Short answer: most AI automation work is four things - connecting systems that don't talk, answering repetitive questions, extracting data from documents and messages, and routing work to the right person. Real projects land between €900 and €5,000. Anything sold as "AI transformation" without naming one of those four is a slide deck.
The category grew faster than its standards. A year of YouTube courses produced a lot of agencies whose main asset is a landing page, and the honest ones are hard to spot from outside.
The four things that actually get built
System-to-system plumbing. A lead comes in by email, someone types it into the CRM, then copies the details into an invoice, then messages the team. That's not AI - that's a workflow, and it's usually the highest-value first project because the savings are countable. €900-1,800.
Question answering on your own material. An assistant that answers from your price list, policies and docs instead of making things up. €1,500-2,500 for one channel.
Getting structure out of mess. Turning inbound emails, PDFs, forms and voice notes into fields your systems understand. This is where AI genuinely beats classic automation, because the input is unpredictable. €1,200-3,000.
Routing and qualification. Deciding which enquiries deserve a human now, which get an automatic answer, which go to the waiting list. €2,000-4,000 with CRM integration.
Everything else offered in this space is a variation on those four.
What the good ones do differently
They ask what you do before telling you what they'd build. A first call that opens with their capabilities rather than your process is a bad sign.
They quote per project, not per "AI transformation package."
They tell you when you don't need them. Half the requests I get for AI agents are better solved with a scheduled workflow costing a fifth as much - and saying so has never cost me a client worth having.
They hand over the accounts. Your n8n instance, your API keys, your CRM. If the automation lives inside their account, you're renting your own operations.
They measure something. Hours saved, response time, percentage of enquiries handled without a human. A project with no number attached can't be judged, which is convenient for whoever built it.
The pricing models you'll meet
Fixed project price. Clearest for both sides. Works when the scope is genuinely known - one process, defined start and end.
Monthly retainer. €500-3,000. Legitimate when someone is actively running and improving things. A problem when it's charged for automations that need no attention. Ask what happens in a quiet month.
Setup plus maintenance. The common structure and usually fair: a build fee plus a smaller monthly amount for hosting, monitoring and small changes.
Revenue share or per-outcome. Rare, and usually more complicated to verify than it's worth.
The model I use: fixed price for the build, €50/hour for changes, no mandatory retainer. Quiet months cost nothing.
Red flags
"We'll audit your entire business for AI opportunities" as a paid first step, for four figures, delivering a document. The audit is a sales process wearing a deliverable's clothes.
No mention of what happens when the automation fails. Every integration eventually hits a downed API, a changed field, a rate limit. Without retries, alerting and readable logs, failures happen silently - and silent failure in a lead pipeline is worse than no automation.
Everything running on their accounts.
Deliverables described in outcomes only ("30% efficiency gain") with no description of the actual workflow.
Heavy emphasis on the tool. Whether it's n8n, Make or plain code matters much less than whether the person understands your process. The tool comparison is a real question, just not the first one.
What a sensible first project looks like
One process. The one that costs the most hours, not the one that sounds most impressive.
Delivered in two to four weeks, in production, with a before-and-after number. Then you decide whether there's a second.
I've watched companies spend €15,000 on a program of AI initiatives and get nothing to production, and I've watched a €1,200 email-to-CRM-to-invoice chain give back 40 hours a month within three weeks. The difference wasn't budget or technology. It was scope.
FAQ
What does an AI automation agency actually do? In practice, four things: connecting systems that don't talk to each other, answering repetitive customer questions from your own material, pulling structured data out of unstructured input, and routing work to the right person. Everything else in the pitch is usually a combination of those.
How much does AI automation cost for a small business? A first project connecting two or three systems runs €900-1,800. An assistant answering from your documents is €1,500-2,500. Running costs are typically €20-100 a month for hosting and model usage, paid directly to the providers.
Is n8n better than Make or Zapier? For anything with volume or custom logic, n8n self-hosted is usually cheapest and most flexible; Make is friendlier for visual, medium-complexity work; Zapier is fastest to start and most expensive at scale. The choice matters less than whether the person building understands your process.
Do I need AI, or just automation? If the input is predictable - a form field, a row, a webhook - plain automation is cheaper, faster and more reliable. AI earns its cost when the input is messy: free-text emails, documents in different formats, voice notes, questions phrased fifty ways.
How do I measure whether it worked? Pick the number before you start: hours spent on the task per week, average response time, or percentage of enquiries handled without a human. Measure it for two weeks before the build and again a month after. Projects without a baseline can't be evaluated, only defended.
Not sure whether your case needs AI or just a workflow? Tell me what your team does by hand - I'll say which of the four it is, and what it would realistically cost.
