The SME Operating Model for an AI-Native Future
An essay on how AI is quietly rewriting what a small business is capable of — and why the owner no longer has to be the thing holding it all together.
The short answer
The traditional small-business operating model made the owner the integration layer — the human quietly stitching together people, providers and software that were never designed to talk to each other. AI is rewriting that model. The emerging shape is different: AI does the repeatable work, specialists handle the exceptions and the judgement, shared context surfaces what siloed providers miss, a trusted advisor holds the whole picture, and the customer simply buys the outcome. The practical consequence is that a small business can now carry the operating capability of a much larger one — which means owners need to think differently about what to build, what to keep human, and what a small business can now be.
THE OLD MODEL
For decades, the owner was the integration layer — and nobody called it that
Here's the thing almost nobody names out loud, because it's so ordinary it becomes invisible.
If you run a small business, you've probably spent years being the connective tissue between everything — the bookkeeper who doesn't quite know what the accountant knows, the accountant who's never spoken to the lawyer, the ops person carrying half the context in their head, and three or four bits of software that each hold a slice of the truth and none of the whole. You are the one who remembers that the invoice dispute from March is related to the supplier issue from May, which is why the cashflow looks odd now. Nobody told you this was your job. It just accreted, quietly, until one day you realised that if you stepped out for a fortnight the whole thing would start to wobble — not because your people aren't good, but because you're the only node connected to every other node.
I've come to think of this as the real operating model of most small businesses, and it's worth saying plainly: the owner is the integration layer. You are the API between your providers. You are the shared memory the software doesn't have. You are the exception-handler, the escalation path, and the person who notices when two supposedly-separate things are actually the same problem wearing different clothes. It works — right up until it doesn't scale, and it never really scales, because the one component you can't clone is you. This is what I've elsewhere called the coordination tax (see /coordination-tax): the compounding, mostly-invisible cost of being the human glue, paid not in a line item but in your attention, your evenings, and the ceiling it quietly puts on how big or how calm the business can get.
And for a long time the honest answer to "how do I fix this?" was: you don't, really. You could hire — but a good general manager is expensive and hard to find, and half of what they'd do is the same stitching you're already doing, just delegated. You could buy more software — but every new tool tends to add another silo to integrate rather than remove one. You could grow into being able to afford proper specialist depth — but that's a chicken-and-egg problem, because you can't grow past the ceiling that the coordination itself imposes. So most owners just carried it. I did. I'd wager you have too.
WHAT ACTUALLY CHANGED
AI didn't add a tool — it removed a constraint that shaped the whole model
I want to be careful here, because the space is thick with hype and I've got no interest in adding to it.
When people say "AI is changing everything", the instinct — a healthy one — is to reach for the pinch of salt. Most of what gets sold under that banner is a feature bolted onto a tool you already had, and a bolted-on feature doesn't change your operating model any more than a faster kettle changes your kitchen. So let me try to say precisely what I think has actually changed, because I don't think it's the tools. I think it's a constraint.
The constraint that shaped the old model was this: repeatable knowledge work was expensive and slow, so it had to be rationed. You couldn't afford to have someone reconcile every transaction the moment it landed, chase every exception the day it appeared, read every contract clause against every other one, or keep a live, connected picture of the whole business updated by the hour. That work existed — it just got batched, deferred, and mostly done by you at 9pm because you were the cheapest marginal option. The rationing of that work is exactly why the owner had to become the integration layer in the first place. There was no other affordable place to put it.
What AI has genuinely done — and I'd argue this is the real shift, not the chatbots — is collapse the cost and latency of a huge slice of that repeatable, context-heavy work to something close to zero and something close to instant. Not all of it. Not the judgement, not the relationships, not the genuinely novel decisions — I'll come back to that, because it matters enormously. But the reconciling, the chasing, the cross-referencing, the surfacing of "hang on, these two things are related" — the very work that made you the glue — is now work that a well-designed system can do continuously, in the background, without getting tired or forgetting March. Once that constraint lifts, the model that was built around it doesn't just get faster. It can be a different shape. That's the part I think owners are underrating.
THE EMERGING SHAPE
The new model has five parts, and the owner is deliberately not the centre of it
If the old model put you in the middle of everything, the emerging one is designed, on purpose, to take you out of the middle — not to sideline you, but to free you to do the handful of things only the owner can do. In my view it has roughly five parts, and the interesting bit is how they fit together rather than any one of them on its own.
AI does the repeatable work
The reconciling, categorising, chasing, drafting, cross-referencing, monitoring — the high-volume, rules-and-patterns work that used to be rationed because it was expensive — runs continuously in the background. Not as a novelty, but as the default engine of the operation. This is the part that only recently became genuinely possible, and it's what frees up everything downstream.
Specialists handle exceptions and judgement
When something falls outside the pattern — a genuinely tricky tax position, a contract that needs a human read, a decision with real consequences — it goes to a person with actual expertise. And crucially, that person spends their time on the exceptions rather than the grind, which means you can afford deeper expertise than you could when specialists were buried in routine work.
Shared context surfaces what silos miss
Instead of each provider holding a slice, the system carries a connected picture — so the thing your bookkeeper sees and the thing your advisor needs to know are no longer separated by a gap that only you were bridging. This is the heart of the connected back office (see /connected-back-office): the context that used to live in your head now lives in the system, where it can actually be acted on.
A trusted advisor holds the whole picture
Someone whose job is explicitly to hold the whole — to notice the pattern across the parts, to bring you the two-or-three things that actually need you, and to be accountable for the outcome rather than a task. This is the human role that becomes more valuable, not less, once the repeatable work is handled: judgement, relationship and synthesis, which are precisely the things AI doesn't do.
