Using AI to run a business sounds like something a tech person says to impress other tech people. I’m a Chartered Accountant. I’ve been inside over 1,000 small businesses. I think in margins, not modules. So when I decided to spend a full month running Clever Operators almost entirely on AI, I wasn’t doing it for content. I was doing it because I needed to know whether the thing I’m selling actually works.
Here’s what I found out.
Why I did this
I wasn’t starting from scratch. I’d been using Claude for months before this experiment, mostly for first drafts and quick answers. What I hadn’t done was commit. I’d been using AI as a finishing touch, not as a foundation.
The experiment was simple. For 30 days, every task that could go through AI would go through AI first. Email. Reports. Content. Client documents. Proposals. Research. If it was on my task list and there was any possible way to route it through a system, I had to try.
I tracked time. Before and after, task by task. I kept a running note of what worked, what didn’t, and what surprised me.
The results were not all sunshine and saved hours. Some of it was genuinely frustrating. But the numbers at the end were hard to argue with.
What I automated, and what it actually saved
Email triage and drafting
Before: I was spending about 90 minutes a day on email. Reading, prioritising, drafting replies, forwarding things with context added. It was the slow leak that never got fixed.
After: I set up a morning email triage system. Claude reads a summary of what’s in my inbox, flags anything time-sensitive, and drafts replies to anything that doesn’t need a personal touch. I review, adjust where needed, send.
The actual saving: from 90 minutes to about 20. That’s 70 minutes a day. Across a working month, that’s roughly 25 hours.
I want to be honest about what that number hides. The first week was slower, not faster. I was writing prompts, adjusting tone, correcting things that were slightly off. There’s a ramp-up cost. By week two it was running smoothly. By week three I stopped thinking about it.
Client reporting
This is where I saw the biggest single improvement.
Before: a monthly client report took me between 3 and 4 hours. Pulling data, writing commentary, formatting it, sense-checking it. Half a working day, gone.
After: I built a reporting workflow in Claude. It reads the raw data I paste in, knows the client context from my brain, and produces a structured draft with commentary and recommendations. I spend about 20 minutes reviewing and signing off.
That’s a saving of roughly 3.5 hours per report. If you do 4 client reports a month, that’s 14 hours back. Every month.
The quality, honestly, was better than my manual reports in some ways. Not because AI writes better than me, but because the structure was more consistent and I stopped rushing the commentary at the end of a long afternoon.
Content production
Before: writing a LinkedIn post took me anywhere from 30 minutes to two hours, depending on how much I was second-guessing myself. A blog post was a half-day or more.
After: I stopped writing from a blank page. Every piece of content starts with a brief, goes through Claude for a first draft, and I edit. The editing is the writing now. I’m shaping and improving, not generating from nothing.
Time saved per LinkedIn post: about 45 minutes. Blog posts went from 3-4 hours to about 90 minutes including editing.
Over the month, I published more than I had in the previous three months combined. Not because I’m more disciplined. Because the resistance dropped.
Research and competitor analysis
Before this experiment, research was something I either rushed or avoided. Reading industry articles, pulling together context before a call, checking what competitors are doing, it all felt like overhead.
I gave Claude a research brief. Specific questions, specific sources, a clear format for the output. What I got back wasn’t always perfect. But it was a solid starting point in 5 minutes instead of 90.
The specific saving here is hard to calculate because research is the kind of thing that expands to fill whatever time you give it. What I can say is that I stopped avoiding it.
Proposals
This one surprised me most.
I’d been spending 2-3 hours on proposals. Not because they’re complicated, but because I’d start each one from scratch, convince myself it needed to be perfect, and drag it out.
I built a proposal template inside my brain. Claude knows the service, the pricing structure, the common objections. I paste in the client context from our discovery call. I get a full proposal draft in about 10 minutes.
Month total: 4 proposals. Before, that was 8-12 hours of work. After: about 1 hour total, including review and edits.
What didn’t work
I promised honesty, so here it is.
Anything that needed real relationship nuance flopped. I tried routing some client communication through Claude early on and the tone was wrong. Not obviously AI, just slightly off. Like it was written by someone who’d read a lot about me but doesn’t quite know me. I stopped doing that by day 10.
Strategy didn’t work as a solo AI job. I’d try to use Claude to think through a positioning problem and end up with something competent and hollow. AI is good at executing inside a frame. It’s not good at building the frame from scratch. For real strategic thinking, I still need a blank page and my own head.
First drafts needed more context than I expected. The quality of output is directly tied to the quality of input. Early in the month I was lazy with my briefs. “Write me a blog post about AI automation for accountants” produces very different output from a 200-word brief with a specific angle, target reader, and tone notes. I had to learn to invest 5 minutes upfront to save 90 minutes on the back end.
The number that mattered
At the end of 30 days, I ran the totals.
Email triage: 25 hours saved. Client reporting: 14 hours saved. Content: 12 hours saved. Research: 8 hours saved. Proposals: 10 hours saved.
Total: roughly 69 hours in a month.
I’m not going to overclaim that number. Some of those estimates are rough. Some of that time was reinvested into better work rather than pocketed as free time. But even at half that figure, 35 extra hours is almost a full working week. Every month. Without hiring anyone.
For a solo operator, that’s not a productivity win. That’s a structural change.
What actually made this work
The honest answer: it wasn’t the AI. It was the system.
Using AI in a one-off, ad hoc way is better than nothing. But it caps out fast. What made the 30-day experiment actually stick was having a brain that knew my business. My brand voice, my clients, my processes, my pricing. Claude wasn’t starting from scratch every time. It was working from a foundation I’d built.
That foundation is what the Clever Operators brain build creates. You teach it once. Then it works from that context every time, getting sharper as you go.
If you want to see whether this kind of setup makes sense for your business, start with a conversation. Tell me where you’re losing the most time, and I’ll tell you honestly whether there’s a system that fixes it.
FAQ
How long does it take to set up AI automation for a small business?
The brain build takes between 3 and 5 hours across the seven steps. Most of that is teaching the system about your business. The initial setup is about 10-20 minutes.
Do I need technical skills to run my business with AI?
No. The brain build is designed for business owners, not developers. If you can use Google Docs and have a conversation, you have everything you need.
What AI tool did you use for the 30-day experiment?
Claude, by Anthropic, running in Claude Code. The Anthropic Pro subscription costs $20 a month. That’s the only paid tool required for the core system.
What tasks should I automate first?
Email, reports, and repeatable documents. These are the tasks that eat the most time and cause the least damage if an AI draft needs a quick review before it goes out.
Is AI good enough to replace a VA or admin assistant?
For repeatable, process-based tasks, yes. For relationship management, nuanced communication, or anything requiring real human judgement, no. The goal is to free you up for the work only you can do.
