Every week a founder shows me a post claiming AI saved someone twenty hours. Every week another founder tells me, quietly, that she has been using AI for a year and her weeks feel exactly as long as they did before. Both of them are telling the truth, and the gap between them is the most useful thing in the 2026 data. The Federal Reserve Bank of St. Louis has been tracking this more carefully than anyone. Its August 2026 update found that the share of employed adults using generative AI for work in the past week grew from 28.2% in late 2024 to 39.2% by mid-2026, while the share of total work hours saved grew from 1.6% to 2.2%. Adoption is racing. Time saved, across the whole economy, is walking.
This article reads the numbers the way a founder should read them. It will lay out what the credible 2025 and 2026 studies actually measured, explain why the average hides the thing that matters, show where the saved time disappears to, separate the founders who see real hours back from those who do not, and give you a two-week method to measure your own business honestly. By the end you will have a realistic expectation, a way to test it, and a clear picture of what would need to change for the hours to become yours.
What do the 2026 studies actually say about time saved?
The short answer: modest at the level of the whole workforce, meaningful for the people who use AI on the right tasks. The best studies measure both, and the difference is the point.
Start with the most rigorous source. The St. Louis Fed's Real-Time Population Survey, run by economists Alexander Bick, Adam Blandin and David Deming, asks a nationally representative sample of working-age adults how much of their work AI assisted and how much time it saved. Between the third quarter of 2024 and the second quarter of 2026, the share of work hours assisted by generative AI grew from 4.1% to 6.3%, and the share of hours saved grew from 1.6% to 2.2%. Read that carefully: even among a workforce where nearly four in ten people use AI weekly, the tool touches a small slice of the hours and saves a smaller one. This is not a failure of AI. It is what happens when a general population uses a general tool on whatever comes to mind.
Now look at small business specifically. Intuit QuickBooks surveyed more than 2,200 US businesses with up to 100 employees in April 2025. 68% said they use AI regularly, up from 48% the previous summer, and 74% of those using it said it makes them more productive. But the survey also asked about the shape of the day. Nearly a quarter, 24%, said their work days are shorter thanks to AI. 11% said their days are longer. Most of the remainder said neither. The top uses were marketing (43%), customer service (36%), administrative tasks (33%), data processing (32%) and bookkeeping (29%). So the honest summary of the small business data is this: most owners feel more productive, and about one in four has actually seen the day get shorter. That one in four is who this article is about.
Why does the average number mislead a founder?
The short answer: because the average mixes people who built something with people who opened a chat window. Your result will look like one of those groups, not the blend.
The clearest illustration comes from a study that could control for this. Microsoft's research team worked with 58 companies and 6,317 employees, randomly giving some access to Copilot and some not. At a consumer goods company, employees with access spent 31% less time reading email, a saving of about 50 minutes a week per person. At a consulting firm, people using AI spent 16% less time in meetings. Those are real numbers. But the same study reports that for many companies there was no statistically significant effect at all, and the stated reason was simply that people were not using the tool. Across the whole sample, people with access read six fewer emails a week. Among high-usage employees, that number tripled to eighteen. The average was made of a few people getting a lot and many people getting nothing.
This is the pattern I see constantly with founders. Two women buy the same subscription in the same month. One uses it to draft a proposal when she remembers to. The other sets it up so that every intake form is summarized before her discovery call, every proposal starts from her template and her past pricing, every follow-up sequence is drafted the night the call ends. Twelve months later, one of them has "tried AI." The other has a system. The average of those two women is a number that describes neither of them. When you read a headline figure, ask which group it was measured on. When you set your own expectation, decide which group you intend to be in.
Where does the saved time actually go?
The short answer: usually into more work, unless you decide otherwise in advance. Saved time is not free time until someone claims it.
There are three places the hours go, and only one of them is good. The first is expansion: the task gets faster, so you do more of it. You could write one newsletter a week in the old time, so now you write three. This is fine if three was the goal. Often it was not. The second is cleanup. Research from BetterUp and Stanford's Social Media Lab, surveying 1,150 full-time US workers, found that 40% had received what the researchers call "workslop" in the previous month: AI-generated work that looks finished but lacks the substance to move a task forward. People who received it spent an average of one hour and 56 minutes dealing with each instance. In a one-person business, you are both the sender and the receiver. Unreviewed AI output does not save time. It moves the time to later and adds interest.
The third place is the good one: the hours go back into the parts of the business only you can do. Client conversations, thinking about your offer, rest. Microsoft's 2026 Work Trend Index, a survey of 20,000 knowledge workers across ten countries, found that 66% of AI users say AI has allowed them to spend more time on high-value work. That is encouraging. But it is a statement about where people say the time went, not a guarantee about where yours will. Time saved is only saved if you had already decided what it was for. Before you automate a task, write down what the freed hour is for. If you cannot name it, the hour will be absorbed by the inbox before Friday.
Which founders actually get hours back, and what do they do differently?
The short answer: the deliberate ones, not the fast ones. The data on this is unusually clear.
