Imagine you have limited resources to help people facing a serious problem.

Who should receive assistance first?

The obvious answer might be those who need it most. But new economic research highlights a fundamental complication:

The people who need help most are not necessarily the people who benefit most from the help available.

These are two different questions, and confusing them can lead to surprisingly ineffective decisions.

Belgian job-search help showed the gap between need and benefit

A working paper published in October 2026 examined how public employment services in Belgium allocate job-search assistance.

Employment agencies increasingly use algorithms to predict which unemployed people are most likely to remain without work for an extended period.

The reasoning seems straightforward: identify those at greatest risk and direct assistance toward them.

But researchers analyzing administrative data from the Flemish Public Employment Service discovered something unexpected.

People at the highest predicted risk of long-term unemployment often benefited less from job-search counseling than people at lower risk.

Counseling improved employment outcomes on average, but its effectiveness declined substantially as predicted unemployment risk increased.

The researchers then examined alternative ways of allocating assistance.

Instead of focusing exclusively on who was most likely to remain unemployed, they considered who was most likely to benefit from receiving counseling.

In their calibrated policy model, this approach produced welfare gains nearly three times those achieved through random assignment.

Remarkably, allocating assistance based only on unemployment risk performed worse than random assignment in the model.

The same gap appears well beyond unemployment

The distinction between need and responsiveness extends far beyond employment services.

Consider healthcare.

A patient with the highest risk of developing a disease is not necessarily the patient who would benefit most from a particular preventive treatment.

The same principle applies to education, where students struggling most may not always benefit most from a specific intervention.

This doesn’t mean people with the greatest needs should receive less assistance.

It means that identifying who needs help and identifying what help will work are separate problems.

A person who responds poorly to one intervention may benefit enormously from another.

The Belgian study illustrates why accurate predictions alone are insufficient for making effective decisions.

An algorithm might correctly identify the people facing the greatest difficulties while still recommending an ineffective allocation of resources.

The research suggests that organizations should consider not only the severity of a problem, but also how much a particular intervention is expected to improve the outcome.

What the study doesn’t show

  • It’s a working paper (October 2026) and hasn’t been peer reviewed yet.
  • It covers one kind of help, job-search counselling, from one public employment service in Flanders, Belgium.
  • The welfare gains of nearly three times random assignment come from a calibrated policy model, not from actually reallocating counselling and measuring the results.
  • The finding that targeting by risk alone did worse than random assignment is also a model result.
  • Predictions of who will benefit are estimates themselves, and they can be wrong.

Need and benefit are separate questions

The broader lesson is simple:

Knowing who needs help is not the same as knowing how to help them.