Artificial intelligence for SMEs: real cases and where to start

I'm Elmar Mamedova, a full stack developer in Málaga, and I have spent a couple of years building projects with artificial intelligence for small businesses. In that time I have seen it all: SMEs that have saved hours every week with a well-tuned assistant, and others that spent an entire budget on a "magic chatbot" that nobody used. The difference almost never lies in the technology, but in choosing the right first problem to solve.

This article is the guide I wish I had had when I started. No empty promises and no jargon: real AI cases that work in an SME, where to begin when the budget is tight, the myths worth debunking and how to choose your first use case without shooting yourself in the foot. If you want to see the full approach of my services, you will find it on the artificial intelligence page.

When I talk about artificial intelligence with a client, the first thing I do is bring expectations down to earth. The AI that really moves the needle in an SME is not a robot that runs your company: it is specific tools that understand language, classify information and generate text. Language models like the ones behind ChatGPT, connected to your business data, do 90% of the useful work you will see in this article.

The important thing is to understand that AI is a complement, not a substitute for human judgment. It works very well for repetitive tasks, for producing a first draft or for filtering large amounts of information. But someone on your team has to review, decide and add the final touch. If someone sells you an AI "that works on its own and never makes a mistake", be suspicious.

The star use case, and rightly so. An SME receives the same questions over and over again: opening hours, prices, availability, how to return a product, where the order is. A modern chatbot connected to your information answers those questions at any hour without your team having to be glued to their phone.

The key lies in the difference between a chatbot from five years ago (rigid button trees that frustrated everyone) and a current one based on natural language. Today's chatbot understands the question even if it is poorly written, replies with natural sentences and, when it does not know something, hands off to a person instead of making up an answer. Set up well, it resolves between 60% and 80% of first-level queries.

• 24/7 support without hiring night or weekend shifts.
• Consistent answers: it always says the same thing, regardless of who is on duty.
• Pre-filtering: the customer with a complex question reaches your team already "warmed up" and with context.
• Lead capture: many chatbots collect the email or phone number before handing off, so you do not lose the lead.

My advice: start with the chatbot only on your website or WhatsApp, fed with your 20 or 30 most frequent questions. Measure it for a month and expand. Do not try to make it answer everything from day one.

This is a step beyond the public chatbot, and it is where I see the most real time savings. Imagine an internal assistant that knows all your manuals, price lists, protocols and old emails, and that any employee can ask "what is the warranty procedure for product X?" and get the exact answer with the source. This technique is called RAG (retrieval-augmented generation): the model makes nothing up, it searches YOUR documents and answers by citing them.

I have built it for consultancies whose knowledge was spread across a thousand PDFs and inside a single person's head. The day that person went on holiday, the team got stuck. With an assistant on top of the documentation, that knowledge stops being a bottleneck. New team members get up to speed faster and nobody wastes half a morning looking for a figure in a shared folder.

The requirement is to have the documentation reasonably organised. If that part is missing, it is usually the first job to do, and it is often where it pays most to lean on automation to gather and structure the information before connecting the AI on top.

If your SME receives forms, emails or messages with requests, AI can read and classify them automatically: distinguish a customer ready to buy from one who is just browsing, detect the urgency of an issue or tag by type of service. That lets your team spend time on the contacts that really matter, instead of going through the inbox one by one.

In practice, this lives inside your Customer Relationship Management, the CRM where you store your customers. The AI reads each new lead, gives it a score and a reason, and leaves it ready for sales to act on. It is one of the fastest returns out there, because it directly targets sales.

1. 1A form or an email arrives with a query.
2. 2The AI extracts the key data: what they want, approximate budget, urgency.
3. 3It assigns a tag and a priority according to your criteria.
4. 4It records it in the CRM and alerts the right person if it is urgent.

Combined with workflows, this becomes very powerful. I explain it with concrete examples in my article on automating business tasks.

Two very different uses that I group together because both free up time from tasks that used to eat entire afternoons. In content generation, AI gives you the first draft of product descriptions, social media posts, replies to reviews or follow-up emails. It does not publish on its own, it saves you the blank page. You review, adjust the tone and keep control of your brand.

In data analysis, the change is dramatic for anyone who is not fluent in spreadsheets. You can ask it in plain language "which product sold the least last quarter?" or "summarise the most repeated complaints this month" and get a clear answer. It does not replace an analyst for complex decisions, but it puts the information within reach of anyone on the team without knowing formulas.

There is a quiet shift here that few SMEs in Málaga are taking advantage of. More and more people do not search on Google: they ask ChatGPT or Perplexity directly "which accounting firm do you recommend in Málaga?". Optimising so that those models cite you is called GEO (Generative Engine Optimization), and it is the first cousin of good old SEO.

The good news is that many of the foundations are the same: clear, well-structured content that answers real questions. Structured data with schema.org helps your business be readable by machines, and the guides in Google's search documentation remain a solid reference. The difference with GEO is that now you also write with the goal of an AI model being able to extract and cite your answer.

• Answer specific questions from your sector with direct, verifiable answers.
• Keep your Google listing and your data (NAP) consistent across the whole website.
• Use structured data so the content is easy to interpret.
• Build real authority: mentions, reviews and your own content that adds value.

I go deeper into how all this fits together at a local level in my guide on automation and AI in Málaga.

Let's be honest: AI has real risks and it is worth facing them head-on. The first is data privacy. If you feed customer information into an AI tool, you have to know where it is processed and whether it is used to train the model. For sensitive data, you have to choose providers that comply with the GDPR and, in many cases, not send identifiable data. This is not optional, it is a legal obligation.

The second risk is dependency. If you build your entire business on a single tool from a single provider, you are exposed to their changes in price or terms. That is why I design projects to be portable and not to tie you down for life. And the third risk, the most common, is "hallucinations": AI sometimes states false things with confidence. That is why I insist so much on human review and on systems like RAG, which force the model to rely on your documents.

There is another myth worth deactivating: that AI is only for large companies with big budgets. It is exactly the opposite. Today the tools are so accessible that an SME can test a use case for less than the cost of a small advertising campaign.

You do not need a big project to get started. My recommendation is to choose ONE single use case, the one that tackles a clear and measurable pain, build it small and measure results before expanding. Starting small is not settling: it is the fastest way to learn what works in your business without risking much money.

1. 1Identify the repetitive task that costs you the most time or money every week.
2. 2Check whether it is one AI does well: answering questions, classifying, drafting drafts or summarising data.
3. 3Build a minimal version and use it with a real case for three or four weeks.
4. 4Measure: hours saved, better-attended leads, questions resolved without human intervention.
5. 5If it works, expand. If not, you have learned cheaply and move on to the next case.

To choose that first case well, look for the intersection of three things: that it is repetitive, that it has a measurable impact and that your data to feed it is available. If all three coincide, you have your ideal candidate. You can see examples of projects I have built with this approach in my projects.

Artificial intelligence for SMEs is no longer science fiction nor a luxury for multinationals. It is a set of accessible tools that, well applied to a specific problem, give you back time and help you sell better. The secret is not to adopt "all of AI" at once, but to start with a clear use case, measure it honestly and grow only when the numbers back you up.

If you take away just one idea, let it be this: the technology is the easy part. The hard, and valuable, thing is choosing well what to solve first. And that is where a realistic approach makes all the difference between an expense and an investment.