Straight answer: Over 30 days, AI saved me roughly 11.5 hours a week on drafting, notes, and admin, but checking and fixing that output ate back about 4.5 hours, so my real weekly gain was closer to 7 hours, not the “AI does everything now” number you see in most posts. The tools matter less than the discipline of knowing which tasks you can trust AI with and which ones will cost you more time than they save.
Worth reading next: I Tracked Every Minute AI Saved My Business for 30 Days. The Real Numb.
Worth reading next: One Client Win, Thirty Days of Content: The AI System That Stopped Me .
Why I bothered tracking this at all
I get asked “how much time does AI save you” at almost every talk I do now. For two years my answer was a shrug and a guess. “Loads.” “A few hours a week.” Vague, and I don’t like vague. So in October I built a simple spreadsheet, tagged every task I did for 30 days as AI-assisted or manual, and timed both. Not a survey of other people. My own hours, my own business, my own clock.
I run a solo-plus-freelancers consultancy. I write, I speak, I advise clients on marketing and now AI, and I still do my own social posts and a chunk of my own admin. So the numbers below are one real business, not an average pulled from a vendor’s white paper.
The method: four columns, nothing clever
I didn’t use fancy time-tracking software. Four columns in a spreadsheet, filled in at the end of each work session:
- Task name (specific: “LinkedIn caption for Tuesday post”, not “social media”)
- Time it would have taken me without AI, based on my own historical average for that exact task type
- Actual time with AI, start to finish including the prompt
- Time spent fixing or checking the output afterward, timed separately
That last column is the one almost nobody tracks, and it’s the one that changes the whole story.
The tasks and the real numbers
Here’s what 30 days of logging showed, broken down by task type, averaged per week:
- Content drafting (blog outlines, email newsletters, first-pass client reports): saved 4.2 hours a week. This is where AI earns its keep. A first draft that used to take me 90 minutes now takes 20, even accounting for edits.
- Social captions and short-form copy: saved 1.8 hours. Real, but smaller than people expect, because I still rewrite most captions to sound like me and not like a template.
- Email replies (client questions, scheduling, routine follow-ups): saved 2.5 hours. Otter and ChatGPT drafting replies from my bullet points is one of the best uses I’ve found.
- Meeting notes and summaries: saved 1.5 hours. Otter.ai transcribes and summarises calls automatically, which used to be 20 minutes of typing after every client call.
- Research and first-pass fact gathering: saved 1.5 hours, but with the biggest asterisk of the lot, which I’ll come back to.
Add it up and you get 11.5 hours a week of time apparently saved. That’s the number that goes on the slide in a keynote and gets a round of applause. It’s also the number that made me suspicious enough to keep tracking the fourth column.
The hidden 4.5 hours nobody puts in the case study
Here’s the uncomfortable bit. When I added up time spent checking facts, rewriting tone, fixing wrong names, and correcting research that AI had confidently made up, it came to 4.5 hours a week. Not zero. Not a rounding error. Nearly 40 percent of the “saved” time came straight back out of my pocket.
The worst single incident: a batch of 40 outreach emails for a guest posting push, where I’d asked ChatGPT to personalise the opening line for each editor based on their site. Twelve of the forty referenced articles that didn’t exist, or attributed a piece to the wrong writer entirely, including one editor I’ve worked with for years who got credited with a post she’d never written. I only caught it because I know that particular relationship well enough to spot it was wrong. If I’d sent that batch as drafted, I’d have looked careless to twelve editors I care about, which is a worse outcome than if I’d never used AI on that batch at all. That’s the piece nobody mentions when they tell you AI writes your outreach for you: it writes it fast, and confidently, and sometimes wrong, and speed plus confidence is a dangerous combination if you’re not the one who already knows the subject well enough to catch the error.
That’s the pattern across all my numbers, not just the outreach disaster. AI saves the most time on tasks where I already know what good looks like, because I can check the output in seconds. It costs me time on tasks where I’m relying on it to know things I don’t, because then I have to go and verify from scratch anyway, which is slower than if I’d just done the research myself in the first place. I wrote about this same trap when I covered what happened when I published 569 AI-assisted blog posts in 45 days: volume without judgment isn’t a shortcut, it’s a bigger pile to check.
The three-tier system I built after the 30 days
Once I saw the real number, I stopped treating “use AI” as a single decision and split every task into three tiers.
- Tier one, never AI first: anything involving a real person’s name, a client fact, a number I’ll be quoted on, or a legal or financial claim. I write these myself, then I might ask AI to tighten the phrasing after.
- Tier two, AI drafts, I verify every fact: outreach emails, client reports, research summaries. AI does the first pass, but I treat every named fact, date, or attribution as unverified until I’ve checked it against a source I trust.
