Back to Blog
    AI for SMEs·22 September 2026

    Before you automate: how to know what your business actually needs

    Why small businesses should analyse their goals, processes and systems before implementing AI, and how a case-by-case diagnosis avoids investing in what isn't needed.

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

    Man in a cluttered office moving floating digital panels with his hands
    AI-generated image

    When a small business starts thinking about artificial intelligence, the natural instinct is to think about the tool. A chatbot for the website. An agent that answers WhatsApp. Something that prepares invoices on its own. It makes sense: tools are visible, you can try them out, and they promise quick results.

    The problem is that the tool is the last piece, not the first. Before it come three questions that are rarely asked out loud: what the business wants to achieve, how the work actually gets done today, and which systems are already in place. Without those answers, any investment in AI is a gamble.

    Why starting with the tool is expensive

    There are at least three ways to get it wrong, and none of them is a technical problem.

    The first is automating a process that doesn't work well. If quotes get lost because nobody is clear on who follows them up, a system that sends them out faster fixes nothing. They go out sooner and still get lost.

    The second is fixing the symptom instead of the cause. A team that can't keep up with messages might need a customer service agent. Or the real problem might be that the same question comes in through four different channels and nobody brings them together.

    The third is buying something new when what you need is already paid for. It's the easiest one to overlook, and it deserves an example.

    What you already have installed (and may not be using)

    Recently I was going through the screens of a practice management system for dental clinics, the classic desktop kind.

    The quotes list could be filtered not only by date and amount, but also by reason for rejection, follow-up actions and their outcome, communication channels and financing status. And the clinical management section had dedicated modules for patient recalls, pending items and appointments.

    In other words, much of what a clinic needs to follow up on its treatment quotes already exists inside its own software. The useful question isn't “which AI should we add?” but some less comfortable ones: is that data actually being filled in? Does anyone review that list every week? What happens to a rejected quote?

    Sometimes the answer is yes, it's being used, and what's missing is a layer on top: an automatic alert, a weekly summary, a follow-up message after fifteen days. Other times nobody touches it, and then the first step isn't technological, it's organisational. Both conclusions are valid. What makes no sense is deciding without looking.

    The three layers of a useful diagnosis

    1. Goals: what you want to achieve

    Everything starts here, because this is what defines what “better” means. Wanting to increase revenue is not the same as getting your team to stop working late, depending less on one key person or preparing to grow without hiring. Each goal puts different processes first.

    2. Processes: how work really gets done

    Not how it's written down, but how it happens on any given Tuesday. Here you look for specific signals:

    • Tasks repeated the same way many times a month.
    • Data copied by hand from one place to another.
    • Waiting times: things that stall until someone remembers.
    • Errors that always show up at the same point.
    • Knowledge that only one person has.

    3. Systems: which tools you use

    Which software you have, what you pay for, how much of it you actually use and how it all connects, or whether it connects at all. Sometimes the improvement lies in getting two existing tools to talk to each other, not in adding a third.

    Each case, on its own

    Two companies in the same sector and of the same size can need very different things. One might have its bottleneck in collecting documents from clients; the other, in handling phone calls. A generic recommendation serves neither.

    That's why each candidate process is assessed on its own, using four simple criteria:

    CriterionThe question
    ImpactHow much time, money or error does it save each time it happens?
    FrequencyHow many times a month does it happen?
    DifficultyWhat does it take to implement with what's already in place?
    RiskWhat happens if it fails? Does it affect clients, sensitive data or legal obligations?

    Impact and frequency tell you what it's worth; difficulty and risk tell you what it costs. What's worth a lot and costs little comes first. What's worth little and costs a lot gets parked, guilt-free. And sometimes the most valuable part of a diagnosis is simply knowing what isn't worth doing yet.

    What a good diagnosis should give you, whoever does it

    • A clear picture of where things stand, in business language rather than technical jargon.
    • Specific ideas applied to your case, not a generic catalogue.
    • Estimates with the assumptions in plain sight, even if they are only indicative.
    • A clear next step, and the freedom not to take it.

    If all you walk away with is the name of a product, ask to see the reasoning behind it.

    Five questions to start with today

    1. If you could take one task off your team's plate tomorrow, which would it be?
    2. Which process grinds to a halt if one particular person gets ill?
    3. What information gets typed in by hand in more than one place in your business?
    4. Which features of your software do you pay for but never use?
    5. Where are you losing clients or opportunities without anyone noticing?

    If any of these made you slightly uncomfortable, that's probably where the first process to review lies.

    How we do it at Isenzo-IA

    The person who carries out the diagnosis is the same person who implements it, if you decide to go ahead. First I get to know your business; the technology comes later.

    Express Diagnosis · €99

    A 60-minute session, 3 to 5 concrete ideas, a short one- or two-page report with indicative estimates, and a recommended next step. Designed for micro-businesses, or for anyone who wants to find out, without a big commitment, whether this makes sense for them. It doesn't commit you to anything else.

    Full Diagnosis · €250 to €500

    A 90-minute session plus a short follow-up session, 5 to 7 ideas analysed in depth, a report of 3 to 5 pages, an implementation roadmap, estimates of cost, timeline and return, and a clear prioritisation by difficulty and impact. Designed for SMEs with 5 to 50 people or more complex structures.

    Frequently asked questions

    Do I need to prepare anything beforehand?

    No documents are needed. It helps to know which software your business uses and to have in mind the tasks that take up most of your team's time.

    What if what I need isn't AI?

    Some of the improvements may turn out to be organisational rather than technological, or the most sensible move may be to wait. I'll tell you either way. Finding out for €99 is far cheaper than discovering it after an implementation.

    Will I have to change my software?

    Not necessarily. The first step is to make the most of what you already have and connect it. Switching systems only makes sense when the current one is the bottleneck.

    What's the difference between the Express and the Full Diagnosis?

    Depth. The Express tells you where to start. The Full Diagnosis gives you the whole plan, prioritised and with numbers.

    If you'd like to know where to start in your case, you can also email me at rafael.vidal@isenzo-ia.com.

    Book your diagnosis
    #AI diagnosis#SMEs#processes#automation#operational strategy

    ¿Te encajaría algo así en tu operación?

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