The customer buys the outcome
You don't buy "bookkeeping" and "tax" and "a piece of software" and then integrate them yourself. You buy the result — books that are right, obligations that are met, a clear picture you can act on — and the stitching happens underneath, where it belongs. The integration stops being your job and becomes part of what you're actually paying for.
THE REAL ARGUMENT
A small business can now carry the operating capability of a much larger one
This is the claim I most want to land, because I think it's both true and genuinely new — and I've sat with it long enough to be reasonably confident it's not just optimism talking.
For most of business history, operating capability scaled with size, and roughly in one direction only. Big companies could afford dedicated finance functions, real specialists, proper systems, someone whose whole job was to keep the picture connected. Small businesses couldn't, so they made do — the owner as glue, the part-time bookkeeper, the once-a-year accountant, the software that half-worked. The gap between what a small business could operate like and what a large one could was structural, and it was mostly a gap of affordability: the capability existed, you just couldn't buy it at your scale. I've written about this before as the back-office capability gap (see /back-office-capability-gap), and for a long time it looked like a permanent feature of the landscape.
What's changed is that the economics underneath that gap have shifted. If the repeatable work — the expensive-to-ration part — now runs at near-zero marginal cost, then the thing that made real operating capability a large-company luxury has substantially eroded. A small business can now have continuous reconciliation, live financial context, exception-based specialist attention, and a genuine picture of the whole — not because it hired forty people, but because the model no longer requires forty people to deliver it. That is a different world from the one most of us built our businesses in, and I don't say that lightly. It means the ceiling that the coordination tax imposed isn't fixed anymore. It means the question is no longer "can I afford to operate like a bigger business?" but "do I understand that I now can, and am I building for it?"
I'd add one honest qualification, because I don't want to oversell it. This isn't a switch you flip, and it isn't free of judgement — if anything it demands more of it. The technology (see /technology) is necessary but nowhere near sufficient; a pile of AI capability with no shared context and no accountable human on top is just a faster way to make a mess. The capability is now buyable at small scale, but assembling it into something that actually holds together still takes design, and still takes people who care whether it works. So the opportunity is real, but it's an opportunity to build something better, not a thing that arrives on its own.
WHAT THIS ASKS OF OWNERS
The harder question isn't what to automate — it's what to keep stubbornly human
If you accept even half of the above, the interesting decisions change, and they get more interesting, not less.
The lazy version of this conversation is "automate everything you can." I think that's not just wrong but backwards. The whole point of moving the repeatable work to AI is to free up scarce human attention for the things that genuinely need a human — so the design question isn't "what can I automate?" but "what should I now be able to keep human, because the grind is no longer eating the time it needs?" In my experience the answer is the things that were always the real work and got crowded out: the judgement calls, the relationships, the decisions that carry consequence, the exceptions where being right matters more than being fast. Those don't get cheaper or more automatable — they get more central, because they're what's left when the routine falls away.
There's also a temptation, once you see the capability, to try and build it yourself — to go and assemble the tools, wire up the context, become the systems integrator on top of everything else you do. I'd gently push back on that, and not because it can't be done, but because it quietly reinstalls the exact problem you were trying to escape. If you become the person who holds the AI stack together, you've just swapped being the integration layer for people and providers for being the integration layer for software instead. Same ceiling, shinier tools. The move that actually changes the shape is to have the whole thing — the AI engine, the specialist depth, the shared context and the accountable human holding it — delivered as a connected outcome, so that the integration genuinely stops being your job. That's the difference between bolting AI onto the old model and actually operating on the new one.
And I want to be honest about the uncertainty, because pretending to certainty here would be its own kind of dishonesty. Nobody fully knows how far this goes or how fast. The tools are improving faster than anyone's org charts, some of what's promised won't pan out, and the right posture is neither breathless enthusiasm nor arms-crossed dismissal — it's paying attention. What I'm fairly confident of is the direction: the constraint that made the owner the glue has loosened, it's not tightening again, and the owners who do best over the next few years will be the ones who stop asking how to survive the coordination tax and start asking what they'd build if they no longer had to pay it. That's a better question to be sitting with, and it's genuinely available to you now in a way it simply wasn't a few years ago.
| Comparison dimension | Dimension | The old model (owner as glue) | The emerging AI-native model |
|---|---|---|---|
| Who integrates the parts | The owner, in their head and after hours | Shared context in the system, held by an accountable advisor | |
| Who does the repeatable work | Owner or a stretched part-timer, batched and deferred | AI, continuously and in the background | |
| What specialists spend time on | The routine grind, with exceptions squeezed in | The exceptions and the judgement calls | |
| Where the whole picture lives | Fragmented across providers and tools | Connected, live, and act-on-able | |
| What the owner is buying | Separate services and software to stitch together | An outcome, with the integration underneath | |
| Ceiling on capability | Scales with size and the owner's bandwidth | No longer bounded by scale in the old way |
If this is a question you're sitting with, I'd be happy to talk it through
There's no pitch waiting at the end of this. If the idea that your business could operate differently — that you might not have to be the thing holding it all together — is one you've been chewing on, I'd genuinely enjoy a conversation about what that could look like for you. No rush, no hard sell. If it's useful to talk it through, reach out and we'll have a proper chat about where you are and whether any of this is worth building toward. And if you'd rather just read more first, the pieces on the connected back office, the back-office capability gap and the coordination tax go deeper on the parts of the argument.