Almost every SME that thinks about using AI gets stuck on the same question: where do I start. There are so many things that "could be automated" that you end up with a list of twenty ideas and none under way. Or worse: you pick the flashiest one, the one that came up in a talk, and three months later you find out your data was not up to it.
This guide gives you a simple method for choosing a single first project. You list your repetitive tasks, score them on a few criteria, place them in an impact and effort matrix, and keep the one that gives you the most for the least work. You do not need to know about technology. You need to know your company and be honest with the numbers.
Why does it have to be a single project?
Because the first one is not just a project: it is how your company learns to work with AI. If it goes well, your team sees the change in their daily work and the second project sells itself. If you open five fronts at once, none gets finished and the conclusion in the office will be "this AI thing doesn't work".
You are not late. According to INE, in the first quarter of 2025, 21.1% of Spanish companies with 10 or more employees used AI, and 13.4% of those with fewer than 10 employees with an Internet connection. In the EU, Eurostat gives 17% among small companies (10 to 49 people) against 55% of large ones. And among European companies that had considered it and did not take the step, the most cited reason was lack of knowledge (70.9%). Choosing the first case well is the cheapest way to gain that knowledge.
There is a second reason, a more practical one. The business AI voucher announced by the IA360 Plan foresees that its rollout will include "prior maturity diagnosis, identification of a specific use case and subsequent measurement of the impact on productivity". One use case, specific and measurable. It is explained in the guide to the Artificial Intelligence Voucher (Bono de inteligencia artificial).
Step 1: what repetitive tasks does your company have?
Start with an inventory. Sit down with the people who do the work, not just whoever manages it, and note everything that repeats. A spreadsheet is enough. For each task, write down:
- What it is, in one sentence: "entering supplier delivery notes into the management software".
- Who does it and at what time of day or week.
- How many times it repeats per week or month.
- How long it takes each time, roughly.
- What data it uses and where it is: email, PDF, paper, WhatsApp, management software.
- What happens when it goes wrong: an annoyed customer, a fine, a duplicated invoice.
A few hints so nothing slips through. Look for tasks where someone copies data from one place to another, answers the same question many times, reviews documents always looking for the same thing or prepares a report that looks like last week's. That is usually where the work AI does well sits. If you do not know where to look, the four places where something almost always turns up (the repetitive, what is lost by not replying, what gets redone, and collections) are covered in what to automate first in your SME.
If you do not know how long something takes, do not make it up. Time it for a week or two. A measured figure is worth more than ten assumptions, and it will also help you later to know whether the project worked. If you want a broader view of where your company stands before starting, take the AI self-assessment.
Step 2: what criteria do I use to compare tasks?
Six. The first three measure impact: what you gain if the task is solved. The last three measure effort: what it costs to get there.
| Criterion | What it asks | Counts towards |
|---|---|---|
| Frequency and volume | How many times a week does it happen? | Impact |
| Time it takes | How many hours go on it? | Impact |
| Errors | Do people often get it wrong? How much does each error cost? | Impact |
| Data available | Is it digital, organised and in an accessible place? | Effort |
| Risk if it fails | What happens if the AI gets it wrong once? | Effort |
| Who uses it | Does the team want it? Does it change their way of working a lot? | Effort |
Two clarifications. Data available scores the other way round: if the data is already digital and clean, the effort is low. If it is on paper or spread across five programs, the effort is high. There is more detail in how to prepare your company's data for AI.
Risk if it fails does not mean the task is unimportant. It means how much human control it needs. Sorting emails carries little risk: if one goes to the wrong folder, someone moves it. Sending a quote to a customer without review carries much more. A high-risk task can be a good project, but it needs human review, and that adds effort.
Step 3: how does the impact and effort matrix work?
You place each task on a table with two axes. High or low impact, high or low effort. That gives four quadrants:
| Low effort | High effort | |
|---|---|---|
| High impact | Quick wins. Your first project comes from here. | Big projects. Worth it, but later, with experience. |
| Low impact | Fillers. Can be done with simple tools if there is time to spare. | Avoid. A lot of work for little return. |
- Quick wins: frequent tasks that eat hours, with data that is already digital and little risk. Your first project is here.
- Big projects: tasks with a big return that need several programs integrated or data cleaned first. Note them down for the second or third project. You will arrive with a team that already knows what working with AI is like.
- Fillers: small things that do not change your bottom line. Do not discard them, but do not turn them into "the company's AI project".
- Avoid: tasks that rarely happen, that depend on data you do not have or that require checking every result by hand. The effort never pays off.
Step 4: how do I score each task?
Give each criterion a mark from 1 to 5. Add the three impact marks and the three effort marks separately. For effort, 5 is the most costly: data on paper, high risk, a reluctant team.
An example. Imagine a distribution company with 15 people. It is a made-up case to teach the method, and the marks are what its team might give, not measured results:
| Task | Frequency | Time | Errors | Impact | Data | Risk | Use | Effort | Quadrant |
|---|---|---|---|---|---|---|---|---|---|
| Entering supplier invoices into the software | 5 | 4 | 4 | 13 | 2 | 2 | 1 | 5 | Quick win |
| Answering "where is my order?" on WhatsApp | 5 | 3 | 2 | 10 | 2 | 2 | 2 | 6 | Quick win |
| Seasonal purchasing forecast | 2 | 4 | 4 | 10 | 4 | 4 | 3 | 11 | Big project |
| Writing social media posts | 3 | 2 | 1 | 6 | 1 | 1 | 2 | 4 | Filler |
| Sorting incoming email | 4 | 2 | 2 | 8 | 1 | 1 | 1 | 3 | Filler |
| Profitability analysis per customer | 1 | 3 | 2 | 6 | 5 | 3 | 4 | 12 | Avoid |
With these marks, the first project would be invoice entry: high impact, low effort, data that already arrives as PDFs by email. WhatsApp support comes second. The purchasing forecast is left for when the company has its history in order.
