The question we get most often from people wanting to start one is what the stack costs. It is the least important question of all.
The tools cost a few hundred a month and they are listed with prices in the stack. That is not where this is won or lost.
What it actually takes, in order
- NOIBeing able to write — not "being able to prompt". Unable to recognise good copy, you cannot judge what the machine produces
- NOIBeing able to sell — the first three clients do not come from a funnel, they come from people who know you
- NOIBeing able to discard — the skill separating an agency from a content generator
- AIA stack that works — necessary, and the easy part
- NOISix months of financial runway — because the first two clients pay on 60-day terms
The first three cannot be delegated or bought. The fourth is an afternoon's work. The fifth is the one almost nobody budgets for.
AI does not make you a good marketer. It makes you a fast one, and speed amplifies whatever you already are.
The real costs of the first six months
| Line | Indicative per month | Note |
|---|---|---|
| AI and automation tools | €200-400 | grows with clients, not with projects |
| Hosting, email, domain | €30-80 | negligible |
| SEO and data tools | €100-200 | necessary from day one |
| Accountant and contributions | varies | the line that always surprises |
| Your own unpaid time | the real cost | 4-6 months before steady flow |
That last row is not rhetoric. Months pass between the first client and an income that supports you, and during those months you work eight-hour days anyway. Starting without that reserve means accepting the first client who turns up, and the first client who turns up is nearly always the wrong one.
The three ways agencies close in year one
Delivering volume. The classic failure of people who understood the technical half and not the editorial one. The client receives forty pieces of content a month, converts none of them, and does not renew. They paid for quantity and wanted results.
Taking any client. Three different sectors mean three different problems and starting from zero each time, accumulating nothing. An agency is worth what it has learned, and learning requires staying.
Competing on price. When your argument is "we cost less because we use AI", somebody always turns up costing less still. A low price is not a position, it is an expiry date.
What to do in the first three months
An order that lowers the risk, and it differs from what most people do.
First pick a sector, tight enough that you could turn somebody down. Then work free or cheap for one real client, with an explicit agreement that you may publish the results: you need a case, not a portfolio of mockups. Then build the stack around that case, rather than building it first and hunting for somebody to apply it to.
And write. Not for rankings — those come later — but because writing publicly about how you work is the fastest way to show that you can. How we alternate machine and people is in the vibe marketing guide.
One thing worth knowing first
The term "vibe marketing" currently carries two meanings: the method described here, and a brand's aesthetic. Open an agency under that name and part of the work will be explaining which one you mean — including to search engines and language models, which at the moment lean toward the second.
That is not an obstacle. It is why whoever defines it well now gets associated with the term later.
Frequently asked questions
What does it cost to start an AI marketing agency?
Tools run between €300 and €700 a month across AI, automation, hosting and SEO, so far less than people assume. The real cost is the four to six months of unpaid work before client flow steadies: starting without that reserve means taking the first client who turns up, which is nearly always the wrong one.
What skills does it actually take?
Being able to write, being able to sell, and being able to discard mediocre work. The first two are obvious; the third separates an agency from a content generator. AI does not make you a good marketer, it makes you a fast one, and speed amplifies whatever you already are, for better and worse.
Do I need to code to run an AI agency?
No, and less than growth hacking demanded a few years ago. The tools configure without code. What you do need is the judgement to assess output, because the part that never gets delegated is deciding a piece is publishable, and that stays an editorial call.
How do you find the first clients?
Not through a funnel. The first three almost always come from people who already know you, and that is normal. What helps in the early months is working for one real client even at a low rate, with an agreement that you can publish the results: you need a real case, not a portfolio of mockups.
Why do most AI agencies close in the first year?
Three recurring reasons: they deliver volume instead of results and the client does not renew; they take clients across sectors too different to accumulate any expertise; and they compete on price, where somebody cheaper always arrives. A low price is not a position, it is an expiry date.
Specialise in one sector or stay generalist?
Specialise, at least at the start, and tightly enough that you could turn somebody down. Three different sectors mean three different problems and starting from zero each time: an agency is worth what it has learned, and learning takes staying in one place long enough.
Running an agency and want to work this way?
Tell us how you work now and we will say what we would change first. Seven questions, founders answer within 48 hours.
ApplyKeep reading
Founder of As Growth, an AI-first growth partner. He runs growth and strategy across partner businesses: OroEtic, Giojello, Equity Plus and others. Apply if you want to work with us.