Where AI genuinely saves time in a working week.
Forget "10x". Here is the honest ledger: the five places the minutes actually come from, what a realistic week looks like, and the four ways AI quietly costs time instead.
Productivity claims about AI come in two dishonest flavours: the demo (a report written in 30 seconds, never mind that checking it takes an hour) and the dismissal (it is all hype, never mind the people quietly finishing at five). The truth is an accounting question, and accounting wants line items.
How much time does AI actually save at work?
On our own modelling, for an office worker with a settled habit, three and a half to five hours a week, built from email, meeting notes, recurring first drafts, spreadsheet work and reading long documents. The honest unit is 15 to 40 minutes per task, with checking time counted in.
On our own modelling, the unit of AI time-saving is 15 to 40 minutes per task. Ten-times claims describe demos, not weeks. Nothing about that is dramatic - until you notice how many times a week those tasks recur. Here is where the minutes are, for a typical office week, assuming you already have the briefing habit from how to use AI at work, which is the method every figure below depends on, and always counting checking time inside the total.
Where do the saved hours actually come from?
Email: the difficult ones, not the volume.
The saving is not typing speed - it is the 15 minutes of drafting-and-deleting on each awkward message, and the messages you stop postponing. Two or three difficult emails a week plus daily thread-untangling is a solid 60-90 minutes.
Meetings: the minutes.
Twenty to thirty minutes of minute-writing per meeting becomes five minutes of checking - a net 15 to 25 minutes each time. For someone in the minutes seat twice a week: 30-50 minutes. Pre-reads are a genuine saving too, but they sit in line five rather than here; counting them in both places is how AI time claims start to inflate.
First drafts of recurring documents.
Status updates, briefs, announcements, job adverts - anything with a known shape. A 25-minute blank page becomes an 8-minute edit. Three recurring documents a week: about 50 minutes, plus the unmeasurable one: the dread is gone.
Spreadsheet formulas and cleaning.
The formula you would have googled for twenty minutes arrives explained in one; the messy export gets cleaned with flags instead of an evening. Bursty rather than daily - call it 30-60 minutes in an ordinary week.
Reading long things you only need part of.
Policies, reports, tenders, terms, and the long pre-read before a meeting: summarised for your question, with the exact sentence quoted back for anything that matters, so you can search the document and see it in context. Ask for quotes, never page numbers - page references are among the details assistants invent most freely. Two long documents a week: 40-60 minutes, and you walk into the meeting having actually engaged with the content.
The addition, so you can argue with it: 60-90 minutes on email, 30-50 on minutes, about 50 on recurring documents, 30-60 on spreadsheets and 40-60 on long reading. That is 210 minutes at the low end and 310 at the high end - three and a half to a little over five hours. Roughly half a working day, every week. Not ten times anything - and, compounded over a year, still the most valuable habit you will build at a desk.
Why is AI not saving you any time?
- Novelty use. Tasks you do twice a year have no learning curve to amortise. The brief takes longer than the task.
- Re-briefing from scratch. If you type context fresh every time, you pay the setup cost daily. Saved prompts - a short library you refine - are where the compounding lives.
- Thin briefs, heavy rework. An output that needs 20 minutes of fixing because the brief lacked context and standards is slower than writing it yourself. The fix is one line of "what good looks like", not a new tool.
- Tool tourism. Evaluating a new AI product every fortnight consumes precisely the hours the last one saved. Choose adequately once and go deep.
One more honest entry: checking is not overhead to be optimised away - it is the part that keeps the saving real. A wrong number caught in a meeting costs more time (and standing) than every minute AI ever saved you. Budget the checking; count it in the totals; never skip it because the draft "looks fine".
How do you measure the time you save?
You do not need a spreadsheet of your life. For each of your top five tasks: time it once the old way, then time the AI-plus-checking version after a week of practice, and count the difference only if the quality held. A fortnight of casual notes gives you a defensible number - and a number changes things. It tells you where to deepen the habit, it survives sceptical questions from colleagues, and it is career evidence: "I cut the monthly report from a full day to under three hours and reinvested the time in X" is the sentence that matters in the employability conversation. That one is the exceptional top end, not the weekly norm - which is exactly why it is worth being able to name.
Which raises the real question: what happens to the morning you get back? Time saved evaporates unless reassigned - into the judgement work, the relationship work, the backlog that never got touched, or simply leaving on time. Decide deliberately. That decision, not the tooling, is what separates people for whom AI changed the week from people for whom it changed nothing.
A last word on compounding, because it is the part the sceptics miss. The first week with a new task usually saves nothing - you are writing the brief, learning what the assistant gets wrong, building the checking reflex. The saving shows up later, once the brief has settled and the checking has got quicker - weeks rather than days, in most people's experience. Every figure above is a settled-habit number, which is exactly why one task at a time, done daily, beats a burst of enthusiasm across ten. Slow is how you reach the top of that range.
Frequently asked questions
How much time can AI actually save at work?
On our own modelling, for an office worker with an established habit, a realistic figure is three and a half to five hours a week, built from email, meeting notes, first drafts, spreadsheet work and reading long material. Claims of "10x productivity" describe demos, not weeks; the honest wins are 15 to 40 minutes per task, several times a day.
Which tasks should I use AI on to save the most time?
Start where high frequency meets first-draft work: email replies and untangling, meeting minutes, recurring documents like status updates, spreadsheet formulas and cleaning, and summarising long documents. One recurring task done daily beats ten novelty uses.
Why is AI not saving me any time?
The usual causes: using it on rare tasks instead of recurring ones, re-briefing from scratch each time instead of saving prompts, accepting outputs that need heavy rework because the brief was thin, or spending the week trying new tools instead of deepening one. Fix the habit, not the tool.
How do I measure time saved with AI?
Time the task once the old way, then time the AI-plus-checking version after a week of practice. Count the difference only if quality held. A simple note - task, minutes before, minutes after - across your top five tasks gives you a defensible weekly number in one fortnight.
The honest time ledger.
- Expect 15-40 minutes per task, several times a day - not miracles.
- Five line items carry the week: email, meetings, first drafts, spreadsheets, long reading.
- Count checking time inside every saving; a saving that skips checking is a debt.
- Kill the four leaks: novelty use, re-briefing, thin briefs, tool tourism.
- Measure once per task, then deliberately reassign the hours - saved time evaporates unless spent on purpose.
Put AI to work in your own job.
AI at Work is Clerq's practical, hype-free guide to getting real work done with AI - written for people with inboxes, deadlines and meetings, not developers. PDF and EPUB, launching soon.