By Jonny Ross, Fractional CMO working with founders and MDs of scaling businesses across the North of England
The short version
- If you’ve been using AI seriously, you probably already have a chat or project that knows your business. The problem is that it knows it for you, not for the business.
- Running your business on AI means three things that live outside any one conversation: memory, roles and rituals.
- What you get back is consistency, not speed. The same job to the same standard, whoever runs it and whatever kind of week it is.
- The first step takes ten minutes and the foundation takes about an hour. Jump to where to start.
Running your business on AI means the knowledge your AI works from belongs to the business rather than to one person’s chat history. Most founders I work with have already done the hard part. They have a long-running conversation that understands their customers, their pricing and how they like things written, and what comes out of it is decent. The catch is that the memory lives in the thread rather than in the business. Start a new chat and it’s gone. Ask a colleague to do the same job and they start from nothing. You have built something useful that you cannot hand to anybody.
The fix is to move that knowledge out of the conversation and into the business, in three layers. Memory is what the system knows about you. Roles are who it’s being when it does a piece of work. Rituals are the jobs it does without anyone remembering to ask. I opened Leeds Digital Festival this month with a talk on exactly this, and the rest of this piece is the argument behind it.
Quick links
- Why doesn’t the AI you’ve built belong to your business?
- What’s the difference between using AI and running on it?
- What does a role do that a prompt doesn’t?
- Why do rituals matter more than rules?
- Do you need to be technical, or an expert?
- What should you expect to get back?
- Where do you start?
Why doesn’t the AI you’ve built belong to your business?
Because it grew up inside one person’s account, one conversation at a time. That’s how almost everyone learns to use AI, and it’s the reason the value stays personal.
At 8am on the first morning of Leeds Digital Festival I asked a packed room of founders, directors and operators two questions. Who has a chat or a project that knows their business? Most hands went up. Who could hand it to a colleague tomorrow and get the same answer out of it? Almost nobody. That drop took about twenty seconds, and it was the whole talk.
The same pattern shows up at scale. When Microsoft and LinkedIn surveyed 31,000 people in 2024, most of those using AI at work had brought their own tools with them rather than being given them by their employer.
78%
of people using AI at work were bringing their own AI tools, which means the knowledge those tools build up sits with the individual, not the business.
Microsoft and LinkedIn, 2024 Work Trend Index, 31,000 people across 31 countries
The commercial cost is quieter than it looks. If the good output only happens when you personally are the one asking, you’ve built one more thing only the founder can do. It doesn’t scale, it doesn’t survive a holiday, and it walks out of the door with whoever built it. Every capable business owner I’ve said that to has recognised at least one of these: the new chat where the quality fell off a cliff, the colleague who couldn’t get the same result, the context limit that lost the lot.
What’s the difference between using AI and running your business on it?
Using AI means asking it things. Running your business on it means the business has written down what the AI needs to know, who it should be for each kind of work, and what it should do on a schedule, so the result no longer depends on who’s asking or when.
Worth being precise about
An AI operating system
The version of what you’ve already built that survives a new chat, a different person and a bad week. You don’t buy one. It sits underneath whichever tools you already use.
If what you have only works in one thread, for one person, when they remember to use it, you have a useful habit rather than an operating system.
Each of the three layers fixes one specific way the thread-based version breaks.
| Layer | What it is | The human version you already know | What breaks without it |
|---|---|---|---|
| Memory | What the system knows about the business, written down once and read by every job | The induction pack for a new starter | Every new chat starts from nothing |
| Roles | Who the system is being for a piece of work, defined so anyone can call on it | A job description | Nobody else can get the same answer you do |
| Rituals | The work it does on a schedule, whether or not anyone remembers to ask | The chair’s summary at the end of a board meeting | It only happens on the days you’re on top of things |
On a phone, scroll the table sideways.
The middle column matters more than it looks. None of this is new. Every layer has a human version that well-run businesses have used for decades. The structure is human. Only the execution is AI.
What does a role do that a prompt doesn’t?
A role holds a standard and applies it every time, including when you’d rather it didn’t. A prompt gives you what you asked for. A role gives you what the job requires.
The clearest example I have is pricing. I have a role that sits across the business watching for underpricing, scope creep and work being quietly subsidised, and it checks figures I put in myself. There’s no exemption for the boss. I was pricing some work for a large institution and put in the number that felt about right. Unprompted, it went and did its own research and walked me somewhere considerably higher. It was right.
The same role has told me, six times now, that I’m delivering too many hours to one particular client. It gets more pointed each time. I’m still ignoring it, for my own reasons, and that’s the point. The judgement stays with me. What changed is that I can no longer not know.
