How to Scale a Service Business in the AI Era

For twenty years, one rule went unquestioned. Services don't scale.

Investors believed it. Founders believed it. I believed it, and I ran a service business.

The rule was wrong, though not in the way you would expect. Services scaled badly for most of the people who tried, but almost never for the reason they thought. And the one part of the rule that was genuinely true, that growth in services costs you headcount, is the exact part that stopped being true in the last two years.

This is what a decade of running division5 taught me about which constraints were real and which ones I invented, and what changes now that the last real one is gone.

Most of what follows is the slow part, the decade before any of this could be automated, because the fixes only make sense once you know which constraint they were fixing. The AI section comes near the end and it is shorter than you would expect. That is the point of it.

The market decides what you build

I did not set out to run a service business. I wanted to build games.

In May 2015 I founded division5. I was 23, with no connections, no money, no references and no experience. My plan lasted until the money ran out, which did not take long. The only way to keep the company alive was to sell software development services, so that is what we sold.

That was the first lesson and it arrived fast. The market does not care what you want to do.

Most founders take this as a defeat, and I did too. You set out to build the thing you dreamed about, the market hands you something less glamorous that actually pays, and the paying work starts to feel like a detour on the way to the real business.

Treating it that way cost me the next several years.

I spent a decade trying to escape

Once we were solvent, I started trying to leave. I kept attempting to pivot the company from services into a product, and I did it more than once.

Every attempt followed the same shape. I would get excited about the product that would finally free us, pour my attention into it, and quietly stop improving the business that was actually paying salaries. The product never got far enough to matter. The service business, starved of the attention it needed, grew slower than it should have.

I did that over and over again before I understood what it was costing me. It taught me the most important thing I know about running a company. If you want to grow a big business, you have to focus.

Not follow the trend. Not follow your passion. That exciting idea you had yesterday, drop it. Don't diversify. Keep doing the one thing until you have improved every part of it. Stay with a service long enough and it compounds. Change direction and you start from zero, carrying nothing forward.

I call this cheating on your business, because that is what it feels like looking back. The energy I spent on the escape plan was energy the real business needed.

But focus alone does not explain why I wanted to escape in the first place. I had three specific reasons, and all three deserve examining, because I now think every one of them was a description of my own limitations rather than the model's.

The bottleneck had my name on it

"Services are tied to the founder." As division5 grew, I noticed our clients were attached to me rather than to the company. I was doing the communication. I was solving the problems. I was the one making sure the work was right. The obvious conclusion was that a service business would always depend on me.

The truer statement is narrower and much more useful. I had not built a structure that could operate without me. In fairness I could not have, early on, because structure costs money and we did not have any. So I did the work myself.

Then we grew, I hired, and I started building systems. The more systems I put in, the less I was needed. Founder dependency turned out not to be a property of services at all. It is a stage, and it feels permanent right up until the moment you build your way out of it.

"Services don't build an asset." The second objection sounds more technical than it is. The argument is that your output is whatever your people did last month, so if a person leaves or a client leaves, you are holding nothing.

What I missed is that the asset in a service business is the expertise, and expertise can be made structural. You capture how the work gets done, you write it down, you build it into how new people are trained. Do that properly and someone leaving is a setback rather than a hole in the company. The asset was available the whole time. I was not building it, because I was busy building a different company.

"Services don't scale." The third I repeated most often, usually to myself. I heard the stories about product valuations climbing overnight. I was struggling to grow. So I concluded the difficulty was built into my kind of business.

Every business scales. They just have different bottlenecks. Often a company does not grow because of its founder's inability to grow it, which was precisely my situation. Build the right systems, hire the right people, solve the right problems, and it moves. Blaming the model is the most comfortable explanation available, which is a good reason to treat it as the last one you accept.

Three objections, one actual cause. The constraint was not the business model. It was me, and everything that finally worked came from accepting that.

Hold on to the third one. It is the only one of the three that was ever partly true, and it is the one that stopped being true recently, for reasons that had nothing to do with me getting better at my job.

What growth actually came from

For the first four years of division5 we acquired clients entirely through word of mouth. No marketing spend, no sales function. We focused on the work and on the relationships around it.

That was not a strategy at first, it was a lack of options. A few months in I was trying to figure out how to stay alive. Without connections it is brutal, especially selling to other businesses, because most B2B work moves through networks you are not in yet. So I treated every client we did win as though the company depended on them, which it did.

Profit was not my priority then. That was probably a mistake and I left real money on the table, but I would not change it. Some clients referred us. Slowly we built a base that kept coming back and brought others with them, until demand outran what we could deliver and we had to hire.

Without meaning to, we had built a service people wanted to share.

It is a slow way to grow and it forces you to get the fundamentals right, because the mechanism only works if the work is genuinely good. To be recommended you need happy clients, and to have happy clients you need to be worth recommending. No marketing budget imposes that discipline on you.

This is also where I stopped thinking about brand as a separate activity.

As a young founder I was fascinated by Apple, by the idea that two people could build something that size. So I read everything I could find about building a brand. Some of it was about names. Some was about logos and typefaces. Those things matter a little. One thing matters far more, which is the experience someone actually has with your service.

