I tried to start an automation agency with no experience

I followed the script from the "start your automation agency" videos step by step: pick a niche, put together the offer and go out and find the first client. Here's what actually happened at each step, with numbers. If I had to sum it up in one sentence: nobody buys an automated workflow, they buy a solved problem.

The script is almost always the same, and it has something of a recipe about it. Step one: pick a niche, say dental clinics or real estate agencies. Step two: build an "irresistible offer". Step three: send cold messages, that is, to people who don't know you at all. Step four: first client within a few weeks. Step five: recurring income every month and life sorted. All with an AI agency, which is what people call a small business that offers other companies AI solutions: chatbots, automations, assistants.

Why try it, when I already apply AI in client systems? Precisely because of that. I wanted to see what happens to someone who starts from zero and follows the recipe to the letter. To step into those shoes I used a new trade name, a new email and a new profile, and I banned myself from drawing on my contacts or my current clients.

A caveat on the title, because "no experience" is only half true. Technical experience I had; what I didn't have was sales experience: I had never gone out looking for clients cold, and I didn't know the industry I ended up choosing from the inside. That's exactly what the script says you don't need.

The rule of the challenge: follow the script without skipping steps or cheating with what I already know. I gave it six weeks, from 13 April to 24 May 2026.

The videos give very clear criteria for picking a niche: businesses with money, with repetitive tasks and with lots of competitors among themselves. Following them, I chose small physiotherapy clinics, with one to four practitioners.

For the research I used AI, as the rules require. I asked Claude to describe what a normal day in a physiotherapy clinic looks like, what repetitive tasks it has, what tools it usually uses and what problems an automation could solve. I also asked DeepSeek the same to compare. The answers were tidy, reasonable and very convincing: appointment management, reminders, last-minute cancellations, answers to frequently asked questions, reviews.

What the research didn't tell me, and couldn't tell me, is what matters. What that business really does every day, who makes the purchasing decisions, how much the problem hurts and how much they're willing to pay to get rid of it. Neither model told me that many of those clinics already have appointment software with reminders included, something I found out later and the hard way.

First warning: the script was going one way and reality another. And I didn't know it yet.

The offer I put together said, more or less: "Recover lost appointments. Automatic WhatsApp reminders, rescheduling of cancellations and answers to your patients' frequently asked questions, without your front desk having to do anything". The automation I proposed connected the clinic's calendar with a messaging system and an assistant that answered basic questions: opening hours, session prices, what to bring on the first day. I built it with n8n and a language model behind it.

To show what I was selling, this is how the demo worked, with made-up data: the patient writes "I can't make Thursday"; the assistant offers the two nearest free slots; if they pick one, it moves the appointment and sends the front desk a message; the day before, the reminder goes out. If the question goes beyond opening hours, prices or what to bring, it replies that the clinic will call and logs the query.

What it included: installation, configuration and one month of adjustments. What it didn't include, and here's the trap I fell into: maintenance from the second month on. I didn't put it in writing because the video said that "gets negotiated later".

As for price, the video gave a very specific monthly fee. I won't repeat it here because I haven't checked it and I don't want anyone taking it as a reference. I set an installation fee and a monthly fee quite a bit lower than the video's.

Before sending it, I gave it to Claude to review through a client's eyes. It pointed out three things: that it didn't say what happened if something failed, that it didn't make clear what was done with patient data, which is health data, and that "without your front desk having to do anything" was a hard promise to keep. It was right on all three. I fixed the last one. The other two, following the script, I left "for the meeting".

This is where the videos speed up and reality slows down.

To sell automations, the script calls for cold messages by the bucketload. And here's something the videos skate over in one sentence: in Spain, the law on information society services generally requires prior consent to send advertising by email, businesses included. I looked into it late, with the campaign already under way. What I did do from the start was write only to contact addresses published by the clinic itself, one message per clinic with no follow-up chains, saying who I was and offering not to write again. And since it's a healthcare sector, the demo ran on made-up patients and appointments: I didn't ask for or touch any real data. If I did it again, I'd start with the phone or LinkedIn and get advice first.

