Pre-benchmarking jobs: +11,000. Private-sector measures not immune to data mess. Trend shows “employers have been cautious with hiring.”
By Wolf Richter for WOLF STREET.
Payroll processing firm ADP conducted its annual preliminary adjustment today to the full-year 2024 Quarterly Census of Employment and Wages data released by the Bureau of Labor Statistics on September 9. The QCEW provides a quarterly count of Paid Employment based on the quarterly payroll tax filings by employers, covering about 95% of payroll jobs, but the BLS releases it annually with a long lag. On September 9, the BLS announced its own preliminary benchmarking adjustment to the QCEW for the 12-month period through March 2025 (we discussed this mess here). Today it was ADP’s turn.
As part of the annual adjustment, ADP reduced its September job count by 43,000 jobs. Companies added 11,000 jobs to their payrolls in September, per pre-benchmarked data. The 43,000-job adjustment to benchmark the data to the 2024 QCEW took this small job gain to a job loss of 32,000.
Same thing for August: ADP applied the 2024 QCEW benchmark adjustment of 57,000 jobs to the August job gain of 54,000 jobs, which flipped it to a job loss of 3,000. The ADP data is seasonally adjusted.
The erratic month-to-month changes and adjustments are somewhat leveled out with the three-month average (blue line in the chart above), which shows a substantially slowing job growth, which is what we have seen in other data as well.
“Despite the strong economic growth we saw in the second quarter, this month’s release further validates what we’ve been seeing in the labor market, that U.S. employers have been cautious with hiring,” ADP said.
Total employment in the private sector, per ADP, was 134.53 million, up by 1.15 million from a year ago, after the benchmark adjustments:
The median increase in wages in September, year-over-year, based on a subset of 14.8 million workers employed for at least 12 months, whose paychecks ADP processed:
- For “Job Stayers”: +4.5%, roughly the same increase as over the prior 6 months.
- For “Job Changers”: +6.6%, down from the 7.0% to 7.1% range in the prior 6 months.
Due to the government shutdown, the BLS may not release its employment report on Friday, and the ADP data may be all we get for now in terms of payroll count and wages for September.
Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the mug to find out how:
A passing thought: Much of this “strong economic growth” revolves around the rapid growth of A.I., where the goal of A.I. is to eliminate the need for human labor.
IOW, this isn’t just another economic cycle, it’s a sign of the future.
Bingo. I am not sure why there has not been a correlation with job losses vs AI. Like the internet age of the 90s, this isn’t new. The only difference is the product being made. The internet was a change for all income levels, even at its infincy. AI will most likely not benefit the average Joe/Jane. Business is ahead on this one.
One of these days, I would love to see a list of jobs that AI will replace (the human).
Coming out of heavy manufacturing, and then the oilfield (engineering and construction), I can’t see the mass of eliminations.
Look up the post from ADP “Yes, AI is affecting employment. Here’s the data.”
Includes a link to a research paper that estimates which jobs are exposed to AI.
That paper’s title is “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”
Really doubt that AI has much to do with the recent bad jobs reports and jobs losses in the US though, that’s a convenient excuse to pump the AI bubble and hype a little more but from talking to companies and just serious failings with AI it’s not the reason. The brutal truth is the AI doesn’t work very well and certainly in no place to replace workers. It’s completely unreliable and constantly gives answers that look nice but are wrong in the worst places, that’s even worse than being obviously wrong and poorly written because managers miss it. It hallucinates like crazy and in fact the bigger and more massive the AI, worse the hallucinations–the AI experts point out that’s not something tech can help, it’s basic to how AI works. And worst of all now the AI’s are starting to take in more of their own AI-produced sloppy content and regurgitate it back, so it’s accuracy and usefulness is falling and start to see model breakdown.
The reality is the value addition from AI just isn’t there, certainly not anywhere close to the tens of billions of dollars (often from margin-juiced debt) being spent on it and the data centers. And again it’s not really fixable–these are problems basic to how AI works. (and AI isn’t really new after all it’s been around for decades, we just have better chat-bots now due to better graphics chips and hardware) So we’re already at diminishing returns for AI ROI and it actually gets worse, because extra billions of dollars thrown at the problem just makes the confident sounding wrong answers, hallucinations and self-regurgitating training even worse. It comes down to a few companies like Oracle and OpenAI pumping each other with spending for something with very little good ROI in the real world for other companies esp given how much spent. And even steps to try to make an AI do more like a worker could do, it would just come with so much extra in costs for the data centers, power and water it’s just cheaper to hire more workers–and the more the AI veers from a set path, less reliable it is. Already even the big firms are cutting back on data centers and cancelling many projects. The ROI isn’t there, or it’s even backwards ROI as all the errors and hallucinations build up.
