Earlier notes. Same idea: what is working for small businesses using AI, and what is not.
There's a free calculator on my site that adds up what you pay for software and outside services every month. Part of it is built to make that number smaller.
There's a free calculator on my site that adds up what you pay for software and outside services every month. Part of it is built to make that number smaller.
It has a whole group headed "we'd tell you to keep paying for these." Your bookkeeper. Your accountant. Payroll. The software your trade actually runs on. You tap them so your total is honest, and they never show up as something to cut.
That's deliberate. If I talk you into cancelling something and it burns you, you never call me again. I'd rather hand you the shorter list and be right about it.
There's a five-question readiness check on my site, and four of the five questions have nothing to do with AI.
There's a five-question readiness check on my site, and four of the five questions have nothing to do with AI.
Do you have customer data. Do you have clear goals. Are you open to the tools. Does your team have basic digital skills. Is there a budget, even a small one.
Only the first is technical — and for most small businesses, "customer data" means the inbox and the invoices.
So "are we ready for AI" is usually the wrong question. You're ready. The real question is which job, and if you can't name the one you still do by hand every week, no checklist is going to find it for you.
I spent part of a morning going through this year's Best of Annapolis winners, looking for businesses that might need what I do.
I spent part of a morning going through this year's Best of Annapolis winners, looking for businesses that might need what I do.
Mostly I found businesses that don't. The ones I checked were already booking online, already answering, already sorted.
Which makes sense. An award list picks the best-run shop in each category, and best-run usually means somebody already bought the tool.
So if you're looking at a winner in your trade and feeling behind — that gap probably isn't how well you run the place. It's that nobody there has had a free week to hand off the job you're still doing by hand after everyone's gone home.
Fair question, and most software answers it badly: stop the subscription, lose the thing.
That isn't how a build works here. The automation is a flat fee — $495 for one workflow, live in about a week — and you own what we build. Plain-English handoff, running on the tools you already use.
There is a monthly and I'll name it: AI Partner, $295/month. Monitoring, fixes, priority support. Cancel anytime. Cancel it and the automation is still yours, because you already owned it.
Two businesses in the same trade, same county. One runs every booking through a platform. The other takes them by phone and text.
The easy read is that one is better run. It isn't. One of them has somebody at a desk. The other is in a van all day.
That's the split I keep finding — not big versus small, desk versus no desk. If you're the van, the question isn't which software to buy. It's what happens to the person who finds you at nine at night while you're finishing a job.
And if you already call everyone back within the hour and it works, don't fix it.
A question I see a lot: what is the 30% rule for AI?
A question I see a lot: what is the 30% rule for AI?
Here's the honest answer. It's a rule of thumb, not a law. Roughly a third of routine desk work is repetitive enough to hand off, and that's where to start looking. That's the whole of it.
Don't take the number literally. It describes an average business, and nobody runs an average business. A third of your week might be two hours, or it might be none.
So look at your own week instead. That's the only sample that counts.
The automations that die aren't the ones that break. They're the ones nobody can explain.
The automations that die aren't the ones that break.
They're the ones nobody can explain.
Six months in, the person who asked for it has moved on. Whoever is left doesn't know what it touches or how to switch it off — so they stop trusting it, and then they stop using it. Nothing failed. It just quietly went unused.
That's why everything I build comes with a one-page description of what it touches and how to switch it off. Not paperwork for its own sake. It's the thing that decides whether it's still running next year.
If your team can't explain it, they won't trust it.
Five questions — what the business is called, what's eating your week — and it comes back with a named automation, an illustrative capacity estimate, and what I'd charge to build it. The questions run in your browser. Business lookup and usage requests may be sent while you use the advisor; an email address is only shared if you enter it to request follow-up. Read the plan, close the tab, build it yourself if you like.
The assumptions are on the screen rather than buried: $32 an hour as a loaded staff cost, about 65% of the task absorbed, 50 working weeks a year. If it's a bad fit, it says so.
Four questions to ask before you sign an AI quote.
How many automations is that, exactly? "AI strategy" and "automation buildout" are phrases, not counts.
Which tools does it run on, and whose account are they in?
What counts as finished? If nobody wrote that down, the job ends when somebody gets tired.
What happens when it breaks, and who can switch it off?
A quote that can't answer those four was guessed, not scoped. Mine is $495 flat for one automation, about a week, agreed before any work starts — and you own what gets built, including the accounts it runs on.
Before you pay anyone to automate something, do the arithmetic.
Before you pay anyone to automate something, do the arithmetic.
Take the job you want to hand off. Count the hours it eats in a week. Multiply by what an hour of that person's time is worth.
If it takes two hours a week and the build is $495 flat, the arithmetic is quick.
If it takes ten minutes a week, do not automate it. Ten minutes a week isn't a problem, it's just something you do — and paying to remove it is how people end up with software they resent.
I'd rather you ran that number before you called me than after.
It's the first question I get and the least useful one.
There is no best one. For writing, ChatGPT or Claude. For connecting things together, Zapier or Make. For keeping track of anything, Airtable or the spreadsheet you already use. That's the honest answer and it takes about ten seconds.
The tool matters far less than picking the right job to point it at. Start with the one you still do by hand every week, and the tool question mostly answers itself.
None of that is a recommendation until I know what you actually do all day.
That's one of the things I actually build, and it's worth saying plainly what it is, because people picture something stranger.
No AI writes to your customer. It's the follow-up you already wrote, sent on day three if nobody has answered — the one that currently waits on somebody remembering.
You approve the wording once. After that it goes out on schedule, whether or not the week got away from you.
One task, about a week, $495 flat.
If you already chase every quote on time, you don't need this.
It's a fair question, and the answer should be no.
AI is confidently wrong sometimes, and it does not know it. So nothing I build goes to a customer without a person seeing it first, and nothing touches your money or your records without a rule around it.
That gets decided when the work is scoped — it isn't a switch you flip afterwards. It's also the difference between an automation you're still using in six months and one you quietly turn off after it embarrasses you once.
Anyone telling you to just let it run has not had to clean up after it.
Most savings calculators inflate the number. Mine has a section built to shrink it.
There's a calculator on my site that adds up what you're paying every month for software and outside services.
Most tools like it are built to make one number look as big as possible. This one has a whole section that shrinks it on purpose — accounting, bookkeeping, payroll, your industry software. You tap those so the total is honest, and they never show up as something to cut.
The interesting part is usually what's left: five separate bills and five logins doing one job.
Under a minute. No email, nothing saved, nothing sent anywhere.