The Work Trend Index identifies a group it calls Frontier Professionals, the most advanced AI users in its sample. Two behaviors separate them. 53% intentionally pause before starting work to decide what should be done by AI versus a human, compared with 33% of everyone else. And 86% of AI users overall say they treat AI output as a starting point, not a final answer. The people furthest ahead are not the ones who automate the most. They are the ones who choose the most carefully. This matches what I see in practice. The founder who gets her Fridays back did not automate fifteen things. She automated the four things that were eating her evenings, reviewed everything that came out for the first month, and left the rest alone.
There is a second pattern in the data worth noticing. In the Anthropic Economic Index from June 2026, the most common outputs of work conversations with Claude were documents and reports (20%), explanations (9%) and email drafts (7%). Those are exactly the tasks a service business founder repeats: the proposal, the summary, the follow-up. Repetition is what makes something worth systematizing. A task you do once a quarter does not need a system. A task you do every Tuesday does. The founders who see hours back are the ones who looked at their own week, found the Tuesdays, and built for those. If you want a fuller picture of what that looks like in a coaching or consulting practice, I wrote about it in Beyond the Grind: How to Run a Modern Business with AI.
What is a realistic expectation for a one-to-ten-person service business?
The short answer: I will not give you a number, and you should be suspicious of anyone who does. What I can give you is a way to think about it.
A service business is not a spreadsheet of interchangeable hours. Some of your week is delivery, which is your expertise applied in real time and largely untouched by AI. Some is connection, which should stay untouched on purpose. And some is logistics: intake, scheduling, reminders, summaries, drafts, sorting, invoicing, follow-up. The realistic ceiling for time saved is the size of that third bucket, and no larger. A founder whose week is 70% delivery has a smaller ceiling than one whose week is 40% delivery and 40% admin, no matter which tools either of them buys. This is why the Fed's 2.2% and the QuickBooks 24% are both true. They are measuring different buckets across different people.
So the useful question is not "how much time does AI save?" It is "how big is my logistics bucket, and how much of it repeats?" I build systems for a living, and I will say it plainly: the businesses where AI changes the shape of the week are the ones where the founder could describe her repeating admin in a list before we ever spoke about tools. If you cannot yet make that list, that is not a failing. It is busyness. It is also the first thing to fix, because a system built on an unclear picture of the work produces unclear results faster. Brand clarity first, intelligent systems second. The same rule applies to time.
How do you measure the time in your own business?
The short answer: track two weeks by hand, sort the hours into three buckets, and only then choose what to automate. It is unglamorous, and it works.
Here is the method. Week one and two: write it all down. Keep a note open all day. Every time you switch tasks, write the time and what you moved to. Do not judge it, do not tidy it. At the end of two weeks you have a real picture instead of a remembered one, and the two are always different. Then sort. Go through the log and mark each block D for delivery, C for connection, or L for logistics. Add up the L hours. That is your ceiling. Then find the repeats. Within the L hours, circle anything that happened more than twice. Intake summaries, proposal drafting, follow-up emails, calendar wrangling, invoice chasing, content formatting. Those repeats are the only candidates worth a system.
Now pick one. Not the biggest, the most frequent. Set it up so it runs without you at the keyboard, in your own accounts, with a written note of how it works. Run it for four weeks and review every output. Then repeat the two-week log. The difference between the two logs is your number. It is the only number that matters, and it will be yours, not a headline's. If the difference is small, you have learned something valuable before spending more. If it is large, you know exactly where to build next. This is the same audit I walk founders through when we build AI systems for women-led businesses: evidence first, then the build, and every system documented so it keeps running after I leave.
Founder to founder: the reason the headlines and your own experience disagree is not that you are using AI wrong. It is that the headlines describe a blend of people, and you are one person with one business and one specific week. The research is actually reassuring once you read it properly. A quarter of small business owners have shorter days. The ones who do are deliberate, they review what AI produces, and they built for the tasks that repeat. None of that requires speed. It requires a list. Your next step this week is to open a note and start the log.
Frequently Asked Questions
How many hours a week does AI save a small business owner?
There is no reliable single number, and the research does not support one. QuickBooks found 24% of small businesses using AI report shorter work days, while the St. Louis Fed measures 2.2% of total work hours saved across all workers.
Why am I not saving time with AI even though I use it every day?
Most likely because you are using it as a tool at the keyboard rather than a system that runs without you. Microsoft's controlled study found little effect where usage was occasional and much larger effects among frequent, structured users.
Does AI make small business owners work longer hours?
For a minority, yes. In the QuickBooks April 2025 survey, 11% of AI users said their work days got longer, usually because saved time was absorbed by more output rather than reclaimed.
What tasks save the most time when automated in a service business?
The ones that repeat every week and require no judgment: intake summaries, proposal first drafts, follow-up emails, scheduling and reminders. Frequency matters more than size.
How do I measure the time AI saves in my business?
Log two weeks of work by hand, sort hours into delivery, connection and logistics, automate one repeating logistics task, then log two weeks again. The difference is your real number.
Is the time AI saves worth the time spent reviewing its output?
Yes, if the review is built in from the start. Research on "workslop" found unreviewed AI output costs recipients nearly two hours per instance, so review is what turns AI from a cost into a saving.
The Scale Atelier builds AI systems for women-led coaching, consulting and expertise-based businesses that run in your own accounts, so the hours you save are actually yours. If you want to know where your week is going before you automate anything, book a free 20-minute audit: no pitch, just a prioritized plan that is yours to keep either way.