SAF Amelia Island 2026 · AI Education Session
Build Your Shop’s AI Brain
Everything from the session, free. Start with the prompt below. The worksheets are here if you want to work it properly, but you do not need them to begin.
Start here: the only thing you have to type
Copy this into your AI assistant word for word, and change the first line to your shop. It hands the AI the whole method, so you do not have to remember any of it. Your job from there is answering questions about your own business.
The prompt
I run a flower shop and I want your help with one specific problem in it. Do not start solving it yet.
First, interview me. Ask one question at a time, wait for my answer, and push back when an answer is vague. Cover these in order: the problem and what it costs me in a normal week; how I will know it is better; what my shop is and who buys from us; how I decide this today, in my own words; what you must never do; what records I already keep and where they live; and what those records probably cannot tell you.
Hold these rules for the whole project. You recommend, I decide. Every number traces back to a file I can open, and you show me the math when I ask. If the data cannot answer, say so and tell me what would fix it. Do not create new work for my staff. And tell me when the data disagrees with me, because I would rather be corrected than agreed with.
When the interview is done, write it all back to me as one page in my own words, and we will use that page as your standing instructions from now on. Then, before you build anything, walk me through your plan in plain English and tell me what you are assuming about my shop that you should check with me first.
Once I approve the plan, ask me which of these fits my shop and recommend one with reasons: I ask you when I need it; you produce the same output every week when I give you a fresh file; or a dashboard I open on my own computer, built from a project folder that holds my files and your instructions. Tell me what each would cost me to set up and to keep running. If I pick the dashboard, describe the screen to me in words first, what is on it, what I look at first, what I can click, and get my approval before you build it. Then build the smallest useful version, using the file I give you, and show me the result with the math behind every number in it.
What happens after you paste it
- It asks, you answer. Expect twenty minutes of questions about your own shop. Answer the way you would train a promising new hire. There are no technical questions in there.
- Save the page it writes you. That one page is the whole point. Keep it where the AI can read it again, not buried in a chat you will never find.
- Come back to the same place every week. It does not remember you. Reopening a brand new chat every time is why most people feel like they are starting over.
- Expect the first real answer to be wrong. You are the one who knows your shop. Say so out loud when it does not match, and it gets better from there.
Before it can really help you
Most free chat boxes forget everything the moment you close them, and cannot open your files. To do this work you need an assistant that can do three things: read files you give it, keep your instructions between sessions, and let you come back to the same project. Some of this works on a free account. What a paid plan buys you, usually around the cost of one wholesale box a month, is enough room to keep all your files in one place and to use it every week without running out. We used Claude. Others can do this too. Those three abilities are what matter, not the brand.
The method, in six steps
Steps 1 to 3 happen with no AI in the room at all. That is the part people skip, and it is the part that decides whether the rest works.
- Pick the one problem. Small, specific, weekly, and it slows you down. Write down what it costs you in a normal week, and how you will know it is better. That last line is your finish line, and you will check the results against it later.
- Teach it your business. Who you are, who you serve, and how you actually decide. What you would tell a new employee in their first week, including the rules of thumb you have never written down.
- Set your rules. The standards you say once, in plain English, before you ask for anything. Borrow ours if you like them.
- Talk through the plan before anything gets built. The cheapest place to change your mind. Ask what it is assuming about your shop that it should check with you first, and ask for the smallest version you would actually use on Monday.
- Check it on the real screen. Against the finish line from step 1, in the place the work actually happens. Run it first on a week that already happened, because you know what the answer should be.
- Fix it, then name the finish line. Corrections make it smarter, and the record of why you corrected it is the valuable part. Decide when it is good enough to start using, so you are not still tinkering in November.
Four rules you can borrow as they are
“You recommend. I decide.”
“Every number has to trace back to a file I can open.”
“If the data cannot answer, say so and tell me what would fix it.”
“Do not make my team type anything they are not already typing.”
And the one people forget: “Tell me when the data disagrees with me. I would rather be corrected than agreed with.” It will happily agree with you about your own shop, and you will have no way to catch it.
Nine things your AI needs to know
This is the long version of steps 2 and 3, and it is the conversation that made the difference at Norton’s. Do not sit down and answer all of it alone. The prompt above already asks the AI to walk you through most of it.
1. The job
What problem am I hiring you to help with, and what does it cost me in a normal week? What exactly should you hand me, and on what day? Who reads it, and what do they do next?