No the job cuts in the USA right now are more from off-shoring, or just cost-cutting with all the buybacks and pumps going with it, with AI offered as an excuse to sound hip with the latest tech and hype. Inflation and housing bubble costs just keep getting worse and it’s harder for companies to expand or attract customers, so cost-cutting with layoffs helps profitability at least for a quarter or two. Of course this just means Americans struggle even more as costs keep going up. And Fed rate cuts wouldn’t help either–those yields went way up when JPow tried and made everything worse, while the dollar just keeps diving even worse and makes things more expensive, while the prolonged Everything Bubble just pushes the US birth rate TFR even lower when people can’t afford homes. If the Fed was smart they’d realize this is a good time to channel Paul Volcker, pop the asset bubbles and mis-allocated capital and nip inflation in the bud. Not easy but this way at least, we start get back to price discovery, market forces and reasonable prices for homes and everything else.
I use Google AI all the time. So much so I practically stopped clicking on most links in the Google search. Is Google planning to monetize my “not clicking/not visting websites” somehow?
Also, when I need to know the truth about the economy I skip AI industrial complex and go directly to Wolfstreet ;-)
Me too!
You’re depriving the publishers whose links you don’t click on of your traffic and ad revenues. But Google, which stole these publishers’ content to give you the answer, is running ads on its search page, and GOOGLE is getting paid for those ads, instead of the publishers that Google stole from to get the content.
I don’t understand how this applies to their AdWords? Publishers pay Google only when their -links are clicked
Alku
So when the traffic stays on the Google search page because AI stole the content from other websites and presents it to readers directly, then those readers only see the ads on the Google page, and if they click on them, Google gets 100% of the money paid by the advertiser.
Without AI, the reader would see a link, click on it to get the answer, land on a website with ads (Google or other), and maybe click on the ads, or view the ads (CPM ads), and the publisher gets paid. If those ads are Google ads, then Google only gets a big cut, instead of all of it. And if those ads are non-Google ads, Google gets none of it. So the publisher gets at least some money for the work they did.
There are links near the AI text on Google search (on laptop screens to the right and below, and sometimes in the text), but apparently few people click on them after reading the AI text. And in my experience, they don’t really contain what that AI text used as sources.
The “zero-click search” strategy has for years been a priority at Google before it rolled out its AI. It used to just present a relevant section of the text from the publisher along with a link, but most people just read that section without clicking on the link. AI has moved zero-click search a step further.
I hoped they get crushed with Copyright infringement issues but like with everything corporations will win. I use AI too but I always click through to the source otherwise it might be inaccurate, or worse, purposeful propaganda. Depends of course on the topic.
i always go straight to your website to get the info. and i initially found out about your website through Google News.
Yes, Google News is different. Google News is a good product.
Wolf,
This is exactly what I don’t understand. If they push for “no click”, this cuts their AdWords revenues.
Read my comment again. That’s not what I said. You’re talking about clicking on a Google ad on the Google search page. As I said, Google gets 100% of that.
Zero-click means you stay on the search page and click on search ads, but do NOT click on the search results, which take you to the publisher’s page.
I’m talking about clicking on the link to the publisher’s site in the search results. These links are NOT ads. They’re search results. Google gets nothing for that, and it is reducing the chances that you will, or even can, click on a link like that, in order to keep you on the search page and get you to click on an ad.
2 ways to lose in today’s stock market:
1) AI replaces enough jobs to cause a consumer spending slowdown and recession,
2) AI fails to replace enough jobs to justify the massive investments, stocks crash in a repeat of the telecom bubble, and a recession occurs.
You’re awesome man, love this comment :)
As a former IT employee, we have been using AI since the late 70’s to early 80’s. I don’t know how many times I have written scripts that optimized my responsibilities so much that I had eliminated my role. AI is here to stay. It will not take all the jobs as humans will still do it cheaper. Data centers need dedicated power plants and that is very expensive. It will change our lives more like the Jetsons and less like Flintstones. That is why my children’s financial literacy includes valuable lessons from Wolf street. Live below your means, avoid debt, don’t buy influencer panhandles, and be happy with what you produce. Relax! There are a lot of people needed in skill trades or healthcare. Innovate and build new things! Life is not over. I digress. AI news like this feels like fear propaganda.
This downward revision caused the major market indices to begin climbing this morning, I would believe, despite the government shutdown beginning.
It seems like the market is drunk on the idea of rate cuts. Bad employment data used to mean recession coming, market down. Now bad employment data means better chance of rate cut, market up. When the correction finally happens it could be epic.
Except the data isn’t that bad. Employment is essentially full. The UE rate for 25-54 year olds is 3.6%. It was 3% in 2022, so it has barely inched up.
I meant to add that the labor participation rate for the same group is 83.7. It was 83.9 last year but around 82 in 2022. Then it flattened for about a year, then accelerated.
Problem is that rate cuts don’t really help employment right now so it’s not like the Fed is doing much for the dual mandate by cutting rates when inflation is already still too high. It’s not just the current inflation, it’s that inflation on top of all the other inflation putting pressure on prices Americans already struggling to afford, and the tariffs effects still haven’t really come in yet because of the huge inventory build companies did to avoid the tariffs hit, now finally drawing down to make tariffs effect felt more.
Remember the yields on Treasuries went UP when the Fed cut last month, not down, which isn’t helping the job market at all. Again there is no free lunch here and it’s not smart to cut rates when inflation is still too high and built on an earlier heavy hit from inflation. The Fed cutting now is just delusional, and would cause the worst of both of the worlds for Americans. Little help for employment but even worse inflation and higher prices, which means less in savings and less to spend. We’re already seeing even higher delinquency rates for all kinds of debt, not just student debt but now for cards and especially cars, and who knows how much for BNPL. Raising prices even more is the last thing American consumers need. The causes of weakening employment and jobs problems are more structural and from bad policy, that’s were fixes have to be. When it comes to Fed interest rates, far better now to take a page from Volcker and pop the housing bubble and asset bubbles at heart of this.
Yep. It’s more that the market is thinking “The data is just bad enough to give the Fed cover to what it wants to do anyway, which is drop rates to 0 (ultimately), but not so bad that there’ll even be close to a recession.”
Not saying I agree, but that’s the narrative.
I agree. I’m still of the opinion that the FED should have ignored the employment mandate and shut off inflation almost immediately. Inflation is just a killer for everyone, job or no job.
I’d rather people have jobs and have some inflation than people not have jobs and have no inflation.
The point of the dual mandate is that it’s reasonable to care about both and that they often go against each other. It’s also why the misery index is defined as the sum of unemployment and inflation. Most reasonable people want a balance.
In case anyone’s wondering, the lowest sustained misery index was from 2014-2020, when it hovered between 5 and 6%.
It now stands at 7% where it has been since Nov 2023. This is still one of the lowest values since 1970.
numbers, I 100% disagree with the false dichotomy you’re presenting. Yes, if the choices were 3% inflation (on top of 20-25% over the past 5 years) and 4% unemployment, or low inflation and 20% unemployment, I’d agree.
But I don’t think letting the employment rate climb to 5-6% would be the biggest deal in the world.
And I don’t think letting the inflation rate climb to 4% is the worst thing in the world.
The Phillips curve describes the magnitude of the tradeoff between unemployment and inflation. Unfortunately, while it was once thought the Phillips curve was relatively constant, the post-Great Recession world showed that it can and does vary. So the problem is that no one can know exactly the magnitude of the tradeoff.
We know we can get 3% inflation and 4% unemployment (both quite low in the history of the US) because we are there now. We don’t know how much unemployment will rise if we try to push inflation down any further. There is a fair amount of evidence that the Phillips curve has gotten a lot flatter recently, meaning that you need big changes to unemployment to get small changes in inflation.
I don’t agree. I think 3% inflation is already awful, and 4% is even worse. 4% means the value of the dollar is halved in under 20 years. I think the economy has been overheated for years, and we still have way too many zombie companies. Allowing unemployment to rise, and for these to be flushed out would be much better in the long run.
Yes, and allowing unemployment to rise 2 percentage points means that 3 million people now don’t have jobs. That’s a lot of people.
