A change of 104,000 seems large, but the number actually estimated is total employment, which changed by less than 0.1%, which is a very small change. However, since people pay attention to the change in employment, it looks like a big number (104,000). Why are there revisions? The numbers are based on a survey and like any survey, some responses come in late. If the late responses differ significantly from earlier responses, the estimated amount of employment may change (for example, the late responders show significantly more job growth than those that responded on time). Also, there’s something called the birth-death model which is designed to capture the effect of new businesses being created and existing businesses disappearing. They used to adjust the model once a year, but now adjust it quarterly, so revisions are incorporated more quickly. Most significant revisions take place at turning points in the economy. For example, at the beginning of the Great Recession, the existing birth-death model was overly optimistic, reflecting the pre-recession behavior of the economy. The revision for August was historically high (higher than just a handful of revisions outside of turning points).
So the numbers are subject to revision and a minor revision to overall employment (such as 0.07% in August) appears to be large when one looks at the change in employment. For more information on revisions to employment data, click here.