeGo.