I Timed Every AI Task For a Week: Here's What It Saved (and Cost) My Small Business

Bottom line: AI saved me 6 hours and 40 minutes in one working week, but it also quietly cost me 2 hours and 15 minutes I hadn't budgeted for, mostly on editing and fact-checking things it got confidently wrong. Net gain, yes. Life-changing, no. And it saved the most time on the tasks that mattered least.

Worth reading next: Why ChatGPT Won't Recommend Your Business (And How To Fix That In 2026.

Worth reading next: I Let AI Write My Sales Emails for 90 Days. Here's What It Did to My L.

I keep reading that AI gives small business owners "back ten hours a week." Nobody ever shows the timesheet. So I made one. For five working days I logged, to the minute, every task where I used AI, how long it took, and how long the same task would have taken me without it. Old-fashioned stopwatch, a Google Sheet, no cherry-picking.

This is the actual data, warts and all, including the bit where AI made things slower.

How I ran the experiment

Nothing fancy. I run a small consultancy, so my week is a normal mix of client emails, social content, a bit of research, invoicing, meeting notes, and one blog draft. For every single task I did that touched AI (mostly ChatGPT and Claude for writing and research, Otter.ai for transcription, and Descript for a short video edit), I noted three things: task, time taken with AI including my editing, and my honest best estimate of how long it would've taken me to do it the old way, based on years of doing exactly that task without help. I'm not pretending that's a perfect scientific method. It's better than the vague "AI saves hours" claims everyone repeats without ever writing down a single number.

The tasks that saved time

• Meeting notes and action points from a 45-minute client call: 6 minutes with Otter.ai plus a quick AI clean-up, versus roughly 25 minutes of me typing from memory and half-listened-to notes. Saved: 19 minutes, once.
• First-draft LinkedIn captions (5 posts): 12 minutes total with AI drafting from my bullet points, versus about 45 minutes writing from scratch. Saved: 33 minutes.
• Research summary for a client pitch (competitor pricing across 8 sites): 20 minutes with AI pulling and summarising, versus an estimated 90 minutes of me tabbing through websites. Saved: 70 minutes.
• Rewriting a stiff internal email into something a human would send: 3 minutes, versus 10 minutes of me fussing over tone. Saved: 7 minutes.

Add those up across the week and you get real, measurable time back. That's the part every AI newsletter loves to quote.

The tasks where AI cost me time

Here's the bit nobody puts in the case study. Two tasks came out worse with AI once I counted my editing time.

• The blog first draft (a 1,800-word piece for a client): AI produced a draft in 4 minutes. I then spent 55 minutes rewriting most of it, because it flattened the client's voice into generic LinkedIn-guru sludge and invented a statistic about email open rates that didn't exist anywhere. Writing that piece from scratch myself, based on past timing for similar posts, would have taken around 50 minutes. So AI added almost 9 minutes to a task it was supposed to speed up, once you count the fact-check.
• An invoice-chasing email to a client who was 40 days overdue: AI's version was polite to the point of apologetic, which is exactly wrong when someone owes you money. I rewrote it firmer, which took longer than if I'd just written it myself in the first place, because I was reacting to a draft instead of starting with my own instinct.

That second one is the uncomfortable truth most people writing about AI and small business skip over: AI is fastest exactly where the stakes are lowest, and slowest, or actively harmful, exactly where tone, judgement, and accuracy matter most. It will happily draft you five identical-sounding social captions in ten minutes. It will also happily hand you a confident, wrong number for a client pitch, and you won't catch it unless you already knew the real figure. The time you save on the easy stuff gets eaten by the vigilance you now need on everything else.

The full weekly total

Time saved across the week: 6 hours 40 minutes.
Time cost across the week (extra editing, fact-checking, rewriting tone): 2 hours 15 minutes.
Net time saved: 4 hours 25 minutes.

That's roughly one working morning back. useful, worth having. Not the "get ten hours back" pitch you see in every AI course sales page. If your business runs mostly on quick, low-stakes, repeatable writing tasks, your ratio will be better than mine. If it runs on relationships, tone, and trust, like most consultancies and service businesses do, your ratio will look closer to mine or worse. I think about this the way Howard Schultz talked about scaling Starbucks: speed is only worth having if it doesn't cost you the thing customers came for. He was obsessive about not letting growth dilute the coffee experience. AI is the same trade. It'll happily let you produce more content, faster, right up until the content stops sounding like you and starts sounding like everyone else's AI output.

Where the time went (and where it should have)

The biggest lesson from the log wasn't the totals, it was the pattern. AI earned its keep on anything mechanical: transcribing, summarising, first-pass research, reformatting. It lost time or added risk on anything that needed judgement: tone with an unhappy client, a statistic that would go on a public page, the closing line of a pitch email that needed to land exactly right for one specific person. That maps almost exactly onto what I'd tell anyone hiring help for the first time, whether that's a freelancer or an AI consultant: hand over the mechanical stuff freely, keep the judgement calls, and never let a tool write the sentence where money or a relationship is on the line without reading it twice yourself.

The habit that made the difference

I run this kind of tracking exercise the way I'd recommend anyone test a new tool or process, and it's the same discipline I picked up writing about Napoleon Hill's business lessons years ago: measure the thing before you believe the story about the thing. Hill's whole argument was that vague belief without a plan and a record gets you nowhere. Most small business owners adopt AI because someone told them to, not because they timed it. A week of honest logging costs you almost nothing and tells you exactly where the tool earns its subscription and where it's just adding a layer of editing you didn't have before.

How to run this experiment yourself

1. Pick one working week, not a cherry-picked good day. Include your normal mix of easy tasks and awkward ones, like the overdue invoice email or the client who's annoyed about something.
2. Log every AI task in real time, not from memory at the end of the week. Use a simple spreadsheet with four columns: task, time with AI, honest estimate of time without it, and a note on quality (did you have to rewrite it, and how much).
3. Be brutal about editing time. If you spent 20 minutes fixing an AI draft, that's 20 minutes of the task, not a separate "quick tidy up" you conveniently forget to count.
4. Separate mechanical tasks from judgement tasks before you total anything up. You'll almost certainly find AI wins clearly on one pile and loses or breaks even on the other.
5. Total both sides at the end of the week and work out your real net time saved, not the number you'd like to tell people at a networking event.

If you want a version of this that your own audience can run themselves rather than just reading your results, this is exactly the kind of thing that works well as an interactive calculator on your site, where someone plugs in their own task times and gets their own honest number instead of trusting yours. I've had more engagement from tools like that than from most blog posts I've written, because people trust a number they generated themselves far more than one I handed them.

What I'm changing because of this

Three things, starting this week. First, AI stays firmly on transcription, first-pass research, and reformatting, the tasks where it won clearly and consistently. Second, anything client-facing where tone matters (chasing money, apologising, closing a deal) gets written by me first, with AI only used afterward to check grammar, never tone. Third, any statistic or fact an AI tool hands me for a client-facing piece gets verified against a real source before it goes anywhere near a page, the same discipline good content teams like Ahrefs apply to their own content, where every claim is checked because their whole brand depends on being trusted with data. None of this means AI isn't worth using. Four hours and twenty-five minutes back in one week, done honestly, adds up to real money over a year. It just means the "ten hours a week back" claims you see everywhere are either measuring the easy tasks only, or not measuring the editing time at all. Track your own week before you believe anyone else's number, including mine.

Frequently asked questions

How much time does AI save a small business owner per week?

Based on my own tracked week, the net figure was 4 hours 25 minutes, after subtracting extra editing and fact-checking time. That's after 6 hours 40 minutes of raw savings on mechanical tasks like transcription and first drafts, minus 2 hours 15 minutes lost rewriting tone and checking facts on client-facing work. Your own ratio depends heavily on how much of your work is mechanical versus how much depends on tone and judgement.

Which tasks does AI speed up for a small business?

Transcription and meeting notes, first-pass research and summarising, reformatting existing content, and rough drafts of low-stakes writing like internal emails and social captions. These are repeatable, mechanical, and low risk if the output isn't perfect.

Where does AI slow small businesses down?

Anywhere tone or accuracy really matters: chasing overdue payments, apologising to an unhappy client, writing public-facing content with statistics attached, or closing a pitch. In each of these, the editing and fact-checking time can equal or exceed the time it would have taken to write the thing yourself from the start.

Should I track my own AI time savings before believing marketing claims about it?

Yes. Run one honest week, log real minutes for every AI task including your editing time, and compare that against your normal time for the same tasks. Most claims of "ten hours a week saved" don't count editing time at all, which is exactly where the real cost of using AI shows up.

Related reading: Beyond Meat Marketing Strategy: How They Built a Brand That Wins and LinkedIn Marketing Strategy: How They Built a Brand That Wins.