How many hours a week do you lose to admin?

The most recent number for the UK is 11 hours a week. That is what 1,000 owners of micro, small and medium businesses told American Express and Small Business Saturday UK in July 2026 they spend on administrative or finance-related tasks — roughly six working days a month, against just 3.6 days a month on sales and business development (Amex SME Barometer, July 2026).

Nearly twice as much time keeping the business running as growing it. That is the headline; the useful part sits underneath it, because "admin" turns out to be several different problems wearing the same coat.

How many hours a week do people actually lose to admin?

It depends on what you count. Narrow definitions — paperwork, invoicing, bookkeeping — land around 11 hours a week. Include meetings, messages and the time spent switching between them and the published measurements get close to half the working week.

Why does admin feel like it takes more time than it does?

Because it does not arrive in blocks. It arrives as interruptions, and interruptions cost more than the task itself. Microsoft's analysis of Microsoft 365 usage signals found people are interrupted roughly every two minutes during core working hours by a meeting, an email or a notification.

The same analysis found 57% of meetings are ad hoc — no invite, no warning — and that meetings after 8pm have risen 16% year on year. That is the mechanism behind the feeling. Ten hours of admin delivered in two focused mornings is annoying. The same ten hours delivered in 90-second slices, all day, every day, is what makes people say they never got anything done.

What does switching between apps actually cost?

About four hours a week. Researchers writing in Harvard Business Review instrumented 137 people across three Fortune 500 companies for up to five weeks and found they toggled between applications roughly 1,200 times a day, adding up to just under four hours a week simply reorienting themselves — around 9% of their time at work.

The cost is not pressing the keys to switch. It is the time it takes to work out where you were, what you were doing, and why.

Harvard Business Review, on the cost of application toggling

Which is why "use fewer tools" beats most tool recommendations, and why bolting an AI app onto each of your ten existing apps can make things worse.

Can you trust these statistics?

Partly. Three limitations are worth knowing before you quote any of them at yourself:

• Self-report inflates. Comparing survey estimates against next-day time diaries, researchers writing in the US Bureau of Labor Statistics' Monthly Labor Review found people overestimate their working hours by 5 to 10 percent — and the overestimation is worst among those claiming the longest weeks (Robinson et al., June 2011).
• Most of it is corporate. The Microsoft, Asana and HBR figures come from knowledge workers inside larger organisations. If you are a sole trader, your meeting load is lower and your admin load is higher, because nobody else is doing the invoicing.
• Vendors funded a lot of it. Asana sells work management software; Microsoft sells Copilot. That does not make the measurements wrong — Microsoft's come from usage telemetry, not a survey — but it explains which numbers get published.

None of that makes the research useless. It makes it a starting hypothesis. The only figure that matters is your own.

How do you measure how many hours you lose to admin?

Track five working days in 15-minute blocks, sort everything into four buckets, and multiply. It takes about 20 minutes of total effort spread across the week, and it will tell you more than every survey in this article combined.

1. Pick a normal week. Not your quietest, not your worst. If the week gets hijacked, start again — one distorted week will send you after the wrong problem.
2. Keep one running note. Every time you change task, write the time and two words. Nothing more elaborate survives contact with a real day.
3. Sort into four buckets at the end of each day. Paid work (the thing clients pay for), admin (invoicing, scheduling, filing, chasing), comms (email, messages, calls), and nothing useful (the reorienting, the re-reading, the tab you opened and forgot).
4. Total the week, then halve your instinct. Given the 5–10% overestimation effect, if your tally disagrees with your gut, trust the tally.
5. Rank the admin bucket by how repetitive each item is, not by how much you dislike it. Repetitive and boring is the profile of a task worth handing over; unpleasant but varied usually is not.

How many of those hours can AI realistically give back?

About 2.2 hours a week for a typical user, on the best current evidence. Economists at the Federal Reserve Bank of St. Louis, using a nationally representative US survey, found that people who used generative AI at work reported time savings of 5.4% of their working hours — roughly 2.2 hours in a 40-hour week (Bick, Blandin and Deming, February 2025).

That average is dragged down by casual use. The same research programme found the share of all US work hours spent using generative AI rose from 4.1% in November 2024 to only 5.7% in August 2025 — most people are dipping in, not restructuring anything. Atlassian's 2026 survey of 12,035 knowledge workers found the same shape: 85% use AI at work, but only 29% have embedded it into how their work actually runs (State of Teams 2026).

That gap between using AI and embedding it is where the difference between two hours a week and something meaningfully larger sits. It is not a tool problem. It is a question of which recurring task you point it at.

[Steve — add a real client example here: the task, the frequency, the before-and-after minutes.]

Which admin should you hand over first?

The most frequent, most repetitive, least judgement-dependent task on your list — regardless of how small it looks. Frequency compounds; size does not.

• Anything you write more than five times a week in roughly the same shape: quotes, follow-ups, appointment confirmations, the same three answers to the same three questions.
• Meeting notes and the actions that fall out of them. The most common first win, because it is high-frequency, low-judgement, and you were doing it badly anyway.
• Triage, not replies. Sorting an inbox into "needs you" and "does not" is smaller and safer than drafting responses, and removes more interruption per hour saved.
• Nothing that requires being right. Anywhere a plausible-sounding error costs you money or a client is the wrong place to start, whatever the tool promises.

For the longer version of that filtering logic, how I choose AI tools for a business sets out the criteria, and is AI worth it for a small business? covers whether the saved hours justify the subscriptions.