Earlier notes. Same idea: what is working for small businesses using AI, and what is not.
An AI savings number is only as good as the assumptions underneath it.
An AI savings number is only as good as the assumptions underneath it.
The planner on my site puts them on the screen instead of burying them: $32 an hour as a loaded staff cost, the automation absorbing about 65% of the task, and 50 working weeks a year. Change any one of those and the answer moves.
That's the myth worth dropping — that the number is the finding. It isn't. The assumptions are. It is an estimate, not a quote for your books, and if the answer looks wrong for your business, it probably is.
So before you trust anyone's estimate, mine included, ask to see what it's made of.
Every automation on my site starts with an event. Not a task — an event.
Every automation on my site starts with an event. Not a task — an event.
A call you cannot answer. A form or work order arrives.
That distinction is the whole thing. Ask a small business owner when their follow-up actually goes out and the honest answer is usually "when I remember to." That isn't a trigger. That's a person, and the person is already busy.
You don't need to buy anything to test this. Take the job that keeps slipping and write down the moment it should start. If you can't name the moment, there is nothing to build on yet — and that is worth knowing before you pay anyone.
I have five town pages on my site. Annapolis, Edgewater, Severna Park, Crofton, Glen Burnie.
Most local pages are one template with the town name swapped in. Skimming, you cannot tell that apart from one somebody actually wrote.
There is a tell, and it is not the design. A generated page cannot afford to be specific, because anything specific can turn out to be wrong. So it stays vague about everywhere it names.
The Severna Park page names the road I spend part of most weeks reading businesses along, names the group up there that has already solved this without me, and tells them the page is not for them.
All three could be wrong. That is what makes them worth writing.
Before you automate anything, sort the job into three lanes.
Before you automate anything, sort the job into three lanes.
One: what the software does on its own. Two: what it drafts and you approve before it goes out. Three: what only a person should ever do — move money, run payroll, sign, hire, fire, file taxes.
Most people start by picking a tool. The sorting is the actual work, and it costs a sheet of paper.
Lane three is the one to write down first. It is the only one that does not move as you get more confident.
Try it on one job this weekend and see which lane it lands in. The longer version is in my DIY toolkit, and part one is free to read.
Most people decide whether to hire someone to automate something before they have ever automated anything.
Most people decide whether to hire someone to automate something before they have ever automated anything.
That is the expensive order to do it in. You find out what you actually want after the first small thing works, not before it.
The cheap version: pick one job, do it yourself, badly, and see whether the week feels any different. There is an ad maker that is free to start on my products page, a stack audit that needs no account, and a $19.99 download if you want something walking you through it in Claude.
Then come talk to me, or don't. Either way you will know something you do not know this morning.
Before you hire anyone to automate part of your business, find out who is actually going to do the work.
Before you hire anyone to automate part of your business, find out who is actually going to do the work.
Not the company name — the person. Whether you can reach them. What happens when you call.
In Anne Arundel County you can usually find that out. Somebody has worked with them, or you ask directly and watch how straight the answer comes back. A vendor who won't tell you who is building it has told you something.
Mine is in writing on my about page: when you call, you get me. Use that as the test on anyone you talk to, me included.
An AI consultant telling you not to pay for AI is an odd look, so it's in writing on my FAQ page.
An AI consultant telling you not to pay for AI is an odd look, so it's in writing on my FAQ page.
Many of these tools have a free tier that's enough for one person — ChatGPT, Gemini and Claude all do. For most small businesses that's enough, right up until you want it doing something on a schedule.
That's the real line. You start paying when you want it running without you sitting there watching it, or when you need it to hold on to your business's information.
Neither is about how clever the tool is. Both are about whether a person still has to be there.
If you haven't crossed either line, you're done. Stay on the free tier.
On one of them I decided there was no booking system at all. I was wrong.
Last week I read a stack of small business websites around the county, looking for how a customer would book or ask a question.
On one of them I decided there was no booking system at all. I was wrong. It was installed, paid for and running — just invisible on the page unless you already knew it was there.
