AI just made 7 out of 10 the new average. Any service you sell, your client can now get a 70% version from AI in an afternoon. Here's why that doesn't terrify me, why we built our own operating system instead of renting more SaaS, and the maths behind the business model I think wins the next decade.
Every service business is about to face the same question: if your client can get a 70% version of what you sell from AI in an afternoon, what exactly are they paying you for?
The businesses that will thrive all answer it the same way: with a service powered by software they built themselves. I call it Service With Software.
Service With Software is a service powered by software in the backend. Not a SaaS company you licence from someone else. Software you built yourself, that your team is trained on, built on your data, your core competencies and your way of working, tailor-made to how you deliver outcomes.
The service is what the client buys. The software is why it's better. And the software is never for sale, because it only makes sense inside the service it powers.
That's the whole model in three sentences. The rest of this page is the detail: what changed, what qualifies, what it looks like in practice, the maths, and who should be building one.
I interviewed Jason Pallant recently and he framed something I'd been circling for months: AI output is the new average. The new 7 out of 10.
Think about what that actually means. Any service you sell, your client can now get a 70% version from AI in an afternoon. An SEO audit. A content plan. A strategy deck. A financial model. Not a perfect version, but a competent one, and it gets more competent every month.
For fifteen years I've watched businesses pay agencies real money for work that was, if we're honest, a 6 out of 10. That market is dead. If AI produces a 7 for close to free, nobody pays for a 6 ever again. The entire industry for average work just evaporated, and a lot of service businesses are still pricing like it didn't.
The flip side matters more. Getting from a 7 to a 9 or a 10 is a completely different process now, and that last stretch is where humans live. It has never been worth more.
The SaaS industry was built on an assumption that held for thirty years: building software is expensive. It took hundreds of thousands of dollars and a team of engineers, so you rented someone else's tool and bent your business around how it worked.
That assumption is crumbling. My team now ships internal tools in days. Plenty of things we used to pay monthly subscriptions for, we now build ourselves, shaped exactly to how we work.
To be clear, I'm not saying software is dead. Deep platforms, systems of record, tools with real network effects: they're fine, and we still happily pay for them. What's changed is the layer of generic tools underneath, and more importantly, what it costs a service business to own its own stack. The question stopped being "which tool do we subscribe to" and became "what should we build that's ours".
Put those two shifts together, AI making 7 out of 10 free and software getting cheap to build, and the logical move for a service business changes: stop renting generic tools, start building your own.
Code can be copied. Data has to be accumulated. People have to be built and kept. That order is the strategy, and it's why the model defends itself: a competitor can clone your features in a week, but they can't clone your years.
Let me be clear about the boundary, because every agency on earth is about to claim they have proprietary AI.
A workflow in ChatGPT is not software. A white-labelled dashboard is not your software. A prompt library is not an operating system.
The bar: your team built an actual application, on your own data, with your own workflows, that your business runs on. Slack is the famous example of the shape. A games studio built an internal communication tool because they had a communication problem, and the tool turned out to be a whole piece of software in its own right. That's the bar. Something real enough that it could stand alone, even though (and this is the difference in this model) you never let it.
So the minimum is all three: owned software, proprietary data feeding it, custom workflows around it. Two out of three is a head start, not a moat.
Our answer was to build our own operating system. HawkOS is the software that powers how StudioHawk delivers SEO. Every client, every integration, every workflow, and the learnings from hundreds of campaigns, all in one platform.
Before, our client knowledge lived everywhere: the CRM, the project management tool, call recordings, onboarding docs, briefs, keyword data across half a dozen sources. Workflows were segregated and knowledge sat in silos. HawkOS centralises all of it into one source of truth per client, so everything we do pulls from the same database and stays consistent.
Then the compounding starts. Take information architecture, one of the most valuable things we build for clients. It used to take dozens of hours of senior time. It now takes a couple of hours, because the system holds all the client's data, knows how we structure an IA, and has learned from every IA we've ever built. Each one makes the next one better. That's the data lake working: not storage, compounding.
We will never sell it. It isn't a product, it's the service. It's tailored to exactly how we work, and that's the point. The system sits at the core of what the service does, and every campaign we run makes it smarter.
There's a reel from Jamie Brindle doing the rounds about the client who says "we'll just use AI". If you run a service business, you've heard some version of that sentence this year, or you're about to.
So, could they? Honestly? Yeah. And this is where most "proprietary tech" stories fall apart, so let me be straight about it.
Building the software is the easy part now. The hard part is orchestrating it to how you do what you do. Knowing what to build took us fifteen years and hundreds of campaigns. The learnings baked into our system can't be replicated in an afternoon, because the input isn't code. The input is reps.
And even if a client used our exact version, without being trained on it, it's just a tool. I can hold a hammer. It doesn't make me a builder.
