How I Automated My Entire Small Business with AI

Eighteen months ago I was working 60-hour weeks running a one-person consulting business. Client emails, project proposals, invoices, social media, research, meeting notes — all of it was manual, all of it was me. Today I work around 30 hours per week on the same revenue. The other 30 hours? Handled by a stack of AI tools that cost me less per month than a decent restaurant meal.

This article is not a theoretical overview of AI automation. It is exactly what I use, what I pay, and what it replaced. Every tool named here is one I have used for at least six months.

I run a small business. For a long time, “running it” mostly meant drowning in it. Then I started handing pieces over to AI, one at a time, over about nine months. This is the honest version of what happened — what worked, what didn’t, what I spent, and what I’d do differently if I started over today.

Where I started: drowning in admin

In early 2025 my week looked like this: about 12 hours of customer support email, 8–10 hours writing content (articles, social, the newsletter), 4 hours scheduling and rescheduling calls, and another 6 hours on bookkeeping and quotes. That’s roughly 30 hours of pure administrative work before I’d done a single piece of actual billable work.

I’d tried hiring a virtual assistant twice. Both times the handoff cost more than the help saved. So when I started experimenting with AI automation in February 2025, the goal wasn’t 10x growth. It was: stop working Sundays.

The three things AI replaced first

1. Customer support email (the obvious win)

Support email was the easy win because 80% of my inbox was the same 12 questions: password resets, invoice changes, nonprofit discounts. I wasn’t being a craftsman about these replies. I was just slow.

In week one I exported six months of sent mail, pasted batches into Claude (Pro, $20/month), and asked it to cluster the questions and draft a reply template for each cluster — in my voice, using my past replies as examples. That gave me a 14-template starting library.

Then I set up a Gmail filter + Zapier ($19.99/month Starter plan) flow. Inbound email goes to ChatGPT (Plus, $20/month) with a system prompt that says, roughly: “Read this email. Decide which of these 14 templates fits. Personalize the greeting and any specific details. Return the draft.” The draft lands in my Gmail Drafts folder. I read it, hit send, or rewrite if it’s off.

I am not letting AI auto-send. Ever. This is a deliberate human-in-the-loop setup, and I will explain in the “didn’t replace” section why I won’t move off it.

Result: support email dropped from ~12 hours/week to ~3.

2. First-draft content (newsletter, blog, social)

The mistake I made for three months was using AI to write finished content. It produced fluent, generic, faintly embarrassing prose. Lots of “in today’s fast-paced world.” My open rates dropped. A reader emailed to politely ask if I was okay. I was not okay.

What worked instead: I use Claude as a structured first-draft tool, never a finisher. My loop is:

• I record a 5–8 minute voice memo on the topic — the rant version, full of opinions and detail.
• I transcribe with MacWhisper ($59 one-time) or Whisper inside ChatGPT.
• I paste the transcript into Claude with a prompt that says: “Here’s a voice memo. Build a structured draft from it. Keep my phrasing wherever possible. Don’t add facts I didn’t say.”
• I rewrite roughly 30–50% of what comes back. The draft is a scaffold, not a final.

That cut content time from ~8 hours/week to ~3, and the writing got more specific, not less, because I was no longer skipping the thinking step. The voice memo is the thinking. AI just types it up.

3. Scheduling and meeting prep

I moved everything onto Cal.com (free tier — Calendly works the same way). Then I added a Zapier flow: when a meeting is booked, send the attendee’s name and company to Claude, which pulls public info from their LinkedIn and website (whatever the booking form provided), and writes me a 1-paragraph briefing. It lands in my calendar event 30 minutes before the call.

Is this an AI agent? Technically no — it’s a chained workflow with one decision step. But it does the job an assistant used to do. Saved another ~2 hours/week and made me look much more competent on calls.

If you’re choosing between automation platforms, I wrote up the tradeoffs in Zapier vs Make. Short version: Zapier is friendlier, Make is cheaper at scale. I started on Zapier and have not bothered to switch.

The three things AI did NOT replace

1. Anything that requires me to actually decide

Pricing a custom proposal. Saying no to a client who’s slowly turning into a problem. Picking what to build next. AI is genuinely good at giving me three options and pros/cons. It is not good at choosing for me, and the times I let it choose, I felt strangely numb about the result. Decisions you don’t own come back to bite you.

2. Sensitive customer email

About 8% of my support email is “I’m upset” or “something bad happened.” AI drafts for those felt clinically correct and emotionally tone-deaf. Worse, it occasionally invented plausible-sounding details (this is what people mean by AI hallucination — confident, fluent, and wrong). Anything containing “refund,” “complaint,” “frustrated,” or “lawyer” now routes to a folder AI doesn’t touch. I write those by hand.

3. Bookkeeping reconciliation

I tried. I gave Claude a CSV of bank statements and asked it to categorize transactions. It was about 92% right, which sounds great until you remember that the 8% wrong is exactly the part the IRS cares about. I’m back to a human bookkeeper ($180/month). Some things are worth paying for.

The stack I ended up with

This is what’s actually running my business as of early 2026. Not what I tried — what survived.

• Claude Pro — $20/month. My main writing and thinking partner. Long context, good at staying in voice.
• ChatGPT Plus — $20/month. Used for the email-drafting flow and image generation. Yes, I pay for both. They’re good at different things.
• Zapier Starter — $19.99/month. Glue between Gmail, Cal.com, my CRM, and the AI APIs.
• Cal.com — free tier.
• MacWhisper — $59 one-time. For voice memo transcription.
• Notion — $10/month. Not strictly AI, but where everything lives.
• Bookkeeper — $180/month. Not AI. Worth it.