It works the other way too, which is the honest bit. I asked my own system four times whether the slides for that Leeds talk were any good. Four times it said they were fine. A human looked at them for thirty seconds and told me they were shouty. An AI that only ever agrees with you is a mirror, not a colleague, which is why the checking, including the humans in it, is part of the system rather than a nice-to-have.
Why do rituals matter more than rules?
Because systems don’t die from being wrong. They die from being optional. A rule that only works when someone remembers to follow it doesn’t work. You just haven’t noticed yet.
I found the proof in my own business. There’s a convention for how time entries get written. Over the same six months, following the same rule, the entries the system wrote itself met it every time. The ones I wrote by hand mostly didn’t.
11 of 11, against 2 of 57
entries meeting the same writing convention, when the system wrote them against when I did. Same rule, same six months, same person.
Jonny Ross’s own business records, one six-month period
Adherence turns out to be a property of the route the work takes, not of the person or the rule. That’s what a ritual is: moving the job onto a route where the standard is applied whether or not anyone is paying attention. It’s also why I’d set up one ritual before I’d write a single clever prompt.
Do you need to be technical, or an expert in every discipline?
No to both. The work is mostly writing down how your business works, in plain English.
I resisted this for years. I never believed in writing down missions, visions and values. I thought you didn’t really need them if you knew your own business. What AI did was brick-wall me. It turns out you need to document everything, because anything that only exists in your head is invisible to the system and to everyone else.
The expertise question is the one that stops most founders, and it has a simple answer. Good practice in most disciplines is already well written down, and a role can be built to work to that standard from day one. You bring the context about your business and the judgement. You don’t have to bring the expertise.
Writing it down has one more benefit people don’t expect. Because the knowledge sits in plain documents rather than inside any one product, it isn’t locked to a supplier. If the tool I use stopped working this afternoon, I could pick the whole thing up and start again on Copilot, Gemini or whatever comes next this evening, without losing anything.
What should you expect to get back?
Consistency, not speed. The same job done to the same standard whether you’re fresh on a Tuesday morning or wrecked on a Friday afternoon.
This matters because expecting speed is the most common reason people decide AI hasn’t worked for them. Setting it up properly is slower than asking a quick question. What you’re buying is the removal of your worst days from the output, and for a business that sells on reputation, that’s worth far more than a few saved minutes. If you want the wider view of where AI helps a marketing function and where it doesn’t, I’ve written about whether AI will fix your marketing and why more AI tools rarely means more growth.
Where do you start?
With a document, not a tool, and not five roles at once. Write down what the AI needs to know about your business, then build one role you’ll actually use. One role in daily use beats a full organisation chart you never open.
The quickest way to get the document started is to let the AI you already use interview you.
Ten minutes, today. Ask your AI these four, in this order
- What do I do each week that you could help with?
- What do you need to know about my business to be useful, that you don’t know now?
- What am I not asking you that I should be?
- What should you be asking me?
Keep the answer. It’s the outline for the document everything else reads, and it’ll be better than the one you’d write from a blank page.
From there, turning that outline into a proper document takes about an hour, done once. If you’re weighing up whether this needs a dedicated AI hire rather than an hour of your own time, I’ve set out what scaling businesses need before they hire a Chief AI Officer. For most, it’s this first.
Frequently asked questions
Which AI tool should I use to run my business on?
Whichever one you already use well. The value sits in the documents the tool reads, not in the tool, so the choice matters far less than people think. Written in plain English, the same memory and roles can move between tools if you ever need to switch.
How long does it take to set up?
The first step takes about ten minutes and the core document takes around an hour. Roles and rituals build up from there, one at a time, as you find jobs worth handing over. Trying to set up everything at once is the most reliable way to stall.
Is this only for larger businesses?
No. It’s most valuable in small businesses, where so much knowledge sits in the founder’s head. A written memory and one reliable role give a five-person business the kind of consistency that usually takes a management layer to achieve.
Does the AI make the decisions?
No. It holds the standard, flags what it sees and does the recall. The judgement stays with you. The difference is that the things you’d rather not see become much harder to miss.
Get the workbook that goes with this
The AI Operating System Blueprint is the six-page workbook from the Leeds Digital Festival session. Six things to build, in the order that survives a normal week, with space to write. Free, no sign-up.
If you’d rather have one idea like this in your inbox each fortnight, The Fractional CMO Brief is where I write them.
About the author
I’m Jonny Ross, a Fractional CMO working inside the leadership teams of scaling businesses across Yorkshire and the North. I’ve spent 27 years in search and visibility, and I run my own businesses on the approach in this article.
Last reviewed September 2026