2. The shop
What kind of shop is this, who buys from us, and what makes us different? How do I make this decision today, start to finish? What are my rules of thumb, including the ones I have never written down?
3. Where your job stops
Which calls are always mine, no matter how confident you are? What are you allowed to decide on your own? When you are not sure, what do you do? And when the data disagrees with me, say so.
4. What good looks like
What is this number today, before you touch anything? Write it down now, because if you skip this you can never prove it helped. What would make me and my team say out loud that this worked? When is it good enough to start using?
5. What you must never do
What must you never do, even if I ask in a hurry? What must you never change or touch? Who must you never create extra work for? That last one is not about the AI, it is about your staff.
6. What you must always show
Every number traces back to a file I can open. When I ask how you got there, show me the arithmetic, not a paragraph. Tell me how old the data behind an answer is, and say plainly when it is too old to trust.
7. What you run on
What records do I already keep, and where do they actually live? How do they get to you, who does it, and how often? What happens the week nobody remembers? What else may you reach on my behalf, and whose computer do you live on?
8. What you cannot know
What is missing, thin, or plain wrong in my records? What questions must you refuse to answer from them? How will you say so, out loud, when one of those comes up? This is the one most people skip, and the one that decides whether anyone keeps trusting it.
9. When you are wrong
How do I tell you, and where does my reason get written down so it sticks? When I overrule you, I want my reason sitting next to the next recommendation, not lost in a chat window. And after I correct you: what did you learn, what let it happen, and what changes so it cannot happen again?
Three more things to type, later
When you doubt a number
“Show me your math. Where did that number come from?”
When it is wrong
“That does not match what I see in my shop. Here is what I actually have.”
After you correct it
“What did you learn, what let that happen, and what changes so it cannot happen again?”
How does it remember? It does not.
This is the question that comes up every time, and the honest answer is that your assistant starts from nothing every session. What looks like memory is a handful of plain documents kept in one place that it reads at the start of every session, and that you add to at the end. Three are enough for a shop:
- What you told it. The one page from the interview. Its standing instructions.
- What happened. A running log, newest at the top. What you asked, what it got wrong, what you decided. Add to it at the end of a session, never rewrite the old entries.
- What it cannot answer. A short list of the gaps in your own records, so its limits travel with its answers instead of being discovered later by a customer.
Every good assistant has somewhere to keep these. In Claude it is called a Project. You make one for your shop, put those documents in it, and start every session inside it instead of in a new chat. Whatever yours calls it, you are looking for one container that holds your files and your instructions and that you can come back to. Setting that up takes about five minutes and it is the difference between a tool and a series of conversations.
That is the whole trick. The memory is not a feature you buy, it is a habit you keep.
A realistic four weeks
- Week one. Interview. Paste the prompt and answer questions. Nothing gets built. Save the page it writes you.
- Week two. First real answer. Hand it one export and ask one question you already know the answer to. Compare.
- Week three. Correct and repeat. Ask it something you do not know the answer to, then check it against the shop.
- Week four. Make it routine. Same day each week, same question, and a place your new exports land so it stays current.
Where people stall
- Wanting to start building. Steps 1, 2 and 3 have no AI in them. Doing them is the work, and skipping them is why most of these projects quietly die.
- Waiting for clean data. You will never have it. Start with the records you already keep and let the gaps surface as questions.
- Accepting a plausible answer. The first answer that looks right is the most dangerous one. When it disagrees with what you know, say so. You are usually the one who is right.
- The export. What actually stops most shops is not the AI, it is getting a sales export out of the point of sale. Find out this week whether you can, and ask your vendor if you cannot.
Where to stop
This is a tool for you and your team, not something you put in front of customers or sell. Keep a copy of the version that worked before you change it, because you will break it at some point. And what you saw on stage is not a four week outcome: that one took an IT director about two months on top of real data work. What you can have in a few weeks is an assistant that holds your files and your instructions in one place and gives you an answer you can check.
Download the sheets
The deck from the session, including the four step method and the walk through of how the tool was built. Large file, best on wifi.
The method in four steps, the three habits, your first 30 minutes, and the words that work. This is the sheet from your seat.
The long version, with fill in lines for each step, prompts to ask at each stage, and a page on how the conversation becomes a working tool.
The nine question groups above, laid out as an interview to work through, with a note on why each one matters.