The hardest place to start with AI isn't a blank page. It's the small business where you already built something that works badly, and you know every inch of it.
The spreadsheet with seven tabs. The email templates you tweak every single time. The folder system only you can find anything in. It's clunky, and it runs the business.
That's exactly where to start, and it's where my builds at SolvStream begin too. Keep the rules that already work and hand AI one step at a time, on one real job. The usual advice is to begin again with a new tool. Don't.
I nearly started again myself
A few weeks ago I nearly spent a Saturday evening moving all my notes into Obsidian, a note-taking app, mostly because it made everything look so neatly linked up.
The setup I had already did the jobs that mattered. It kept a record of business decisions and ran the tasks I repeat. Obsidian would have done fewer of those. I'd have been rebuilding working parts of the business for the sake of a nicer screen.
AI sets the same trap. The new thing looks clean and what you've got looks like a mess, so the instinct is to throw it out and rebuild. But the mess is where your rules live. Every odd column and every template line you never change is a decision you've already made and tested on real work.
Why do so few businesses get AI to stick?
In Goldman Sachs's survey of US small businesses this year, 76% said they use AI. Only 14% have it fully built into how they run.
Using AI has become normal. Building it into the way a business works is where nearly everyone gets stuck. That's not a skills problem on your side. It's what happens when a new tool meets a system somebody built by hand over years.
The six steps
1. List what you already built
Write down every piece of your current setup that touches the job you want help with. The spreadsheet, the templates, the saved replies, the checklist taped to the wall. Name each one the way you'd explain it to someone covering for you for a week.
It's the step that makes every later one easier, because you can't hand over what you can't see.
2. Mark what already works
Go through the list and tick the parts that do their job. The column you check every morning. The line in your quote that clients never question.
Your existing process is the input, not the throwaway. Those ticked parts are the rules AI will need to follow, and you've already tested them on real customers.
3. Find the step that keeps coming back to you
Every homemade system has one step you always end up doing by hand. Copying figures from one place to another. Rewriting the same paragraph for a slightly different client.
That step is your first candidate, because you can already tell a right result from a wrong one. A quick test: if you could do it half asleep and you still do it every week, that's the one. If you're torn between two, this guide to picking what to automate first walks through how to choose.
4. Keep the rules, hand over the typing
Give the AI the rules from step 2 and the job from step 3. Not a new process. Yours, written down. The AI does the typing, and your rules decide what counts as good.
The rules can be plain. For a quote, it might be: always include the call-out charge, and ask for a deposit on anything that runs past two days. Two lines like that, taken from how you already work, are worth more to the AI than a page of instructions about tone.
That's the whole reason to fix the process before adding AI. A tool can only follow rules that exist somewhere it can read them.
5. Move one step on one real job
Pick one client or one project. Run that single step through the AI and leave everything else exactly as it is, even when it's tempting to move a second step early.
One real job teaches you more than a month of testing on made-up examples, because real jobs have the awkward exceptions your made-up ones leave out.
6. Keep the old way running until the new one earns its place
For a fortnight, do that step both ways and compare. If the AI version is as good and quicker, switch the step over and go back to step 3 for the next one. If it isn't, you've lost nothing, because the old way never stopped.
Keep a note of everything the AI got wrong during that fortnight. That list is your next set of rules, and it's how the second attempt ends up better than the first.
Nothing gets switched off until the replacement has proved itself on real work. That's what stops a sensible experiment turning into a weekend of rebuilding.
Where the free map fits
If you'd rather not do step 1 on paper, the AI Workflow Map is free. You describe how you work and it shows you which parts of your week suit AI, and which don't. It takes about two minutes.
Where this leaves you
You don't need a new system to start using AI properly. You need the one you've already got, written down, with one step handed over at a time.
Common questions
Do I need to replace my spreadsheets to use AI?
No. Most of the value sits in the rules your spreadsheet already holds: what goes in which column and what counts as finished. Write those down and hand one step over. The spreadsheet can stay.
Which step should I hand over first?
The one you keep fixing by hand, where you already know what the right answer looks like. A step with a clear right answer is easy to check, so you'll know quickly whether the AI is doing it properly.
What if the AI gets it wrong?
That's why the old way keeps running for the first fortnight. You compare the two on real work, and nothing depends on the AI version until it has matched yours.