Isn’t 3% to 3.6% a 20% increase?
Sure but who cares. If unemployment increased from 0.01% to 0.1% that would be a 1000% increase but we’d all still be ecstatic.
Do I really need to explain the law of small numbers to you?
This is like those studies where something doubles in risk but the initial risk was like .1%
You people are ridiculous.
Again the Labor force participation rate for this group is basically unchanged with ups and downs since 2023. It is actually exactly where it was throughout most of the 90s
Numbers said it nicer than me lol
Numbers great comments above. Thanks for the information. I learned something today.
Thanks, Eric! Great to hear you appreciated them and great to have someone who wants to have a good conversation about these things. Makes this place an even better read!
But it’s not .01.
Your point is true, and it’s one that I have made different circumstances any number of times over the years, but there’s also the point that velocity of a change is meaningful when put in contact with magnitude.
As long as we’re picking segments of the population to look at, here are some statistics from the St. Louis Fed about the July report:
“ Young college graduates between ages 23 and 27 are experiencing unemployment rates that average 4.59% in 2025—a stark contrast to the 3.25% rate this same demographic experienced in 2019.1
This 1.34 percentage point increase represents more than just a statistical noise; it reflects a significant shift in how the economy is absorbing newly educated workers”
Not saying one point of view is more valid than the other, just that it kind of depends on personal experience and which conversation you’re trying to have.
When you want to check context, you look at a graph. And here’s what you see:
1. It increased from 3 to 3.6% from July 2023-July 2024, and has been flat ever since, so currently the velocity is zero.
2. It has been lower than 3.6% for only a very few short periods of time in history (2017-2020, 1998-2001, 1972-1974, 1964-1970, 1955-1957, 1950-1953), i.e. about 20 years out of the last 75.
You mean a graph like the one that was posted in another article which showed that since the 1950s, every time the unemployment rate has increased in the way that it has recently, it has continued up to much higher increases?
But back to percentages… you would agree that a 6% unemployment rate is higher than most people would like, right?
So with that same .6% increase, to 6.6%, that would be a 10% increase, right?
10% still seems meaningful.
Caused might be a little strong.
Looking at futures markets bottomed around 4am est and began climbing from there and then slightly pulled back after this report before chopping around and then trending higher again.
More clear now then ever that markets aren’t rational especially stonks
Yep Buffett indicator of 221%. Schiller PE of over 40. Regular PE of 31. Markets are as bubblicious and overvalued as they were in 2000.
US Supreme Court defies Trump with huge move on Fed governor
The US Supreme Court has rebuked President Donald Trump for attempting to fire Federal Reserve Governor Lisa Cook.
That’s not even close to an accurate representation of what the court said or did.
That is so fucking misleading and you know it. It said she can stay until the heading in January. Why do you post click bait headline bs?
I mean, the “cause” for which he tried to fire her turned out to be made up, so now we’re just arguing about whether Trump can explicitly break black letter law.
Irrelevant. The Supreme Court did not rule on the merits (whatever they happen to be). They just let her stay in the position while the case plays out. That isn’t a “rebuke” or “defiance” by any reasonable definitions of those words.
Agreed on that count.
That isn’t what the headline said though. It is click baity bs and headlines should be ignored.
This is why people hate the media and why misinformation is wide spread.
Covering all politics like a game or a horse race where the only thing that matters is who won or lost and not what the decision actually means has been absolutely toxic for our understanding and ability to trust media.
All of the recent Trump SCOTUS rulings are similar, the real SCOTUS merit hearings are far off.
MW: Dow and S&P 500 head for record closes, even as government enters shutdown and ADP reports jobs decline
Wolf, how does deportation affect ADP data? My question is if people with jobs getting paid with adp payroll under some migrant #ITIN number get deported they fall off the monthly payroll. Correct? So if 5k per month with migrant #itin get deported who had paying jobs in the surveyed data would be 5k fewer workers worked, correct? Or wrong? Please provide clarification if you know the answer. I know migrants work under migrant #itins without proper green cards and or h1 visa, I understand it’s easy to get a migrant #itin number without work authorization, technically they could be be counted and are paying our taxes for us without representation or benefits to themselves except income for time traded. Is ADP data flawed due to deportations?
Deported or self-deported… or just going on a long trip: If a person on a company’s payroll leaves the country and no longer works for the company, that company’s payroll drops by 1.
The company then tries to hire a replacement, but while the company is looking, there’s 1 job opening and 1 less job.
Everyone acknowledges — including the Fed — that the crackdown on illegal immigration is having some effect on various aspects of the employment data. It reduces the supply of labor, it may reduce employment, it may keep unemployment rates relatively low, etc. We’ve been talking about this for months. Here is one of them, from August 1:
https://wolfstreet.com/2025/08/01/slowing-supply-of-immigrant-labor-slowing-job-growth-keep-the-job-market-in-balance-and-the-unemployment-rate-low-push-up-wages-powell-talked-about-it/
I work in Big tech and I can tell you job market is extremely rough.
This is the first time where big Techs have historically high valuations and profit but they have laid off 100s of thousands of people so far.
AI would be a big job killer from what I see since I do work in AI a lot.
Yeah have some old friends also working in tech and layoffs have been awful, but doubtful AI is the reason. Put up comment above on it, talking to companies and users of AI. It’s just way too unreliable, often makes software and data even worse with all the good-appearing text and coding that turns out to be wrong, and then the hallucinations and collapse of the models as the AI starts to train on and spew back out it’s own messy regurgitated content. The ROI just isn’t there for these big GPU purchases or data center investments. So the reason for the job losses in tech has very little do with AI. It’s more off-shoring and just cost-cutting to try show a profit. At least for next couple quarters, after that it’s the next guy’s problem.
The AI hype will remain until reality sets in, now the chip makers are investing cash into data centers to feed their demand, it’s so circular and obvious, pathetic really as all this capacity will be utilized maybe 10 percent at most with years of excess capacity…large language models will NEVER be artificial intelligence…the occult trying to find their fountain of youth, so desperate to upload their mortality to live forever, trying to be creation itself…and of course they are grifters….
The AI hype will remain as long as suckers keep clicking on the links.
Honestly I think the fed should have let rates remain or even increased them to cut this AI hype train off at the knees. LLMs have their uses but as stated by others they have been vastly oversold. Much of the gains have already been realized, and the massive investments will likely go the way of Meta’s “Metaverse” underutilized and ultimately abandoned.
People fail to see the two problems with AI, time and time again:
1 Power is not free and is a variable cost.
2. Data centers and the hardware that power then are not free and are a variable cost.
3 Human content that feeds them is quickly becoming not free, therefore will be a variable cost and a growing one at that.
4. Humans that maintain and tweak the models are definitely not free and expensive talent. Another variable cost.
5 Each query costs money and compute power. Yet another variable costs.
Labor is itself a variable cost. So what are we actually doing here with AI? It’s certainly not saving in variable costs. It’s a shell game, clear and simple. And slowly but surely investors are picking up on this. It’s not going to be pretty.
Jon,perhaps those workers should “learn to code”?!
I will be interested to see what long term/full time jobs AI will provide(or not provide).
We might only be seeing the first signs but entry level positions(new college graduates) are starting to struggle. If that is combined with people retiring later that may not impact the overall numbers but certainly has significant societal ramifications. Sort of the expression is that if one person is killed it is a murder, if 20,000 people( or whatever) then it is a statistic. The real world implications is most important.
But again the unemployment rate for 16-24 year olds is 10.5, with labor force participation at 70 for 20-24 year olds.
The labor force participation for this group has been like +/- 2 from this number since 2011.
The unemployment for 16-24 is basically where it was in 2016. Sure it is up from the low of like 7.3 in 2023, but 10.3 is still pretty damn low historically speaking.
We basically are seeing headlines of this stuff with no historical context
All the AI comments crack me up. Humans are innovators and will not sit idly by for automation to rue the world. I believe AI will allow us to work smarter and not harder. Yes, it will take some jobs and some humans will have to adapt. Just like automobiles to horses. Computers to paper. Email to snail mail. AI will have opportunities too. No, I have no investment in AI. Just my 401K/Pension does. I believe my fellow humans will do the right thing. I choose faith instead of fear.
If I see anything that smells of AI, I stop using it immediately. AI is legalized piracy stealing legitimate content and pawning it off as one’s own. People need to fight back against it and boycott any company using it.
This quantum computing, perhaps hyperbole, could jettison those giant data centers and crypto processing facilities.
Watch out y’all get fired for reporting bad news!