ADP Employment Report: Payroll Records, Not a Jobs Survey

The ADP National Employment Report is the release most often introduced as a preview of the government payrolls number. Its publisher does not describe it that way, and the two are not built from the same kind of information at all.

What follows is what ADP states about its own report in the monthly releases it issues, what the number is drawn from, when it lands, and what a trader can and cannot settle with it.

Where the ADP Number Comes From

ADP is a payroll processor. Employers who use it run their wage payments through its systems, which means the company holds the payment records of a very large number of American workers as a by-product of doing its ordinary business.

The monthly release states the scale directly. The report is built on payroll data that is anonymised, collected weekly, and drawn from over 26 million American workers on private-sector payrolls. It is produced by ADP Research with the Stanford Digital Economy Lab.

Two words in that description carry most of the weight. Payroll means the figures come from money that was actually paid, recorded at the moment it was paid, rather than from anyone reporting what they believe happened. Weekly means the underlying records arrive at a finer interval than the monthly figure that is published from them.

The workers counted are those of ADP client firms. That is a large population, and it is not a randomly drawn one, which matters for the comparison in the next section.

Why the Publisher Does Not Call It an NFP Forecast

Market commentary routinely presents the ADP number as an early read on the government payrolls figure due later in the week, and treats a gap between the two as an error on ADP’s part.

ADP does not describe the report in those terms. Each release presents it as a measure of the labour market in its own right, produced from its own data, with no claim that it estimates or anticipates what any other body will publish.

That distinction changes how a difference between the two should be read. If one number is an attempt to predict the other, a gap is a miss and the smaller number is wrong. If they are two independent measurements of overlapping but different populations, taken from different kinds of evidence, a gap is information about where those populations and methods diverge.

Nothing here says ADP is the better measure. It says the widespread framing sets up a test the publisher never entered, and that reading a release against a standard it does not claim to meet produces conclusions about it that its own description does not support. The government series has its own construction, examined in our page on non-farm payrolls.

Payroll Records and a Survey Answer Different Questions

An administrative record and a survey response are different kinds of evidence, and the difference is not a matter of one being more careful than the other.

A payroll record is a transaction. It exists because a payment was made, it is complete for the firms whose payments run through the system, and nobody had to remember anything or fill anything in for it to exist. What it cannot do is describe firms that are not clients.

A survey works the other way. It is designed so that a sample can stand in for a whole population, which lets it speak about employers of every kind rather than about one processor’s customers. The cost is that it depends on responses arriving, and on estimation to bridge what did not arrive.

The practical form of that limit is worth being concrete about. A firm appears in the ADP data because it is a client, not because it was selected to represent firms like it, so any part of the economy where ADP has few clients is thinly represented however large that part is. Scale does not fix that on its own: twenty-six million records drawn unevenly still describe only the places they came from.

So the two are answering questions with different shapes. One asks what happened inside a very large and specific set of firms. The other asks what is likely to have happened across the economy. Neither answer substitutes for the other, and a page that treats one as a rehearsal for the other has collapsed the distinction.

The Release Clock: 8:15 Eastern, and Which Weekday

ADP publishes at 8:15 in the morning, Eastern time. The release naming the date of the next one is the reliable source for it: the report covering July 2026 was issued on 5 August 2026, and the report covering August 2026 was scheduled in that same release for 2 September 2026.

The three consecutive dates visible across those releases are 1 July, 5 August and 2 September 2026. All three fall on a Wednesday, which is the pattern the report has followed rather than a rule the publisher states as one.

A widely repeated shortcut says the report always lands a fixed number of days before the government payrolls release. That relationship depends on two separate calendars, and it is read from them rather than assumed. Treat the position of any month’s release as something to look up in an economic calendar, in the same way the position of the JOLTS report has to be looked up rather than counted forward.

ADP Revises Its Own Previous Month

The report is frequently discussed as though each month produces one permanent figure. It does not.

The release issued on 5 August 2026 restated the previous month: the June total, first published as 98,000 jobs added, was moved to 95,000 in that later release. The revision is stated inside the newer release rather than announced separately.

Both rows are read from the ADP releases dated 1 July 2026 and 5 August 2026. The report covering August 2026 was scheduled in the later of those two for 2 September 2026.

Three thousand jobs is a small movement against a total near a hundred thousand, and the size is not the point. The point is that the number a trader reacted to in early July was not the number the series carried a month afterwards, which is the same property that governs the government release and the CPI release.

The practical consequence is narrow and worth stating plainly: a first print is a first print. Any comparison built on ADP figures from several months back is being built on revised values, while the figures that moved prices at the time were the unrevised ones.

The Pay Figure Most Coverage Drops

Each release carries two measures, and coverage usually reports one. Alongside the change in employment, ADP publishes a year-over-year change in pay.

In the release covering June 2026 that figure was 4.4 per cent. In the release covering July 2026 it was 4.4 per cent again. The employment number between those two releases moved a great deal, from 98,000 as first published to 44,000, while the pay measure did not move at all.

The pay measure is a change against the same month a year earlier, which is why it moves slowly compared with the jobs figure. A monthly employment change can swing by tens of thousands between releases; an annual rate of pay growth is averaging over twelve months and will not.

That is the argument for reading both. A month in which hiring slows while pay growth holds steady is a different month from one in which both fall, and a reader who only ever sees the jobs figure cannot tell those apart.

What the Report Can and Cannot Settle

Five checks are enough to use the release without over-reading it.

1. Read the release itself rather than the headline written about it, since the revision to the prior month sits inside the same document.
2. Note both figures: the change in employment and the year-over-year pay change.
3. Take the date of the next release from the current one instead of counting forward from another body’s calendar.
4. Treat any comparison against the government series as a comparison of two different measurements, not as a scoring of one against the other.
5. Treat the first print as provisional, and check whether a figure you are relying on from an earlier month has since been restated.

What the report cannot settle is what any market will do when it is published. It is a measurement of employment at a set of firms, released at a known time, and everything beyond that is a question about positioning and expectations rather than about the data.

Frequently Asked Questions