Back to Blog
    Automation·18 August 2026

    The chatbot works. The phone still rings just as much.

    You rolled out the AI and the repetitive work is still there. You didn't buy a bad tool — you bought a good tool for a problem that sat one layer below where you thought it was.

    Some articles are produced with AI-tool assistance and human review before publication. See our privacy policy for details.

    Folded hands on a glass table with tangled pipes and cables visible beneath it
    AI-generated image

    You set it up three months ago. It replies fast, never trips over opening hours, answers at 2 a.m. In the demo, everything lined up. And yet something is off: the repetitive work you wanted off your plate is still there. It changed shape, but it's still there.

    Before we go on, an uncomfortable question: did the chatbot solve your problem, or did it solve the problem that looked like yours?

    They aren't the same. And the gap between them is where much of what a small business spends on AI quietly evaporates — not on bad tools, but on good tools aimed at the wrong place.

    Start with the symptom, not the solution

    When you first went looking at AI, you probably arrived with a ready-made phrase: "I need a chatbot," "I want to automate bookings," "I need an agent for WhatsApp." Clean, closed phrases, each with a product name attached.

    And there's the tell: you arrived with the solution already chosen. Not with the problem.

    So answer this yourself, not me: who chose that solution? You, after looking at your operation top to bottom? Or the vendor, who happened to sell exactly that?

    A chatbot vendor sees chatbots everywhere. A CRM vendor sees CRM problems. Each diagnoses what they know how to cure. It's human, and it's predictable. But it means the tool you bought didn't come from your business — it came from the catalogue of whoever sold it to you.

    Now go one layer down

    Let's do an exercise. Think of the specific process you wanted to fix. The real one, the one that eats your afternoons. And answer three things:

    One. When that process gets stuck, does it stick for lack of information or lack of decision? Put another way: is the problem that nobody knows something in time, or that someone knows but nobody acts?

    Two. The piece of information you need for things to flow — the pending quote, the unconfirmed appointment, the client who said "let me think about it" two weeks ago — does it already exist somewhere in your company? Is it in your software, in a spreadsheet, in someone's head?

    Three. If that information already exists, your problem was never capturing it. It was using it. And that isn't solved by a tool chosen for its name — "chatbot," "agent," "automation" — because the name describes what the tool is, not what your operation needs.

    Notice what just happened? The problem moved. It's no longer where you placed it at the start.

    Three businesses, the same mirage

    Let me show you the same thing through three different doors. Recognise yourself in whichever one is yours.

    The professional firm asks for a chatbot "to stop losing hours answering client questions." Go one layer down: those questions repeat because the client never knows what stage their case is at. The problem wasn't answering faster. It was that the client had to ask at all. A chatbot that answers a question that shouldn't exist is a plaster over a leak.

    The online store wants to "automate support" because the team spends its days putting out after-sales fires. Go one layer down: almost every fire starts from the same upstream cause — a badly described size, a delivery date that overpromises. Automating the reply to the fire costs more, in the long run, than fixing what starts it. But to see that, someone has to look upstream.

    The service business with a booking calendar asks for "an agent to confirm appointments." Go one layer down: the gaps in the calendar aren't opened by unconfirmed bookings — they're opened by the sleeping revenue, the work a client already agreed to that nobody followed up on. That data already sits inside their software, in a list nobody reads for what it is: an unworked sales pipeline. Confirming appointments is fine. But it's the tip of the problem, not the problem.

    Three sectors. Three tools requested. And in all three, the real problem sat one layer below where the owner had placed it.

    So what was this really about?

    Go back to the chatbot that works and hasn't taken any work off your plate. Now you know why.

    You didn't buy a bad tool. You bought a good tool for a problem that wasn't quite yours. Nobody looked at your whole operation before putting a name on the solution — not even you, because you arrived with the name already chosen. And the real problem, the one that kept ringing the phone, was one layer down, waiting for someone to ask instead of sell.

    That's the one idea I wanted to leave you with, and I'd rather you reached it yourself: next time you feel you need "an AI for X," pause before the ready-made phrase. Don't ask which tool to buy. Ask which process breaks, where exactly, and which data you already have that you're not using. The tool is the last decision, not the first.

    Almost no one works in that order. Which is why there are so many chatbots that work sitting next to so many phones that still ring just as much.


    Before choosing a tool, look at the process. The AI Diagnosis reviews three processes in your business and shows where the real problem sits.

    Request a diagnosis (€99)

    If this way of looking at problems resonates, here's who I am and how I work.

    Meet Rafael Vidal
    #chatbot#automation#diagnosis#AI for SMEs#processes

    ¿Te encajaría algo así en tu operación?

    Empieza por el diagnóstico de 99 €. En 72 h sabes qué automatizar primero.