8 min read
AI adoption for small teams: where to start, in order
Six steps. Most small teams do them in the opposite order.
A small team should start AI adoption with the work, not the tools. Spend a week finding where the hours actually go, then pick one workflow. Decide who checks the output before anything goes live. That order is the whole answer, and here is the reasoning behind it.
The sequence
Find the hours before you find the tools
Pick one workflow, not one department
Choose fewer tools than you think
Decide who checks the output
Write down what you are not going to use it for
Measure the net, not the saving
What actually stops small teams adopting AI?
Eurostat asked EU enterprises that had considered AI and decided against it why they stopped. Lack of relevant expertise came first, named by 70.3%. Cost was named by 38.4%.
The pattern holds at every company size: 70.9% of small enterprises named expertise, against 65.1% of large ones. Expertise was the leading barrier in this survey, but cost still mattered. The survey covers enterprises with at least ten people in selected sectors.
70.3%
named lack of relevant expertise as a reason for not adopting AI, compared with the 38.4% that named cost.
Eurostat, ICT enterprise survey, 2025.
These figures do not show that budget or access is irrelevant. They make a practical case for understanding the work before buying another tool.
Start by looking at your own work. Include subscription costs, training and review time when you decide whether a trial is worth running.
How many European small businesses use AI?
EU27 enterprise AI adoption reached 20.0% in the 2025 reference year, up from 13.5% the year before. That comes from Eurostat's ICT Enterprise Survey. Adoption is climbing fast. Four in five enterprises covered by the survey still do not use AI in any way it counts. Businesses with fewer than ten people are outside its scope.
In Portugal it is lower. INE's IUTICE 2025 survey put it at 11.5% of firms with ten or more workers. That is up from 8.6% in 2024.
Enterprises using AI
EU27 adoption grew by half in a single year, and Portugal is climbing at the same pace from a lower base. On a 0–100 scale the other half of the story stays visible: most enterprises covered by these surveys do not use AI in the ways measured.
The size gap is the striking part: 9.4% among firms with 10 to 49 people, against 49.1% among firms with 250 or more. The European Commission's 2026 Digital Decade report for Portugal names low business digitalisation as a weakness, and asks for more support for AI take-up by companies.
Adoption is mostly a company-size story
Same country, same year. A large Portuguese firm is more than five times as likely to use AI as a small one. That gap is what this article is about closing.
These figures provide context, not a ranking of your agency. The surveys cover many industries and do not show where a particular studio sits among its competitors.
1. Find the hours before you find the tools
Before evaluating anything, spend one week noting where time actually goes. Not a formal audit. A shared note, and everyone adds tasks that felt slower than they should have been.
You are looking for three properties: it repeats, it produces text or images or structured output, and nobody enjoys it.
That list is your shortlist. Choosing a tool first can leave you looking for a task to justify it. Start with a problem your team already needs to solve.
2. Pick one workflow, not one department
The instinct is to say "let's use AI in marketing". Too broad to act on and impossible to evaluate. Pick one workflow. Not marketing, but "the monthly client report write-up". Not admin, but "turning call notes into follow-up emails".
One workflow gives you a before and an after. It gives you a person who owns it. The first one often does not work. When that happens you learn something specific, instead of concluding that AI does not suit your business.
Set a first review after four weeks. Keep, adjust or stop the trial based on quality, total time and how often people use it. Extend it only if there is a clear question still worth testing.
3. Choose fewer tools than you think
BCG researchers surveyed 1,488 full-time US workers at large companies, and published the results in Harvard Business Review in March 2026. Productivity rose with each AI tool up to three, then dipped after that. That finding does not establish a universal tool limit for small teams. Our recommendation is to start with one general assistant and add another only for a defined need.
Tool sprawl looks like progress and functions as cost. Every additional tool is another subscription, another login, another set of habits. It is also another place where client data might sit, and another thing to explain when someone leaves.
If you take one thing from this article, take this one. Check which tools people actually use before renewing or adding subscriptions.
4. Decide who checks the output
This is the step that gets skipped, and it is the one that determines whether any of this actually saves time.
