How I got the word out

Setting up the service was the easy part. Getting someone to pay me for it took five weeks, €170 in ads and flyers, and a lot of shoe leather. Here are the channels, how I used artificial intelligence in each one and what each euro brought in, which in some cases was exactly nothing.

In the second part I left Mostrador Digital, the service that builds FAQ assistants for local shops up and running, with the website, reviewed legal texts and an assistant installed for free at a friend's bakery. I had spent €149 and 40 hours, with zero income.

The goal is still the one from the first part: €1,000 net a month within six months of launch. For this phase I set aside €200 of the total budget of €500.

"Post every day with AI and customers will come on their own." "Ads that write, design and optimise themselves." Those are two of the classics, and both have the same trap: they assume the problem is producing content, when the problem is getting someone to want to see it and, on top of that, trust you enough to pay you.

My expectation, with all the honesty I can muster: five clients in the first month. I had the website, the images, a real case working and an assistant capable of writing a hundred posts in an afternoon. What could go wrong?

Spoiler: the hardest part of the business wasn't setting it up. It was this. In the first part I already said that what scared me most was selling, and the fear was well informed.

My customer is the owner of a neighbourhood shop, who doesn't spend the day online looking for chat assistants. With that in mind I chose six channels and ruled out two.

• In-person visits to shops in my area, with the demo on my phone. Target: 20 shops visited.
• Word of mouth from the pilot: asking my baker friend to talk about it. Target: none; whatever came.
• Flyers to leave at the shops I visited when the owner wasn't in. Target: 5 of those owners getting in touch.
• Ads on social media and Google, two small tests limited to my area. Target: one client for every €60.
• Instagram and LinkedIn, unpaid, sharing real cases. Target: three posts a week.
• Local SEO, which is showing up when someone searches for something in your area, starting with the business's free Google profile. Target: none in the short term; it takes months and I knew it.

I ruled out TikTok, because my customer isn't looking for suppliers there, and email, because I had nobody to write to without buying lists, which is a bad idea and legally dubious besides.

The workflow ended up like this, every Monday afternoon:

1. Ideas and calendar. I'd give Claude what had happened that week (curious questions to the bakery assistant, conversations with shopkeepers, a mistake) and ask for three posts.
2. Text for each network. The same message in an Instagram version, a LinkedIn version and, when needed, an ad or flyer version.
3. Images, based on real photos and screenshots, as I describe in My business's images, made with artificial intelligence.
4. Scheduling, with Meta's scheduling tool for Instagram and LinkedIn's own.

Social media and ads took me about 3 hours a week; without the machine I reckon about 7. That's in line with what I say about using artificial intelligence for a business's social media: it helps enormously with producing, but it doesn't decide what's worth telling. The rest of the week's hours, between 10 and 12, went on the street and on building assistants, where the machine helped a lot less.

An example from start to finish. Request: "A bakery customer asked the assistant whether they could come in with their dog. Write me a LinkedIn post about what a shop's customers really ask, with humour and without sounding like an ad, 100 words maximum". First answer: correct but with an "AI is revolutionising local retail!" and three rocket emojis. Second round asking it to drop the ad tone. Published: "The question a bakery's assistant got asked most on its first day wasn't the opening hours. It was whether the dog could come in. That's why I don't trust assistants that write themselves". It was the most commented post of the five weeks.

There were three things the assistant couldn't do for me.

The first, having something interesting to say. The first week I posted what it suggested, and posts like "5 ways AI can help your business" went by without a single comment. What worked always came from something real.

The second, trust. No shopkeeper hired me because of a post. The ones who said yes had seen my face, had tried the demo in front of me or had heard about it from someone they trusted.

The third, getting the tone right. The model leaned towards fairground enthusiasm and words like "transform" or "empower" that sound like a scam to a hardware shop owner. I redid roughly one in three posts.

And yes, a made-up claim slipped through, on paper: the flyer text said the assistant "cuts calls by up to 70%". That figure came from nowhere; the model put it there to sound good and I didn't check it. I spotted it with 200 flyers already printed. I crossed out the sentence in pen on the ones I handed out, which at least got conversations going.

Data from 22 June to 26 July, taken from the website, each channel's statistics, my notebook and my invoices.

By "contact" I mean someone who asked for information or a demo, not a "like". The three clients were a bicycle shop, a physiotherapy clinic and a hardware shop, and on top of that the pilot bakery started paying the fee. In total, €500 collected in the period.

Now the calculation that matters. The ads cost €120 and brought zero clients: €30 per contact and not a single sale. The flyers, €50 for one contact that didn't close. Money down the drain. Everything that worked was free in euros and very expensive in hours: pounding the pavement and having someone speak well of me.

And this ties in with what I say about whether you can make money with artificial intelligence: those €170 aren't the cost of getting the word out. You have to add about 68 hours of mine over these five weeks, plus the tools. That's what the videos never subtract.

Ads for half the province. I targeted the first campaign at "small business owners" within a 40-kilometre radius. I heard from a shop an hour's drive away and a gentleman who wanted a chatbot for his residents' association. €60 to learn that a neighbourhood service is advertised in the neighbourhood or not at all.

The demo with no signal. On the third visit, in a bar with the counter at the back, my phone had no signal and the demo wouldn't load. I ended up explaining with my hands how a chat works. I didn't close that one. Since then I carry a recorded video version just in case.

Not checking what I sign off on. The flyer's 70%. It didn't cost much money, but it could have cost credibility, which in a neighbourhood is all you have.

As of 26 July, eight weeks after starting: three paying clients plus the bakery, €500 collected and 19 contacts. Accumulated spend was €362, counting tools, server, model usage, ads and flyers. The hours, 108, well above the ten a week I promised myself.

In part 4 it's time to sit down with all the numbers: income, expenses and hours, week by week. And a test that scared me a little: stop pushing for two weeks and see whether the business held up on its own or ground to a halt as soon as I stopped turning the crank.