Here are the numbers:

Of those hours, about 20 went on the research and building the demo, about 25 on finding contacts and writing messages, and the rest on meetings and proposals.

Of the nine replies, five were a polite "no, thanks", two asked for the price and vanished once they got it, and two led to a meeting, plus a third that came from a late reply.

The most revealing replies, paraphrased and without anyone's details:

• "We already have reminders, they come with the appointment software".
• "And if the assistant tells a patient something wrong, who's responsible?".
• "I'll think about it", which in small-business language means "no".

And then there's the silence. Seventy-odd messages with no reply. And the "I'll let you know"s, which are like silence but with good manners. There's something very educational about sending an email you've polished three times and getting nothing back. It brings you down to earth faster than any course.

The three meetings taught me more than all the niche research.

What a real client asks for isn't an automation. They ask you to explain what changes in their day-to-day, what happens if it fails, who fixes it and what's done with their data. One of the physiotherapists put it very plainly: "If one Saturday the system changes a patient's appointment and he shows up and I'm not there, the problem is mine, not yours". And she was right. I was going to sell technology; she wanted peace of mind.

Then there's how fragile an automated chain is. An AI automation agency connects tools so that certain tasks happen on their own, and those connections are delicate. All it takes is for an app, a password or an email format to change and the chain stops without warning. It happened to me during the demo itself: the messaging system had changed a permission and the workflow that worked on Friday sent nothing on Monday. Bad time to find out.

That maintenance, moreover, is work you don't get paid for unless it's in writing from the start. And I hadn't put it in writing.

Finally, the competition. When I looked for other agencies offering the same thing to clinics, I found dozens, almost all with the same website structure, very similar phrases and ever lower prices. Hundreds of people watched the same video and chose similar niches. Competing with all of them on price alone is a race to the bottom.

Changing shoes. Now I'm speaking as someone who applies AI in real client systems.

None of my clients came from a cold message. They came through previous work, acquaintances and recommendations from other clients. And the conversation with them never starts with the tool. It starts with the problem and the lost hours: how much time goes into that task, how many errors come up, what happens when it fails.

Starting with something small and measurable is far more convincing than promising a transformation. One case, anonymised: a small renovation company was losing several hours every week copying data from accepted quotes into its tracking sheet. I started with just that. When they saw it worked and that I answered the phone if it failed, they asked me for the next thing.

And something that has changed in a short time: part of what used to be a complicated workflow is now handled by an agent, an AI that takes several steps on its own from a task given in plain language. Which, by the way, makes selling "workflows" as a product even more fragile. If tomorrow the client can ask an assistant for the same thing in two sentences, the only thing that still has value is having understood their problem and standing behind the result.

If after all this you still feel like it, this is how I'd do it:

1. Know an industry from the inside before the tool. Better if you've worked in it.
2. Start with a measurable problem: hours, errors, customers left without an answer.
3. Put in writing what the price includes and, above all, the maintenance.
4. Don't compete on price alone.
5. Be clear about what you'll do when something breaks at ten o'clock on a Friday night.

None of these steps is technical, and that's no accident. The technical part is what the videos teach the most and what least decides whether someone buys from you. What decides it is whether the client trusts you'll be there when something goes wrong, and that trust isn't won with a nice demo but with time, with clarity about what you charge and with delivering what you promise, even if it's little.

What's true: you can sell AI to other companies, and there's demand for solved problems.

What's smoke: the niche copied from a video and clients within weeks. The challenge cost me about 55 hours and about €60 in tools and the domain for the trade name, with zero clients.

And a limit worth keeping in mind: this is one niche, one channel (cold messaging), six weeks and one person who couldn't use his contacts. With another industry or more time the result might be different. What I will stand by is that the script, as it's told, didn't hold up in my case.

My recommendation: read this before spending anything on a course about starting agencies.

This was the first attempt, and I'm telling it as the first chapter of a series about selling automations to small businesses. After finishing the Mostrador Digital challenge and putting that business on hold, in August 2026, I tried again, changing what failed here: the industry, the way of reaching clients and what I was selling. The next three chapters tell that second attempt: the new offer and how I looked for clients, the first job I delivered and the accounts for the whole series, with the usual rules: with numbers, however it turns out.