Artificial intelligence for business: where to start without wasting budget
Before you buy an AI assistant, choose the process you will give it. Five questions to help you do that, with 2025 Eurostat data on which European firms use AI.
· Giordano Vinci · 6 min read
What will you find in this article?
Artificial intelligence for business pays off where a task comes up often, the right answers are already written down somewhere and a person can check the result. Anyone who wants to buy an AI assistant should therefore start from the process and check these three requirements. The tool comes later, because those who start from the product often end up paying for a licence for a job nobody had ever defined.
Artificial intelligence for business: what the Eurostat data say
According to Eurostat, in 2025 17% of small enterprises in the European Union used AI, as did 30.36% of medium ones and 55.03% of large ones. Eurostat counts firms with 10-49 employees as small, 50-249 as medium and 250 or more as large.
The smaller the firm, the less it uses artificial intelligence
European Union firms with 10 or more employees that use AI, by size, 2025. Small firms in pink.
Source: Eurostat, Statistics Explained, Use of artificial intelligence in enterprises, 2025 data (extracted December 2025). EU figure, not Italian.
The figure covers European firms with 10 or more employees, self-employed people included. Micro firms are left out, and this is not a figure for Italy alone. For that group the EU average is 19.95%. If a supplier tells you that “everyone uses it by now”, these numbers say otherwise: in 2025 more than eight in ten small firms were not using it yet.
Among the firms that use it, 34.70% apply it to marketing or sales, and 31.05% to organising administrative and management processes. These are the functions where similar requests and documents arrive every day: the kind of work a chatbot or an agent can support.
Why many companies stop before they start
Eurostat also asked the firms that had considered AI but not adopted it why. The first reason is the lack of adequate skills (70.89%). Next come uncertainty about legal consequences (52.52%) and concerns about privacy and data protection (48.83%).
Here is how I read the figure: the biggest risk is choosing the wrong first use case. So I would put the initial budget on a single process, chosen against clear criteria, not on a platform that then goes looking for something to do. The European Commission says its Apply AI Strategy aims to boost AI adoption, particularly among SMEs. But adoption without a chosen process still ends up as a purchase made blind, even when the supplier is good.
Five questions to choose the process
The five questions below help you tell whether a process suits an AI assistant. If one answer is not a yes, that process is not the first one to hand over for now.
Five questions before handing a process to an AI assistant
If one answer is not a yes, that process is not the first one to hand over for now.
Does the task come up often?It happens many times a week, with similar requests or questions.
Are the right answers already written down?They are in a catalogue, a price list or an up-to-date procedure, not only in someone’s head.
Can someone check the result?A person reads a sample of the answers and says whether they are right.
Can a mistake be put right?If a mistake is costly, the assistant prepares and a colleague approves.
Do you know how it looks today?You have a starting number, even a rough one, taken before you begin.
LU3G diagram
The first question is whether the task comes up often. The tool pays off on what happens many times a week, such as the same information requests or the same product questions. A task that comes up twice a year hardly justifies the time it takes to set it up.
The second question is where the right answers live: in a catalogue, a price list or an up-to-date procedure. If they live only in one person’s head, the chatbot either makes them up or has nothing to say. We cover this in more depth in the article on AI agents and company data.
The third is whether someone in the company can check the result: a person who reads a sample of the answers and says whether they are right. The fourth is what a mistake costs and whether it can be put right. A wrong opening time is fixed in a minute. A price quoted wrongly to a customer, however, is a real problem: in that case the chatbot drafts the reply and a colleague approves it.
The fifth is whether you already know how things stand today. If you do not know how many hours the team spends on that task, in three months you will not be able to say whether the chatbot made a difference. A starting number, even a rough one taken before you begin, lets you compare the results with today.
A well-ordered catalogue is half the work
For Metalsistem Lombardia, a shelving and logistics company, we built a platform of over 100 pages with navigation by application, sector, service and product, plus the Google Ads campaigns with tracking in Google Tag Manager. It is not an artificial intelligence project and no assistant is connected to it.
