AI for businesses: where to start in 2026

Artificial intelligence is no longer an experiment reserved for large companies: mature tools and affordable costs make it useful for an SME too. The real risk is starting from the wrong end, with the technology rather than the problem. Here is how to set up a first project that delivers measurable results, and which rules to keep in mind.

Where AI really helps

The projects that work best share three traits: a repetitive task, a lot of text or documents to handle, and a person who stays responsible for the outcome. Some concrete examples:

• Customer support: an assistant that answers frequent questions on the website or in chat, based on your own documents, and hands complex cases to a person.
• Documents and data: extracting information from invoices, orders, contracts or emails and entering it into your business software without retyping it.
• Internal knowledge: a search engine over procedures, manuals and project history that answers in natural language.
• Content: first drafts of product pages, descriptions and translations, always reviewed by a person.
• Automations: connecting different tools (CRM, email, calendar, business software) to remove the manual steps between one system and another.

How to choose your first project

1. Start from a measurable problem. "Answer clients faster" is vague; "cut the response time to quote requests from two days to two hours" is a goal.
2. Look at the data you have. AI works well when it has documents, history and procedures to draw on. If the information is scattered or out of date, the first step is to put it in order.
3. Start small. A pilot of a few weeks on a single process tells you more than any preliminary study.
4. Keep a person in the loop. Especially at first, the AI proposes and a person approves. It is safer, and it lets you improve the system through real use.
5. Measure before and after. Time saved, errors, requests handled: without numbers you cannot tell whether the project works.

Data, security and privacy

Before putting company data into an AI tool, check where it is processed and stored, whether it is used to train models and who can access it. Personal data is subject to the GDPR: a legal basis, a privacy notice, contracts with suppliers and, where needed, an impact assessment. Business versions of the main services generally offer contractual guarantees that free versions do not.

The rules: the EU AI Act

The EU AI Act applies in stages. Since February 2025 certain practices deemed unacceptable have been banned, and organisations using AI systems have had to look after their staff's AI literacy. The so-called "Digital Omnibus", in force since July 2026, clarified that duty: it does not require a specific level of skill to be guaranteed, but for it to be supported with appropriate measures, such as training and instructions for use.

Since 2 August 2026 the transparency obligations have applied: people talking to a chatbot must know they are dealing with an AI, and images, video and audio generated or altered with AI that look real (so-called deepfakes) must be disclosed as such. The most demanding obligations, those for "high-risk" systems such as the ones used to select staff or assess creditworthiness, have been postponed to 2 December 2027.

Individual countries add their own rules too. Italy, for example, has had a national AI law in force since 10 October 2025 (Law 132/2025), which among other things requires employers to inform staff when they use AI systems that affect them, and professionals to tell clients whether and how they use AI in their work.

For most SMEs, which use AI for support, documents and automation, all this comes down to a few concrete steps: train the people who use the tools, be transparent with clients and employees, and keep track of which systems you use and for what.

The right AI project is not the most ambitious one: it is the one that solves a specific problem and that people genuinely use every day.

Conclusion

Getting started with AI in 2026 does not take a large investment or a dedicated department. It takes a clear problem, orderly data, a well-defined pilot and attention to the rules. From there, you scale what works.

This article is for information only, is up to date as of September 2026 and does not constitute legal advice.