It’s 9 pm. The kids are in bed. You didn’t see them go up.
You’re at the kitchen table with a laptop, asking ChatGPT to write a follow-up email to a customer. It gives you something polite and useless. It has no idea who the customer is, what you quoted them, or that they’ve already paid the %50 deposit.
So you paste the details in. From the CRM. Except nobody has touched the CRM since March.
Next, you check the spreadsheet. Then your inbox. Then you text Dave, because Dave knows.
Twenty minutes later you’ve got your email. You could have written it yourself in five.
Sound familiar?
In Part 1, I said your AI tools aren’t saving you any time because they’re not sitting inside a system. Shared data, repeatable workflows and clear rules. Without those, you’ve bought a very clever junior admin bod who doesn’t know where anything is.
This is Part 2. It’s the boring bit.
Which is exactly why most people skip it.
I have a thesis:
Every AI tool is only as good as the data that you give it.
Rubbish in, rubbish out.
Plenty of businesses can’t find their own data, never mind letting a machine read it.
Start With Your Data
So start with the data. Not the tools.
Think of it like building a house. Nobody posts photos of their foundations on Instagram. Nobody invites the neighbours round to admire the concrete.
But if you get the foundations wrong, it doesn’t matter how nice the kitchen is. It’s going to crack.
Right now, in most service businesses I speak to, the foundations look like this.
A CRM nobody fills in. Spreadsheets held together with sticky tape. An inbox with thousands of unread emails.
And the rest lives in people’s heads.
That’s not a criticism. That’s just how businesses grow. You were busy winning work, not building filing systems.
But it’s why the shiny AI tool you signed up for isn’t helping you that much.
Why am I banging on about this? I’ve been writing code since the early 2000s and building automation since 2007. I’ve worked in software teams, from tiny start-ups to big corporations.
And here’s the thing. Almost every automation I’ve seen go wrong went wrong for the same reason.
Not the code. Not the tool. The data underneath it.
Rubbish in, rubbish out. It was true in 2007. It’s even more true now, because AI will take your rubbish and make it sound like it isn’t rubbish.
So let’s run through the three principles that sort it out.
#1 – One Source Of Truth Per Thing
Three Things, Three Homes
Your business runs on three kinds of information.
Customers. Who they are, how to reach them, what they’ve bought.
Jobs. What you’re doing for them, where it’s up to, who’s on it.
Money. What you quoted, what you invoiced, what they’ve paid.
Each of those needs one home. Not two. Not “mostly the CRM but also Sarah’s spreadsheet”. One.
When a customer’s phone number lives in three places, you’ve got three versions of the truth. One of them is wrong. You don’t know which. Neither does the AI.
The Spreadsheet Nobody Dared Touch
Let me tell you about a construction client.
We were building an automation to help them put tenders together faster. Tenders are a nightmare in construction. Loads of pricing, loads of detail, a hard deadline, and the whole thing sits on the estimator’s shoulders.
The estimator had a spreadsheet.
And it was brilliant. No bones about it. Years of pricing knowledge baked into one file. Formulas on formulas.
It worked, and it worked because the person who built it knew every cell.
But nobody else dared touch it.
Not the directors. Not the rest of the team. It was the estimator’s spreadsheet, and everyone treated it like an unexploded bomb.
Now, I’m going to offend some people, but a brilliant spreadsheet that only one person understands is not an asset.
It’s a risk.
What happens when the estimator goes on holiday the week a big tender lands? Or gets the flu? Or leaves?
The business stops pricing work. That’s what happens.
I believe repeatable processes beat intelligence, every single time. Intelligence walks out the door at 5 pm.
So when we started on the tender build, the first job wasn’t AI at all. It was getting what lived in that spreadsheet, and in the estimator’s head, into a shape that something else could read. Once the pricing had a proper home, the automation had something to work with. Before that, any AI tool would have been guessing.
I’m not knocking the estimator. The spreadsheet was brilliant. That was the problem. Nobody else could build on it.
Pick The Boring Option
So how do you choose where each thing lives?
Grab a brew and a sheet of paper. Write down customers, jobs and money.
Next to each one, write every place that information currently lives. Be honest. Include the whiteboard in the office and the notes app on your phone.
Then pick one home for each.
Pick the one your team already uses, even if it’s not the fancy one. Complex software doesn’t equal the best software.
I’ve seen businesses pay for a massive CRM with hundreds of features and use a handful of them, badly. A well-kept spreadsheet that everyone trusts beats an expensive system that nobody fills in.
Everything else either feeds that home, or you switch it off.
#2 – Decide What AI Is Allowed To See
It Doesn’t Need To Know Everything
Once people get their data in order, there’s a temptation to throw the lot at the AI. Every email, every file, every folder on the shared drive.