There is a direct line between that experience and your brand. Good experience, good brand. Bad experience, bad brand, no matter what the logo looks like. Apple reads as premium because that is what people repeatedly experienced, until the experience hardened into an identity. You cannot install a perception that is not rooted in what you actually deliver. A good name will not hide poor work, and a clean logo will not cover for bad communication.

Which makes the fastest route to a strong brand embarrassingly simple. Deliver better work. And the first ingredient of that is not eloquence, it is understanding. You become a good communicator by becoming a good listener.

Being worth recommending stays vague until you make it concrete, and for us it came down to one word. Clarity.

Clarity in the process. Not just knowing your own steps, but communicating them to the team and to the client. When everyone understands how the work actually runs, mistakes drop and expectations get met instead of guessed at.

Clarity in expectations. Great service is not about meeting whatever expectation the client walked in with, it is about setting the right one at the start. Be direct about what is possible, what is not, and how long it will take. Clients value that far more than a promise you cannot keep.

Clarity when something breaks. Problems are inevitable, and how you handle them defines your service more than the smooth work ever does. Be direct, own it early, explain the situation and lay out how you are fixing it. People forgive the problem. They do not forgive being managed around it.

All three converge in the same place. Clarity builds trust, and trust is what makes someone put their own reputation on the line to recommend you. Which turns out to be the thing everything else in this piece depends on.

Then word of mouth ran out

All of that made referral work better. It did not make it bigger.

Word of mouth has a ceiling and we hit ours after about four years. From the outside everything looked stable. We had recurring clients, the work was good, the reputation was real, and we simply could not grow past that point. The mechanism that had carried us was now the thing holding us in place, because referral only travels as far as your existing clients happen to talk. You do not choose which markets you enter or how quickly. You wait to be mentioned.

There are three ways out of that and I tried all of them.

The first was partnerships, and a good local partner is genuinely hard to beat. They arrive with the connections, they speak the language, they are already trusted, and they know what the market will actually pay, which is knowledge that takes years to build from the outside. When these arrangements fail it is rarely about competence. It is a mismatch in values, expectations nobody made explicit, or communication that thins out once the novelty wears off.

The deeper problem with partnerships took me longer to see. They hand control of your growth to somebody else. If you slip down a partner's list of priorities your pipeline in that market goes quiet, and there is very little you can do about it from where you are standing.

The second was networking, which I was bad at for years because I had decided it was a personality trait I had not been issued. It is not. It is a skill, it responds to practice like any other, and treating it as something you either have or lack is a convenient way to avoid getting better at it.

The third is marketing and sales, and it is the only one of the three that leaves you holding the controls. Marketing gets people to your door. Sales gets them into the building. Both are slower to build than a partnership and neither arrives with somebody else's client list, but they belong to you, you can measure them, and when you need more you can turn them up.

There is no silver bullet in any of it, and the longer version is a story of its own. What matters here is the shape of the answer, because it is the shape of every answer in this piece. Depending on referral means depending on other people's conversations. Depending on a partner means depending on another company's priorities. The way out was to build the one thing I controlled, which is precisely what building a structure that could run without me had been.

Building that engine meant hiring for it, and at our size that meant choosing between growth and margin every time. That trade is the one that has since collapsed, though it would be another few years before I understood how completely.

Culture is the system that carries all of it

Everything above depends on systems, and in a service business the systems are people. That is why culture stops being a soft topic and becomes operational.

In mid 2020, right after lockdown, demand for our services jumped. We were around 15 people and needed to hire quickly.

Until then we had never written down a set of values. Culture was not something we had built deliberately, it was something we lived. Under about ten people you shape it directly through your own behaviour. Past that your personal influence thins out faster than you notice, and you need values that work when you are not in the room.

So I defined some, and I got it wrong. I focused on the social side of culture and lost sight of the business side. It took a couple of years going the wrong direction before I went back and started over.

The correction was to build culture around what clients actually valued, the things that made them come back, instead of around what made the office pleasant. Those are not in tension. Good results are a large part of why people feel good about their work.

We settled on three.

• Trust, which has to run both directions. If I do not trust you to do good work we cannot work together, and if you do not trust my decisions you cannot follow them.
• Responsibility, meaning ownership of the outcome and not just the task.
• Accountability, meaning a willingness to be measured against what you said you would do.

None of that is unique to us, which is the point. They are fundamentals about how work gets delivered, so they survive growth, new hires and people who never met the company when it was small.

The same principle extends outward to clients. Business gets described as a shark tank and I have never found that true. Kindness is not softness, it is leverage. Clients are used to being treated as revenue and used to suppliers who retreat behind the contract when something goes wrong. Listening when they are under pressure, owning a mistake before they find it, remembering that an angry email usually comes from someone being squeezed by their own boss. All of that builds trust, and trust is the thing that actually compounds. A satisfied client pays the invoice. A client who trusts you brings you the next three.

That is the real engine behind the word of mouth. Not a growth tactic, just trust, accumulated slowly.

Then the last real constraint disappeared

Everything so far I learned the slow way, before AI. It is worth being precise about what has changed, because it is not everything.