Workday's January 2026 survey covered 3,200 full-time employees who actively used AI at companies with at least US$100 million in annual revenue. Of those respondents, 85% reported saving one to seven hours a week. Nearly 40% of reported savings went into rework; that is not an agency-specific estimate. IDC research published by Sage in July 2026 found 48% of surveyed finance decision-makers spending at least 15 hours weekly checking AI output. Its sample covered companies with 20 to 1,999 employees. It named the effect the verification tax.
The verification tax
An approximate split of the time savings reported in Workday's sample, not a prediction for every team. It is the difference between a workflow that pays for itself and one that only feels faster.
Time saved is not the number that matters. Time saved minus time spent checking and fixing is. Measure the result in your own team.
So before a workflow goes live, answer one question in writing. Who reads this before it matters, and would they catch an error? If the answer is nobody, you have not automated a task. You have moved the risk somewhere you cannot see it.
5. Write down what you are not going to use it for
A short list. Half a page. It covers three things: what data never goes into a tool, what work is never AI-drafted, and what always gets human sign-off.
Two reasons this matters more than it sounds. The practical one is that people are already using these tools, whether or not you have a policy. A written boundary is the difference between informed use and quiet use.
The regulatory reason is that the EU AI Act places duties on organisations using AI. The AI literacy duty under Article 4 has been in force since February 2025. Transparency obligations under Article 50 apply from 2 August 2026. A short internal document can support responsible use, but it does not establish compliance. Your duties depend on your role and how you use AI.
Dates worth knowing
February 2025: the AI literacy duty under Article 4 came into force. Already applies to you.
2 August 2026: transparency obligations under Article 50 start to apply.
6. Measure the net, not the saving
Pick one number before you start. Hours on the task, turnaround time, or output volume at constant quality. Record it for two weeks beforehand. Then measure the same thing afterwards, including the checking time.
This sounds fussy for a small team. It takes about ten minutes a week. Without it you can carry a tool for a year because it feels faster.
What good looks like after three months
The three-month checklist
One workflow that has genuinely changed
A named person who owns it
One or two tools the whole team can name
A written rule about what stays human
A number showing whether it worked
What to do next
Start the shared note this week. One week of observation costs nothing and it determines whether everything after it is aimed at the right thing. When you choose the workflow, the five AI workflows that hold up in agencies make a good shortlist.
If you would rather not build the sequence from scratch, ViraOps has it laid out. The AI Foundations track walks a team through it. The Monthly Workflow Playbook adds a complete worked example each month.
ViraOps is €89 a month, up to five seats included, with a 14-day free trial.
Written by Catarina Mestre, Founder of Viravesso. Over a decade in creative operations across agencies and studios in the UK and Portugal, now writing practical AI guidance for small marketing and creative teams.
Common questions
The cost of starting with AI includes subscriptions, training and checking the output. Eurostat found cost mattered to 38.4% of covered enterprises that considered AI but did not adopt it. Start with one defined workflow and budget for the whole trial, rather than assuming the subscription is the only expense.
Start with one general assistant that suits the chosen workflow. Research on workers at large US companies does not establish a universal three-tool ceiling for small agencies. Add a tool only when it has a defined job and earns its place after subscription costs, review time and team effort are counted.
Set an initial review after four weeks, rather than treating a month as a universal deadline. Give the workflow an owner and compare quality and total time with the previous process. Keep, adjust or stop it on that evidence. Extend the trial only if a specific unanswered question is worth testing.
Sources
- Eurostat, 20% of EU enterprises use AI technologies, 11 December 2025
- Eurostat, The use of artificial intelligence technologies in the European Union: key results, 2026 edition, 26 March 2026
- INE, 11,5% das empresas utilizam Inteligência Artificial (IUTICE 2025), 21 November 2025
- European Commission, Digital Decade 2026 country report, Portugal, August 2026
- Harvard Business Review, When Using AI Leads to "Brain Fry" (BCG study of 1,488 US workers), 5 March 2026
- Workday, New Workday Research: Companies Are Leaving AI Gains on the Table, 14 January 2026
- Sage, Finance leaders demand AI transparency (IDC white paper The Verification Tax), 6 July 2026
- Regulation (EU) 2024/1689 (the AI Act), Articles 4 and 50
Read next
31 August 2026 · 9 min read
The five AI workflows that hold up in agencies, and where each breaks
31 August 2026 · 8 min read
What the EU AI Act's transparency rules actually ask of a small agency