In our view, though, a catalogue organised this way is what a chatbot answers from best: every piece of information has a page and a precise place. It is the second question on the list, and the one people neglect most. The longest job, often, is putting the content in order, well before choosing the software. That is why the company website matters as much as the chatbot, because that is where the chatbot finds the answers it gives customers.
How much budget to put on the first case
The reasoning is the same as for online advertising. Set an amount you can afford to lose, assign it to a single case and raise it when the numbers justify it. For the method, see the article on how much to invest in advertising. The first case is the process that answered yes to all five questions, and the budget grows only if the starting number improves, that is, if the time spent on that task falls compared with before you started.
On the AI chatbot page you can see what we build: AI models connected to your site, WhatsApp and Messenger, to answer and capture leads. From there you can ask us to talk through the first case to automate. For wider processes there is the page on AI agent development for SMEs.
Eurostat, Statistics Explained: Use of artificial intelligence in enterprises (2025 data, extracted December 2025): EU firms with 10 or more employees using AI (small 17%, medium 30.36%, large 55.03%, average 19.95%), business functions where it is used (marketing or sales 34.70%, administration and management 31.05%), reasons for not adopting it (skills 70.89%, legal issues 52.52%, privacy 48.83%).
Giordano Vinci is CEO of LU3G, a digital agency that works alongside companies and SMEs on websites, SEO strategy, advertising campaigns and online growth. His approach combines consulting, method and strategic vision, with the aim of turning digital into a practical business tool.
Giordano Vinci
Giordano Vinci is CEO of LU3G, a digital agency that works alongside companies and SMEs on websites, SEO strategy, advertising campaigns and online growth. His approach combines consulting, method and strategic vision, with the aim of turning digital into a practical business tool.
More stories, ideas and insights from the Magazine.
CPQ software: when you need it and how to link it to CRM and ERP
When CPQ software pays for itself, which data it should pass to the CRM and the ERP, who keeps that data up to date and how to tell if it works, with Istat 2025 figures on CRM and ERP in SMEs.
READ MORE
Social media editorial plan: how to align content, brand and community
A four-step method to connect a social editorial plan, the brand and the community, with an LU3G case and the signals that show whether the plan works.
READ MORE
How to evaluate an AI agent: metrics, checks and acceptance criteria
An AI agent can answer well and still work badly. The metrics, checks and criteria that show whether it works inside a real process, with the sources behind them.
READ MORE
Before asking how much to spend on advertising, ask what growth is worth
The right advertising budget comes from customer value, margins and the demand the company can serve. With ISTAT, Eurostat and Politecnico di Milano data.
READ MORE
AI agents and MCP: from connecting to data to a verifiable workflow
MCP is the standard AI agents use to connect to company data and tools. What the specification says, what ISTAT data show and how to design a verifiable workflow in an SME.
READ MORE
AI search: content that answers straight away and AI bots to govern
Questions put to AI tools are longer and more precise. A four-step method for pages that answer straight away and for deciding what AI bots can read.
READ MORE
Automated Google Ads: how to give it the right goals
A guide to setting up automated Google Ads campaigns on the company’s own numbers: targets worked out from margin, leads qualified in the CRM, verified measurement and a checklist before raising the budget.
READ MORE
Marketing automation and CRM: why the rules come before the tool
A CRM automation works when data, rules and responsibilities are clear before the tool. How to set it up, where to keep human control and how to measure it, with ISTAT 2025 data.
READ MORE
E-commerce growth: how to coordinate UX, advertising and CRM
An online store grows when advertising, on-site experience and CRM read the same customer journey. ISTAT 2025 and Baymard data, a ROAS calculation, the numbers to share and a roadmap.
READ MORE
AI agents and company data: how to bring them into a controlled process
A five-step guide to connecting an AI agent to company data: choose the process, sort out the sources, decide autonomy, human review and metrics. With ISTAT and Eurostat 2025 data.