The old trade off was clean. Scaling a service meant hiring, which meant recruiting, training, retaining, absorbing the friction that comes with people, and constantly tightening process just to hold quality steady while growing. Services were easy to start and genuinely hard to scale. Products were the mirror image, hard to build and nearly frictionless once they worked, since serving ten thousand users cost about what serving a hundred did.

Investors were not being irrational. Given that trade off, they were right.

The trade off is what changed. The human bottleneck is dissolving. You no longer need to add analysts, writers and researchers in proportion to clients. You need agentic workflows built to carry the load instead. In 2022 doubling your client base meant doubling your delivery team. Now a small group of senior people can carry what used to take a department, with agents handling research, analysis, drafting and production.

A service business can reach margins that used to belong to software. Growth has come apart from headcount, and that was the one objection in the whole list that was ever actually true.

At the same time the other side of the trade is eroding. Building software used to be its own moat, because writing code was hard enough to keep competitors out. If your product can be rebuilt over a weekend with current tools, that moat is gone. Standard SaaS is commoditising, and positioning is far harder when your users can generate a passable version of your product themselves. I expect that to continue.

What to do about it

If you run a service business, stop operating like a traditional agency and start operating like a systems architect. Productise the work, not by selling generic packages but by building proprietary, agent driven workflows that let you deliver bespoke value at close to zero marginal cost.

What that looked like for us

In our case that meant we stopped buying software and built the system the company runs on. It now exposes more than 350 tools to our agents across thirty areas of the business, from CRM and recruitment through to finance and reporting. An agent here is not a chat window bolted onto the side of the product. Each one has a playbook, a cron schedule, and a specific set of tools it is allowed to touch.

Recruitment is the clearest example, because for us that is delivery work rather than back office admin. Thirty of those tools cover one pipeline. An agent wakes on its schedule, finds the candidates that have gone stale, pulls their email threads, redacts a CV before it leaves the building, drafts the note that presents someone to a client, and reads back the interview transcript afterwards.

What it does not do is send any of that on its own. Every tool is either automatic or queued for a person to approve, and the default is decided by one rule. If a tool only reads, it runs. If it writes, it waits for a human. That second case covers every message that would ever reach a candidate or a client, unless somebody has deliberately decided otherwise. That rule is the one that matters, and it is worth being precise about what it means. Nobody was taken out of the decision. Every message that reaches a candidate or a client still goes out because a person read it and decided it should. What stopped needing a person was the work of getting to that decision.

The effect on capacity is the part that still surprises me. Running that pipeline properly takes about five people, an operations officer, two SDRs, a recruiter and someone on HR. We run it with two, one SDR and one on HR, and they are not working longer hours to manage it. The work that went is the work nobody ever wanted, assembling context, chasing threads that went quiet, pulling email history, reformatting a CV, writing the first version of a message somebody else would rewrite anyway. What is left is the part that was always the actual job, deciding who is worth talking to and what to say to them.

It also runs better than it did, which I did not expect. A candidate going quiet for three weeks used to depend on somebody noticing. Now something watches the whole pipeline on a schedule, so the follow up that used to fall through the gap between two busy people just happens. That is the quiet part of this shift. The obvious win is doing the same work with less, and the real win is that the work stops being subject to whether anyone had a good week.

Which parts transfer

Notice that this is the same instruction as before, with the cost removed. Decouple yourself. Make the expertise structural. Fix your own bottleneck. Those were always the answers, and AI made executing them dramatically cheaper. Which is also why this favours the founders who did the unglamorous work already. If you have systems, you have something to plug agents into. If you are still doing everything personally, there is nothing there to automate.

Then own the channel that brings the work in. Referral will carry you further than you expect and then stop, and when it stops the temptation is to hand the problem to whoever offers to sell on your behalf. Build marketing and sales you control instead. This is also where the arithmetic has shifted most sharply, because research, qualification, follow up and content were exactly the parts that made an in house growth function too expensive for a company our size. What used to need a department is now within reach of a small team that has systems and knows where to point them.

If you build products, features are not a moat and your code is not your advantage. Move toward what is genuinely hard to copy. Proprietary data, deep integration into legacy systems, complex B2B workflows, or a community and brand that took years to earn.

And if you are building outside Silicon Valley, as we are in Albania, this matters more than it does elsewhere. We do not have the luxury of burning millions in venture capital to discover whether the unit economics work. We need durable businesses solving real problems, where cash flow is the measure rather than a funding announcement. The model that suits that constraint, a service business with real clients, honest fundamentals and disciplined systems, is exactly the model that just became scalable.

The ecosystem does not need more rockstars. It needs more mechanics.

What I would tell myself at 23

Stop trying to escape the business you have. The service is not the consolation prize, it is the asset, as long as you are willing to build the systems that turn it into one. The bottleneck is you and not the model, which is bad news for your ego and very good news for your options. And focus will beat every clever idea you have this year.

If I had understood that in 2015, I would have spent a decade compounding instead of a decade pivoting.

The old playbook is dead. The founders who notice that the trade off has inverted are the ones who will take the next decade.