This model sits in the middle of a spectrum that's worth understanding if you run any kind of service business.
The middle model: better margins than a pure service, more defensible than a thin SaaS, and it compounds with every campaign.
A pure service scales with headcount. Every new dollar of revenue needs more people. Benchmark most agencies past 20 people and the gross margin lands somewhere in the 30-50% band, before the overhead that comes with all that headcount.
A pure SaaS scales with servers at 80%+ gross margins, which is why everyone spent a decade trying to build one. The thin point solutions will feel pressure as customers realise they can build a 70% version in-house, but the deep platforms, the systems of record, the products with real network effects or data nobody else holds, keep their economics.
Service With Software sits in between: roughly 50-70% margins, scaling with your data. You still need people, because people make the calls AI can't. But every campaign makes the system better, and the system makes every person more effective. The service funds the software, the software compounds the service.
"The moat isn't the code. It's not even the data. It's the people trained on both."
So if anyone can build software, what actually protects you?
Not the code. A competitor can copy a feature in a week. The data lake is part of it: hundreds of campaigns of what worked and what didn't, structured and feeding one system, compounding every single day. Is our data moat insane? No, and I'd be suspicious of anyone who tells you theirs is. Is it good? Yes, and it gets better every day.
But the data isn't the whole moat either. The moat is really in the people. The system is only as good as the operators making calls with it: people who've done the reps, who know which of its answers to trust and which to override, who take what it produces at a 7 and turn it into a 10. Software plus data without those people is a very expensive dashboard.
Code, data, people. One can be copied, one has to be accumulated, and one has to be built and kept. That order is the strategy.
Here's the part of this model I find genuinely interesting. The lighter touch your service gets, the better your margin, and the more replaceable you become.
Automate 90% of your delivery and your profit looks brilliant right up until the client realises they can automate the same 90% themselves. Unless you're protected by a real data moat or a legislative one, light-touch is just a countdown to becoming a commodity.
Which means the people matter more, not less. The strategic decisions, the forward calls, the judgement on what to do next quarter rather than what happened last quarter: that's the layer that makes the whole thing defensible, and it needs genuinely good people running it.
Strauss Zelnick, the CEO of Take-Two (the company behind Grand Theft Auto), was asked whether AI could build the next GTA. His answer stuck with me: AI is backward-looking, it can only compute from data that already exists, while big hits are forward-looking and have to be "created out of thin air".
My shorthand for it: AI is brilliant at progression and terrible at prediction. It will take what exists and extend it, summarise it, execute against it, all to a solid 7 out of 10. What it won't do is the forward-thinking creative leap. The call that doesn't follow from the data. In services, that's strategy, and it's still entirely human.
AI gets everyone to a 7. People get you to a 10.
I think this is the future of most service businesses. If you're five people or more, you should be working on some kind of software layer under your service, and the build cost has fallen far enough that you have no excuse not to start with one tool.
Below that it gets harder. I think this genuinely disrupts freelancers: not because they lack skill, but because they don't have the time or the accumulated data to build a moat like this. A solo operator using AI well is exactly the "new 7 out of 10" the market now gets for cheap.
Where's it going next? Anywhere a service runs on repeatable judgement over accumulated data. SEO and marketing are early because the data is digital and abundant. Accounting, legal, recruitment and consulting are the same shape.
It's the evolution of it. Tech-enabled meant using tools. This means owning them.
No, and the difference matters: they say replace your people with agents, I say arm them.
Because it's worth more inside the service than out of it. Sold, it's a product competing with every other product. Inside, it's the reason the service can't be copied.
Honestly, yeah. Building the software is the easy part now. What they can't shortcut is the years of accumulated results the software is trained on, or the operators who know which of its answers to trust.
If you run a service business, three moves.
Find your 7. Be brutally honest about which parts of your service AI already does to a 7 out of 10. Stop charging a premium for those parts, because your clients will work it out even if you don't. Reprice around the parts that get clients to a 9 or 10.
Start building your OS. Not a product. Not something to license. One internal tool for the workflow you repeat most often, shaped exactly to how you work. Then another. The goal is software that operationalises the way you deliver, because that's the version no competitor can buy off the shelf.
Own your data lake, then train your people on it. Start capturing and structuring everything now: campaign results, client context, decisions and outcomes. Your accumulated data is the input your competitors can't buy. But a data lake without trained operators is just storage, so invest in the people who know what to do with it as hard as you invest in collecting it.
AI didn't kill services. It killed average ones. The 8s, 9s and 10s have never been worth more, and the businesses that pair real expertise with software they own are going to take the lot.
If you're building something similar, I'd genuinely love to compare notes. Get in touch.
I'm publishing one of these a fortnight. Each one takes a piece of the model above and goes deep on it.
Subscribe to the fortnightly newsletter. No fluff, straight to the point.