That is worse than not having one. You bought the thing, and the customer still can't find the door.
There's a line on my own site: they showed up, and the door was locked. Turns out the lock is sometimes on the inside.
Open your site on your phone and try to book yourself.
Before you hand anything over to someone automating part of your business, read what they are asking you for.
Before you hand anything over to someone automating part of your business, read what they are asking you for.
My own onboarding page says it in writing: do not enter passwords or sensitive customer records, share only the details needed for the agreed project, and submitting the form does not change the scope or price. It runs five to eight minutes, and you can come back to it later.
That is not generous. It is the floor.
If a vendor wants your logins before there is an agreed scope, or the intake form quietly turns into the quote, you have your answer either way.
There's a free prompt on my site that turns ChatGPT or Claude into a prospecting researcher.
There's a free prompt on my site that turns ChatGPT or Claude into a prospecting researcher. No code, no signup. It interviews you first, then hands back short dossiers you can call from — five at a time, on purpose.
The part worth reading is further down the page: how to check its work.
Pick two of the five. Verify one fact on each yourself — open the site, dial the number. If either one is wrong, bin the whole batch and run it again rather than fixing it by hand. A model that made something up once has usually done it more than once, and the one you caught is not the one that would have embarrassed you.
A flat price is easy to publish. The useful part is knowing what moves it.
A flat price is easy to publish. The useful part is knowing what moves it.
Mine is on the page: one workflow, automated end-to-end, is $495 and goes live in about a week. Two or three workflows that talk to each other is $1,500.
What changes between those isn't how hard any single task is. It's the joins — the handoffs where one thing has to tell the next what happened, and both have to agree. That's where the work goes, and it's why "can we add one small thing" is rarely small.
Before you ask anyone for a price, count the handoffs, not the tasks.
The job that drove you mad in July is still broken — it's just not bothering you this week.
Around here the season just turned over. If your summer was flat out, the next few weeks are the quiet part.
Here's the trap in that. The job that drove you mad in July is still broken — it's just not bothering you this week. A problem that isn't bothering you doesn't get fixed, so it'll be waiting in June, exactly as expensive as last time.
So do the boring thing while you still can. Write down the three that cost you the most, in your own words. Not what you'd automate — what actually went wrong, and which part hurt.
You can fix it any time. You can only remember it now.
AI got very good at making the ad this year. It did not get any better at giving you something to say.
AI got very good at making the ad this year. It did not get any better at giving you something to say.
There's a free ad maker on my site. Among the layouts it offers are before/after, testimonial, and comparison — three formats that work, and every one of them runs on a fact the software doesn't have. A real before. A customer who actually said the thing. A comparison you'd defend out loud.
That's the half nobody automated. Making the thing got cheap this year. Having something true to put in it didn't move at all, and it was always the expensive half.
So if you're stuck on an ad, check which half you're stuck on. It usually isn't the design.
Every automation example on my site has three lines under it: the trigger, the workflow, and your part.
Every automation example on my site has three lines under it: the trigger, the workflow, and your part.
The third one is the one worth reading.
On the missed-call build, your part is to review the enquiry and call back. On the paperwork build, it's to review and approve the record before it goes anywhere.
That isn't hedging. An automation with no "your part" hasn't removed the judgment — it has moved it somewhere nobody is looking, and you find out which way it guessed months later.
So if someone is selling you one, ask what your part is. If there isn't an answer, that's the answer.
If you have never booked a free consult with anyone, it is fair to assume it is a sales call with a calendar invite on it.
If you have never booked a free consult with anyone, it is fair to assume it is a sales call with a calendar invite on it.
So here is the whole agenda for mine.
You describe one repetitive task — what gets repeated, how often, and which tools your team uses. You do not need a technical brief. Then we check the fit: what could be handled automatically, what needs a person, and whether your current software already does the job.
That last one is a real answer, not a courtesy. Sometimes the tool is already paid for and nobody turned the feature on.
Nothing is installed, signed or paid for during the call. Fifteen minutes.