Was the business worth it?

We've reached the end of the series, and the question I asked myself at the start no longer allows for gut feelings: it's time to answer with numbers. Here's the verdict on Mostrador Digital, my service of FAQ assistants for local shops, what's true in the promise, what's smoke, and what I'd change if I started tomorrow.

A quick recap. In the first part, Can I build a business that pays me a salary?, I went over what the internet promises and chose a small, very specific service: setting up FAQ assistants for local shops, with a fixed price and a monthly maintenance fee. In Getting the business off the ground I told you what it took to set it up; in How I got the word out, how I got the first clients. And in The business in numbers I put every figure into a table.

The rule was to tell it however it turned out. I'm keeping it to the end.

The bar was set by one of those hugely popular videos promising "your AI agency billing more than your job in 60 days, with no experience and no investment". I won't link it: there are dozens just like it.

From the table in part four I'll keep four figures, those for ten weeks, from 1 June to 9 August 2026:

The expenses include what almost nobody mentions in the videos: the server, model usage via API, the accountant, an image tool I didn't cancel in time, a template I never used and €170 in ads and flyers that didn't bring in a single client.

And that gives the figure that hurts most: €230 over 118 hours works out to €1.95 per hour of work. Any part-time job would have paid me quite a bit more, without walking into a hardware shop to explain why they need a chat.

It's not negative. It's something. But it looks nothing like the video.

In the first part I defined what "a salary" meant to me: €1,000 net per month within six months of launch. I haven't used up the deadline, but the gap is what it is. What counts towards a salary isn't the setups, which are paid once, but what comes in every month: four fees of €25, €100 a month, which come down to about €70 net once I subtract the server and model usage. I'm at 7% of the goal.

I also wrote down what would count as failure: not reaching half the figure within the deadline, or reaching it at the cost of more hours than I have. The first isn't entirely lost; the second has already happened: I promised myself ten hours a week and I did almost 12 on average, with weeks of 14 and 15.

The acid test was the one I described in part four: two weeks, from 27 July to 9 August, with no visits, no posts and no ads. In came €140, fees plus an extension for the bike shop, which was already a client. Zero new contacts. The business doesn't run on its own; if I stop pushing, it stops.

And the curve? It's going up, but slowly and on foot. Of the three paying clients I signed, not counting the pilot bakery, two came from in-person visits and one from the baker's recommendation; the ads, the flyers and social media brought in none. At about €17.50 net per month per client, it would take nearly 57 shops to reach the 1,000, and each one cost me walking through the door.

AI saved me about 35 hours of writing and coding: copy, the chat code, legal drafts, posts and the first build of each knowledge base. That's real.

The part of the video that turned out to be true was starting almost without investment. €410 in total won't ruin anyone. The real investment is the hours, and the "no experience" bit is false: without knowing how to set up systems, I couldn't have offered this.

AI makes faster a job that someone already wants to pay for. It doesn't bring clients. It doesn't replace having something to offer or the trust that leads someone to pay you. My clients chose me because they saw my face, tried the demo in front of me or heard about it from someone they trust. No shopkeeper hired me because of a post.

And there's a distinction worth keeping in mind when people talk about making money with AI: using it to make money, which is what I tried, is one thing; selling it through courses or agencies, which is where most of the promises are, is another.

As for passive income: an assistant answers on its own in the middle of the night, but someone has to update its opening hours and make sure it doesn't make anything up. Even on standby, it asked for 10 hours in two weeks.

Income without subtracting the costs isn't income, it's a headline. And headlines don't pay the rent.

The split that taught me the most:

In hours, of the 118, about 28 were work where the machine did the bulk and I reviewed; the other 90 were pure work of mine: deciding, selling, serving, setting up and fixing. And of the 35 hours it saved me, I spent about 15 reviewing what it got wrong.

The most expensive failure wasn't the one with the most euros: it was the model's tendency to fill in gaps. It started on 11 June, when the bakery's assistant calmly replied that they were open on Sunday from nine to two. They aren't. It cost me 3 hours and forced me to create the set of 40 trick questions I run on every assistant before installing it and every time the shop changes something, about 45 minutes per assistant per review that don't show up on any invoice. The same habit crept into the flyer: that 70% fewer calls, made up and printed in 200 copies. A text that sounds good isn't a true text.

That failure isn't in the revenue table but in the hours, where nobody looks. The machine speeds up the part I already knew how to do and leaves the part I didn't untouched: selling, dealing with clients and responding when something goes wrong.

If I started today I'd change four things, and they all come from the data:

• The price. €25 a month as a launch price is too little for what it costs to keep an assistant up to date. I'd charge more for the fee from the first client, even if I closed fewer.
• Where I put the money and the hours. The ads cost €120 and brought zero clients, and the flyers, €50 for one contact that didn't close. What worked was free. I would have started by asking the baker to introduce me to the neighbours on his street.
• What I left to the machine too early. The assistants' answers and everything I was going to print or sign. The "if you don't know, say so and ask them to call" rule goes in from the first minute, not after the scare.
• What I thought about too late. Spending limits per assistant, which I set after losing €10 in one night, and the privacy of the conversations, which I only sorted out because the accountant asked. The machine warned me about neither.

Of the three paths I laid out in part four (keep pushing, pause or shut down), I'm going with the pause. I'm not shutting it down: the four shops, the three paying ones and the pilot bakery, are happy and I owe them the maintenance they pay for. I won't look for new clients or spend any more on advertising, and I'm cancelling the image tool.

If anything changes in the coming months, especially the price, I'll share it in an update to this series. What stays: what I learnt, four working assistants and the set of trick questions.

If a video promises you a monthly figure with AI, ask yourself three questions. What does it cost, counting everything. How many hours are behind it. And who buys, with first and last names, not "the market". In my case they had names: a bakery, a bike shop, a physiotherapy clinic and a hardware shop.

We don't sell data, we sell knowledge and experience. And this time experience says you can earn something, but not a salary, and not in weeks.

Before this business, in the winter, I'd already tried another promise: AI-made books on Amazon. I tell it in its own series, with the same rules: for real and however it turns out.