Don’t.
More data doesn’t mean better answers. It often means worse ones. And it definitely means more risk.
The New Starter Test
Imagine this. You hire someone new on Monday. Bright, keen, works fast.
Would you hand them the keys to the building, the payroll login, the HR files and your personal email on day one?
Of course you wouldn’t. You’d give them what they need to do the job, and you’d open up more as they proved themselves.
Treat AI the same way. It’s a new starter. A very fast one, but still a new starter.
Split your information into three piles.
Fine to see. Job notes, quote templates, service descriptions, your standard terms. The stuff you’d happily show a customer anyway.
Needs a person to check. Customer contact details, pricing, anything going out the door with your name on it.
AI can draft it. A human signs it off.
Never. Payroll. Staff sickness and HR records. Anything personal about your team.
Anything you’d be shit scared to see on the front page of the local paper.
Write those three piles down. It takes roughly half an hour. It saves you from the conversation nobody wants to have with a customer about where their details ended up.
Rules Beat Trust
Here’s the thing. You can’t rely on everyone in the team remembering what they should and shouldn’t paste into a chatbot.
People are busy. They’ll take the shortcut.
So make the rule part of the system, not part of someone’s memory. If the AI can only reach the “fine to see” pile, nobody has to remember anything.
That’s the whole point of Part 1. Clear rules, built into the system.
#3 – Keep It Somewhere You Control
Renting Versus Owning
I learnt this one the hard way.
A few years back I had a software product on AppSumo, a big marketplace for this sort of thing. It didn’t sell enough, so they kicked us off the platform.
Their platform. Rules that they had decided. Their decision. Not mine.
Crap, I know. But that’s the deal when you build on someone else’s land. They can change the rules whenever they like.
Your data works the same way. If your customers, jobs and money all live inside one tool, and you can’t get them out, you don’t own them. You’re renting them.
That’s fine, right up until the tool doubles its price.
Or gets bought, shuts down. Or changes how its AI uses your information.
Then you’re stuck. Your whole business history is behind someone else’s front door.
The Export Test
So here’s a simple test. Pick each of your three homes, customers, jobs and money, and ask one question.
Can I get all of this out today, in a format a spreadsheet can open, without ringing support?
If the answer is yes, brilliant. Crack on.
If the answer is “I’m not sure”, find out this week. Not when you need it.
If the answer is no, that’s your biggest risk, and it’s worth sorting before you plug a single AI tool into it.
For some businesses, keeping control goes further than that. If you handle sensitive client information, you might want your AI running on your own kit rather than someone else’s servers.
I’ve written a whole on-premise series on exactly that, so I won’t repeat it here. Have a read if that’s you.
What This Won’t Do
I always tell people where an idea breaks. So here goes.
It Won’t Be Exciting
Nobody gets a buzz out of deciding where customer phone numbers live. There’s no demo. No “look what it can do” moment.
You’ll be three weeks in, wondering why you’re not playing with the good stuff yet.
Stick with it. The good stuff sits on top of this.
It Won’t Fix A People Problem
If your team doesn’t fill in the CRM now, moving to a new CRM won’t change that.
I hear this on discovery calls all the time. “Staff don’t do what I’ve asked them to do.”
Sometimes that’s a people issue. More often it’s because the system asks too much of them. Twelve fields to fill in after every call? Nobody’s doing that at 4:45 pm on a Friday.
Make it easy to put data in the right place. Better still, make it automatic. But know that tidying your data won’t, on its own, change how people behave.
It Doesn’t Need To Be Perfect
You don’t need every record cleaned before you start. Progress over perfection.
Get customers into one home first. Then jobs. Then money. One at a time.
Put in the reps.
It’s the same as the gym. Nobody walks in on day one and deadlifts 200kg. You start with the bar and build up.
It Won’t Mean Buying Loads Of New Software
Most of the time, you already own the tools you need. The problem isn’t the tools. Nobody decided which one wins.
Conclusion: Start With Your Data
Here’s the recap.
- One source of truth per thing. Customers, jobs and money each get one home. Brilliant spreadsheets only one person understands are a risk, not an asset.
- Decide what AI can see. Treat it like a new starter. Fine to see, needs checking, never. Build the rule into the system, not into people’s memories.
- Keep it somewhere you control. Run the export test. If you can’t get your data out, you’re renting it, not owning it.
Every AI tool you’ll ever use sits on top of this. Get the foundations right, and the tools start saving you time. Get them wrong, and you’ll still be at the kitchen table at 9 pm, texting Dave.
In Part 3, we’ll build on this and get into the workflows.
I hope this helps.
Oh, just one more thing.
If you want a hand working out where your data actually lives and which process to sort first, map out one of your processes with me here.
