The BLS cannot survey every business that opens or closes in a given month. To account for firms too new or too small to appear in the sample, they use a statistical model called the birth-death model. It estimates net job creation from business formation and closure.

In stable conditions, the model performs reasonably well. After major economic disruptions, it tends to overshoot.

In August 2024, the BLS released its annual benchmark revision and subtracted 818,000 jobs from the prior 12 months of payroll data. That is the largest downward revision in 15 years. The model had been crediting the economy with small-business hiring that, on net, was not occurring at the assumed rate. The methodology was later adjusted to incorporate real-time sample data starting January 2025. That adjustment is itself an acknowledgment that the prior approach was lagging reality.

When you see a strong NFP print, the birth-death component is worth checking. That is where revisions tend to accumulate.