According to INE, Statistics Portugal, in 2025 only 11.5% of Portuguese companies with 10 or more employees used artificial intelligence, and among companies with between 10 and 49 workers that figure fell to 9.4%. In the European Union, Eurostat puts it at around 20%. At the same time, Gartner predicts that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. In other words: the technology is accelerating, and most Portuguese SMEs have not even started. That is a risk, but also a huge opportunity for those who act with method.
AI agent or chatbot: what is the difference?
I often see business owners confusing the two. A chatbot answers questions: the user types, it returns some text. An AI agent goes further. It receives a goal, decides the steps, uses tools (email, the CRM, the spreadsheet, the invoicing software) and carries out tasks with a degree of autonomy, coming back to you when it needs a decision.
As an engineer, I like a simple definition: the chatbot is a good adviser; the agent is an assistant that does work. And precisely because it does work, it demands more care.
What an agent can already do in an SME
We do not need futuristic scenarios. Today, with tools available on the market, a well configured agent can:
- Triage emails: read the shared inbox, classify requests (quote, complaint, supplier), suggest replies and route them to the right person.
- Follow up leads: spot contacts left unanswered for more than three days and prepare a personalised follow up message for the salesperson to approve.
- Prepare quotes: gather the details of the request, check the price list and put together a first draft of the proposal.
- Reconcile documents: match supplier invoices against purchase orders and delivery notes, flagging any differences.
Notice that almost all of this touches directly on the 5 Ways we work on at ActionCOACH: more leads handled, a better conversion rate because nobody is forgotten, more transactions because the team gains time to sell, and faster proposals that protect the average sale value and the margin.
Microsoft, in its 2025 Work Trend Index, reports that 80% of workers globally say they lack the time or energy to do their job, and that they are interrupted on average every two minutes. In an SME, where everyone wears several hats, that recovered time is gold.
The risks you cannot ignore
In my experience, enthusiasm without method is the fastest route to disappointment. Gartner itself predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, due to rising costs, unclear business value or inadequate risk controls. It also estimates that, of the thousands of vendors claiming to sell agents, only about 130 actually do. The main risks are three:
- Errors: an agent can misread a request, make up a figure or apply the wrong price. Without review, the error reaches the customer.
- Too much autonomy: granting permission to send emails or change records without approval is like handing the company keys to an intern on their first day.
- Data: you need to know where customer data is stored and comply with GDPR. According to Eurostat, among European enterprises that considered using AI but did not, 70.9% cited a lack of expertise and 48.8% cited data protection and privacy concerns.
How to start with method: a 90 day pilot
McKinsey's 2025 State of AI study shows that 62% of organisations are at least experimenting with agents, but only 23% are scaling them in any part of the business. The gap between experimenting and getting results almost always comes down to method. I suggest four rules:
- Start small: choose a single repetitive process, with volume and clear rules. No "transforming the whole company".
- Human in the loop: in the first weeks, the agent proposes and a person approves. Autonomy is earned through a track record, not handed out.
- Metrics before and after: measure the starting point (time per quote, unanswered leads, hours spent on reconciliation) and track it every week.
- Decision at 90 days: at the end of the pilot, decide with numbers whether to scale, adjust or switch it off.
A hypothetical example: an air conditioning company with 18 people receives around 60 quote requests a month and takes, on average, five days to reply. For 90 days, an agent prepares the draft of each proposal and the sales manager reviews and sends it. The indicator to track is simple: average response time and quote conversion rate. If both improve and errors are negligible, there is a basis to expand.
This week's action
Set aside one hour with your team and list the five most repetitive tasks in the business. For each one, note how many hours a week it takes and whether it follows clear rules. Choose just one, the one that steals the most time from sales or customer service, and define the indicator you will measure. That will be your pilot.
If you would like to do this exercise with support, the ActionCOACH Porto team can help. The Free Business Analysis is a good starting point to see where the greatest potential lies, and the 90 Day Planning Workshop helps turn the idea into a concrete plan, with goals and owners.