In short: You don't sell technology, you sell an outcome in a niche. The five building blocks: a niche, one offer, visible proof, a channel for first contact, and fulfillment that runs without you. The bottleneck is almost never the AI – it's first contact and fulfillment.
What you're really selling
Nobody buys “AI”. People buy a state of affairs: more inquiries, less typing, appointments that don't fall through, quotes that go out the same day. The technology is the how, and it doesn't belong in the first call.
The sentence that works in sales has three parts: for whom, which outcome, by when. Everything else – models, tools, providers – only matters once someone wants to buy.
The five building blocks
1. A niche you can explain
Not “small and medium-sized businesses”, but a group whose working day you can describe: restaurants, dental practices, roofers, driving schools, solar sales teams. The reason isn't marketing, it's fulfillment: in a niche, the tenth job is the same as the first – and only that makes the business sustainable.
2. One offer, not seven
One outcome, one price, one process. Seven offers mean seven ways to deliver and no practice in any of them. My own example: a website plus integration for restaurants at €799 one-time and €49 a month – a sentence you can finish on the phone.
3. Proof people can see
The biggest difference between “I can do that” and “here it is”. Before the first conversation, I built 616 finished demo websites – every business could look at its own result before talking to me. That's where AI really makes the difference when you're starting out: proof at scale, before the sale.
4. One channel you work every day
Letters, phone, network, content – one of them, consistently. I sent 522 letters and followed up afterwards, because paper actually reaches these businesses and, legally, it's the least critical route in Germany. On the legal side of each channel, see Using AI in sales.
5. Fulfillment that runs without you
The difference between a service business and self-employment. Templates, fixed steps, a checklist – and whatever repeats gets automated. Not the sales conversation, but the groundwork. Skip this and after five clients you won't have time for the sixth.
What it takes in Germany
I work in Germany, so this is the German checklist. The details differ from country to country; the categories – registration, outreach rules, data protection, legal notice, advertising claims – show up almost everywhere.
- Register a trade (Gewerbe), choose a legal form, check the small-business VAT rules – questions for a tax advisor, not for me.
- First contact: marketing emails generally require consent under Section 7 of the Unfair Competition Act (UWG), including B2B. Calling businesses may be allowed in some circumstances, calling private individuals isn't.
- GDPR: if you process data for clients, you need a data processing agreement – in both directions: with the client and with the providers you use.
- Legal notice (Impressum) and privacy policy on your own website, from day one.
- Advertising claims: whatever you claim about results, you have to be able to back up. That's not just decency, it's competition law.
This is my practical overview, not legal or tax advice. Two hours with a lawyer and a tax advisor are the cheapest line item in the plan at the start.
What really eats your time
| Expectation | Reality |
|---|---|
| Building the offer | describing the outcome you're selling |
| The tech at the client's | the data the client doesn't have or won't hand over |
| Finding clients | reaching clients – every day, for weeks, without a reply |
| The first job | the fifth, if fulfillment isn't in place |
| Negotiating the price | defining the outcome so there's nothing to negotiate |
What I advise against
- Reselling courses you haven't done yourself. It works briefly and costs you your name.
- Income promises. Not to clients and not to yourself. What you can control is conversations per day and results per job.
- Collecting tools instead of selling. Twenty subscriptions aren't a company. One offer someone has paid for is.
- An agency without a project of your own. If you haven't built anything, you have nothing to show in the first call – and showing beats every argument.
And the one question I ask everyone at the start: which process have you already automated for yourself? If you can answer that, you have proof. If not, start there – see AI automation for business.
Frequently asked questions
Do I need to know how to code to start an AI agency?
To build, not necessarily anymore; to take responsibility, yes. If you can't read code, you can't check what you're installing at a client's. My way: build something for yourself first, then sell – not the other way round.
How do I find my first niche?
Where you can describe the working day – from an old job, your network, your family's trade. Not the niche with the highest price, but the one where you can deliver the tenth job like the first.
How do I get my first clients?
One channel, worked every day, with visible proof. For me that was letters plus follow-up, with a finished demo for every recipient. The number that counts isn't the number of channels but the number of first contacts per day.
How much can you earn with an AI agency?
I deliberately don't give a number. What a business like this brings in depends on niche, price, first contacts per day and close rate – in other words on your work, not on the industry. Any number I named would be a claim about you. What you can control is conversations and fulfillment; the rest follows from that, or it doesn't.
What do I need legally to get started?
In Germany: trade registration, a settled legal form and VAT question, a legal notice and privacy policy, data processing agreements with clients and providers, and a first-contact channel that complies with Section 7 UWG. A tax advisor handles legal form and tax, a lawyer handles the channel. Other countries have their own equivalents.
Is it still worth it in 2026, or is the market saturated?
The market of people who talk about AI is saturated. It's thin among those who deliver a finished outcome in a niche and take over the operation afterwards. The difference is fulfillment, not timing.
Mario De Luca I build sales systems and AI-powered software for businesses and founded the Clone Looping System (CLS). Every figure in these guides comes from my own projects, not from studies. Anything I haven't measured myself is labeled as an estimate. More about me · Journal
Who's writing this
I'm Mario De Luca, an AI systems builder from the DACH region and founder of NextCore and the Clone Looping System. My story, my projects and the principles behind these guides are on my about page. In the journal I write about what I've built, tested or messed up myself.
About Mario De Luca →