Put the cut-off where it makes sense for you. With three criteria from 1 to 5, each axis runs from 3 to 15, and 9 is a reasonable midpoint. What matters is not the exact figure but comparing tasks with each other using the same criteria and the same people scoring.
Before deciding, make one last check: can you measure the result? If you choose invoices, note down today how many arrive per month and how long each one takes. Without that starting point you will not know whether it helped. To turn those hours into euros, use the guide to the profitability of an AI project. And once you have chosen, what comes next is the 5 phases of putting AI in place: understand the process, build on your tools, see it working, measure and adjust.
What signs point to a bad first project?
If your candidate meets any of these, think twice:
- Nobody knows how long it takes today. With no starting point, there will be no way to show the improvement.
- It happens a few times a year. Automating something done four times does not repay the effort of building it.
- The data is on paper or in one person's head. It first has to be got out of there, and that is another project.
- A mistake has serious consequences. Diagnoses, legal decisions, money going out without review. It can be done, but not as a first experience.
- The team has not asked for it and does not want it. If whoever does the task does not see the benefit, they will keep doing it by hand "just in case".
- It is just buying a licence. Paying for a tool without changing the process is not a project. The AI voucher, in addition, will not fund mere licence subscriptions.
- It is described in big words. "Transform customer service" is not a use case. "Answer order status enquiries out of hours" is.
You will find more common mistakes in mistakes when putting AI into an SME.
What first project makes sense in each Campo de Gibraltar sector?
These examples are illustrative. They are not clients or results: they are typical candidates worth running through the matrix. In your company others may come out.
Logistics and customs at the Port of Algeciras. Freight forwarders and customs agencies move a lot of paper with a repeated structure: commercial invoices, packing lists, bills of lading. Extracting that data to prepare the documentation usually scores high on frequency and errors. Mind the risk: a wrong figure in a declaration has consequences, so human review is mandatory. More in AI for logistics and customs at the Port of Algeciras.
Hospitality in Tarifa. The season concentrates the enquiries: bookings, availability, opening hours, allergens, how to get there. Answering those repeated questions on WhatsApp or the web, with access to real availability, usually falls under quick wins. Seasonal staff forecasting is a big project for later. You will find ideas in AI for seasonal businesses in Tarifa.
Workshops. Calls to book an appointment and ask "is my car ready yet?" interrupt the mechanic. An agent that books and reports the status usually scores well. Setting up automatic quotes from photos has more risk and more effort. See AI for workshops in Algeciras and La Línea.
Law firms and accountancy practices. Classifying and extracting data from the documentation clients send (invoices, payslips, receipts) is a clear candidate: high volume, repeated format. Drafting legal documents or giving legal answers without review is a bad first project because of the risk. Read AI for lawyers and accountancy firms.
Retail in La Línea. Enquiries about stock, opening hours and orders by WhatsApp, often in two languages. Handling them with a chatbot connected to your inventory is usually a quick win if the stock is already in a program. If the inventory is in a notebook, it has to be digitised first. More in AI and WhatsApp for shops in Algeciras and La Línea; and if you import through the crossing, the paperwork the open border brings is in invoices and paperwork in La Línea.
Clinics. Appointments, rescheduling, reminders and pre-consultation questions. High volume and data that is already in the diary. Anything touching diagnosis or medical records stays out of the first project: high risk and health data, which data protection law treats as a special category. You will find it in AI for dental clinics.
Notice the pattern. In almost every sector, the quick wins are of two kinds: reading repeated documents (invoices, delivery notes, client documentation) and answering repeated enquiries with real data from your business. That is why invoice automation and AI agents are usually the first step.
How does this fit with the AI voucher?
It fits almost word for word. The IA360 Plan describes a rollout in three pieces: a maturity diagnosis, a specific use case and measurement of the impact on productivity. The method in this guide gives you all three:
- The inventory is the basis of the diagnosis: what processes you have, what data they use and where the time goes.
- The matrix leaves you with a specific, justified use case, not a general idea.
- The starting measurement (how many times, how long, how many errors) is what will let you show the impact afterwards.
What is not yet known: the amount per company, who will be able to apply and how it will be requested. As of 25 September 2026 there are no published rules. The plan foresees a pilot in the first half of 2027 and the general call before the end of 2027. In the meantime, the preparatory work does not depend on any call. If you are looking for other lines, go through the grants for putting AI in place in 2026.
What if I am not sure how to score the effort?
That is normal. You know the impact better than anyone: you know which tasks eat your week. The technical effort is harder to see from the inside. Whether your data works as it is, whether your management software lets you connect to it, which part needs human review.
That is exactly what we do in the business audit: we go through your processes with you, score the tasks with these same criteria and leave you in writing with what your first project is and why. If you prefer to talk it over first, write to us.
Sources
Xora IntelligenceAI consultancy and custom software in Los Barrios (Cádiz). We write about what we build for businesses in the province. If you have a question about your case, write to us.