Total AI-related spend: roughly $70/month, plus the bookkeeper. For a list of tools I tested and dropped, see Best AI Tools.

10 Automation Plays Most Small-Business Owners Have Not Tried

The story above is one path. The 10 plays below extract specific high-leverage moves you can copy regardless of business type.

1. Start with your highest-frequency admin task

Look at last week. What did you do 5+ times? That is the right first automation target. High-frequency simple tasks compound faster than rare complex ones.

2. Map the workflow before automating

Write the workflow on paper first. Most failed automations skipped the manual-mapping step. Once mapped, surprises in the automation come down materially.

3. Pick tools that connect, not best-in-class isolated

The MOST connected tool beats the best isolated tool. Workflow speed comes from integration, not feature lists. Optimize for the stack, not the individual node.

4. Build a personal Skill for your single biggest decision pattern

You make the same kind of decision repeatedly (pricing, client triage, vendor evaluation). Encode the decision logic as a Skill. Quality improves and decisions get faster.

5. Customer-communication templates with personalization

Onboarding, follow-up, milestone-celebration, win-back emails. AI personalizes templates with customer-specific details. Every customer feels uniquely served at near-zero marginal cost.

6. Daily-data-pull dashboards instead of monthly reviews

Monthly reviews catch problems weeks too late. Automated daily pulls from your accounting, sales, and operations data surface trends in real time.

7. Recurring-revenue surfacing in client interactions

Most small businesses leave recurring revenue on the table. AI surfaces conversation moments where a maintenance contract, subscription, or recurring package would naturally fit.

8. The one-hour-Friday-review ritual

Every Friday, one hour with Claude reviewing the week. Patterns emerge across weeks that you would not see one week at a time. Iteration discipline compounds.

9. Document your workflows so they can transfer

Even if you stay solo, documented workflows make hiring or selling possible. AI helps you produce the SOPs you would otherwise never write.

10. The exit-optionality discipline from year one

Build the business so it could be sold even if you never sell it. AI makes the documentation, process, and customer-relationship-portability work feasible for solo operators.

Before and after: the actual numbers

I tracked my hours in Toggl for two months before I started and again in March 2026. The comparison:

• Support email: 12 hrs → 3 hrs/week
• Content: 8 hrs → 3 hrs/week
• Scheduling and prep: 4 hrs → 0.5 hrs/week
• Bookkeeping/quotes: 6 hrs → 4 hrs/week (smaller win — see above)

Net: roughly 19.5 hours back per week. I did not use that to grow revenue 2x. I used 8 hours of it to take weekends off, and the rest to do deeper, slower work. Revenue grew about 35% over the same period, which I attribute mostly to the deeper work, not the automation. The automation just made the deeper work possible. That’s the honest AI ROI story for me.

What I’d do differently

Start with one workflow, not five. My first month I tried to automate everything at once. Each broken Zap took 90 minutes to debug, and broken Zaps compound. If I started over I’d pick the single most painful task — for me it was support email — and only move on once that was boring and stable.

Don’t use AI to write things you wouldn’t be embarrassed to publish unedited. The version of me that pasted a Claude draft straight into the newsletter was the version that lost subscribers. AI is a co-author. It’s not a ghost-writer who knows your voice better than you do.

Track hours before and after. “I feel like I have more time” is not a result. Toggl is free. Use it for two weeks before you change anything, then again three months later. If the number didn’t move, something’s wrong with the workflow, not with AI.

Keep a “human only” lane. Decide upfront which work AI never touches. For me: anything emotional, anything legal, anything financial that the IRS cares about. Naming the lane in advance stopped me from drifting into bad automation.

If you’re at the start of this, the path I’d point you at is Start Here. If you want the weekly version of this kind of thinking, it goes out in the newsletter, and if you want to compare notes with other small operators doing the same thing, that’s what the community is for.

FAQ

How long until I see real time savings?

Honestly, the first month is usually a net loss. You’re learning the tools, building templates, and fixing your prompts. I didn’t break even on time until about week 6, and didn’t see real compounding gains until month 3. If anyone tells you “automate your business in a weekend,” they are either selling a course or lying. Or both.

Do I need to know how to code?

No. I don’t, in any meaningful sense. Zapier and Make are visual — you connect boxes. ChatGPT and Claude are chat boxes. The hardest “technical” thing I did was paste an API key into a Zap, and there’s a YouTube video for that. The actual skill is writing clear instructions and noticing when the output is wrong, both of which are normal-business-owner skills, not engineer skills.

What if my customers find out I’m using AI?

Mine know. I told them in a newsletter, and the response was a collective shrug. What customers actually care about is whether you reply quickly, whether the reply is correct, and whether they feel respected. AI helps with the first two. The third is on you, every time, no matter what’s drafting the email. That’s the whole job, really — and it’s the part that didn’t change at all.

You May Also Like

• AI Automation for Small Business: Where to Start
• How to Start an AI Automation Agency: Complete Roadmap
• AI Passive Income: 10 Ways to Earn While You Sleep
• Build and Sell AI Agents for Local Businesses
• How to Use AI: A Step-by-Step Guide