- Tier three, AI does it and I barely look: meeting summaries from a call I was on, caption variations of something I already wrote, formatting and rephrasing tasks with no new facts involved.
That sorting took me about two hours to do once, for every recurring task type in my business, and it’s the single change that turned an 11.5 hour gross saving into a genuine 7 hour net one, because I stopped wasting the check-time on tier three tasks that never needed it and stopped skipping the check-time on tier two tasks that always did.
What this looks like week to week now
A normal week for me now: Otter sits on every client call and I get a summary in my inbox before the call has even ended, no typing required, tier three, barely checked. My newsletter draft comes from a ChatGPT first pass based on a voice note I record on my walk, tier two, I read every line before it goes out because my name is on it. My outreach for guest posts gets a first-pass personalised line, then I manually verify every article title and author name against the actual site before sending, because of the twelve wrong ones I caught before. Canva’s AI features resize and vary my social graphics across platforms in minutes rather than the half hour that used to take, tier three, quick visual check only.
None of this needed expensive software. Otter, ChatGPT, and Canva together cost me under £40 a month combined. The saving isn’t in the tool, it’s in the sorting.
Where a proper setup is worth paying for
If you’re doing this alone, the spreadsheet method above will get you real numbers within two weeks, not thirty days, because patterns show up fast once you’re tracking. But if you’re running a team of five or more and everyone’s doing their own version of tier sorting in their head, inconsistently, that’s usually the point where it’s worth bringing in outside help to build a proper system rather than five different personal workarounds. I’ve seen small businesses waste more time reconciling everyone’s different AI habits than they ever saved from the tools themselves. That’s exactly the gap a good AI implementation coach closes, not by teaching people prompts, but by building the tier system for the whole team so everyone’s checking the same things and skipping the same things, consistently.
If you’re not technical and the whole idea of “building a system” sounds like more admin than it’s worth, start smaller. I’ve written before about how I use AI every day without being technical, and the honest starting point is exactly what I did here: pick one task, time it for a week, decide if it’s a tier one, two, or three, and only then add the next task.
The uncomfortable truth in one sentence
AI doesn’t remove work from your day, it moves work from doing to checking, and whether that’s a good trade depends entirely on whether checking is faster than doing was, which for a genuine expert in their own field it usually is, and for someone leaning on AI to cover a gap in their own knowledge, it usually isn’t.
That’s not an argument against using it. I use it every single day and I’m not going back. It’s an argument against the version of the story where AI saves you 20 hours a week and costs you nothing, because that version has never once matched what my spreadsheet showed after four weeks of honest logging. The same discipline that turned my website’s traffic around after years of neglect, which I wrote about in the month my website started paying me again, is the same discipline that makes AI worth the subscription: track it, don’t just trust it.
Tools worth trying if you want to run your own version
You don’t need a long list of tools to run this experiment. Otter.ai for call notes, ChatGPT or Claude for drafting, Canva for visual variations, and a plain spreadsheet for the tracking is enough to get a real number within your first two weeks. If you want to see what else is out there beyond the obvious names, I keep adding to a running list of lesser-known websites worth knowing about, several of which are small, free AI utilities most business owners have never heard of.
Frequently asked questions
How much time does AI save a small business per week?
In my own 30-day tracked test, AI saved roughly 11.5 hours a week on content, email, notes, and research before accounting for the time spent checking the output. After factoring in fixing and verifying, the real net saving was closer to 7 hours a week. Your number will depend heavily on how disciplined you are about which tasks you trust AI with.
Which AI tasks save small business owners the most time?
Tasks where you already know what a good result looks like save the most time, because you can check them in seconds. Meeting note summaries, first-draft emails you’ll edit anyway, and repurposing content you’ve already written into different formats are the biggest genuine wins. Research and fact-based writing save far less real time because you still have to verify everything.
Does using AI save money as well as time?
Usually yes, but only once you account for the checking time honestly. A £30 a month AI subscription that saves you 7 real hours a week is a strong return if your time is worth £30 an hour or more, which for most small business owners it is. It stops being a good deal the moment nobody is checking the output and mistakes start reaching clients.
How do I track whether AI is saving my business time or just moving the work around?
Log four things for every AI-assisted task for two to four weeks: the task name, your normal time without AI, the actual time with AI, and a separate timed entry for checking or fixing the output. Add the fourth column up separately. That’s the number most people never track, and it’s usually the one that tells you the truth.
Related reading: I Cut Client Onboarding From Three Weeks to Three Days Using AI. Here’s What Broke Along the Way and I Let AI Write My Invoice Chase Emails for 90 Days. Here’s What Happened to My Cash Flow.
