Tax Observatory › Digital and compliance › Artificial intelligence in business: what can really be automated

Artificial intelligence in business: what can really be automated

Invoices, bank, deadlines, email, management accounts: many repetitive tasks can already be handed to a machine, provided the starting point is the process, a person keeps the final check and the rules on data and employment are respected.

Everyone talks about it; few use it well. Many businesses have tried artificial intelligence to write an email or summarise a document, and stopped there. It is like buying a van and using it to fetch the newspaper.

The real point lies elsewhere. Every business has tasks that are repeated in the same way every day, that nobody enjoys and where mistakes creep in through fatigue. Those are the tasks that can now be handed to a machine, leaving the decisions to people.

Where time is lost, and where it can be recovered

Look at an ordinary day in the accounts office. Invoices, bank statements, customer emails, reminders and documents to file all arrive. Someone opens them, reads them, keys them into a program, puts them in a folder, replies. Much of this work always follows the same rules.

Here is what is already being automated today, with tools that are often within reach of a small or medium-sized business.

• Invoices and bookkeeping entries. The system reads the electronic invoice, suggests the account and the VAT rate based on how that supplier has been recorded in the past, and flags anything that does not add up. A person checks and confirms.
• Bank. Most bank statement transactions are matched automatically to invoices and receipts. A person checks the result and concentrates on the items with no match.
• Deadlines. Taxes, social security contributions, contract renewals, expiring documents: a calendar that updates itself from the data and gives timely notice to whoever has to act.
• Email. Incoming emails are read, sorted and filed in the right file. For routine messages a draft reply is prepared, which someone reviews before it is sent.
• Management accounts. Margins per job, staff costs, cash position over the coming weeks: figures updated every day, instead of a report that arrives after the quarter has closed.
• Meetings. Recording, transcription and minutes showing the decisions taken and who has to do what, with participants informed beforehand.

This is not science fiction, and it does not always take a large group’s budget. It takes a thorough knowledge of the process, because a machine is quick to do both the right things and the wrong ones.

Three mistakes to avoid

Starting from the tool. Businesses buy a subscription and then look for something to do with it. The starting point should be the work: which activity takes the most hours, how often it is repeated, what happens if it is done badly.

Removing human oversight. Artificial intelligence makes mistakes, and when it does, it makes them with great confidence. Any serious automation includes a point at which a person looks and approves, especially where money, customers or employees are involved.

Forgetting the rules. The use of these systems is not a lawless area.

The European regulation on artificial intelligence (Regulation (EU) 2024/1689) asks those who use these systems in a business to take steps to build the knowledge and skills of the people who work with them. For the most sensitive uses, such as recruitment and personnel management, it lays down heavier obligations, which will apply from December 2027.

In Italy, Law No. 132 of 2025 recalls the employer’s duty to inform workers when it uses fully automated systems that decide on or monitor aspects of the employment relationship: hiring, duties, assessments. And when automation touches people’s work, as with employees’ email or recorded meetings, those concerned must be informed first, and it must be checked whether a trade union agreement or an authorisation from the Labour Inspectorate is needed.

Then there is the data protection regulation. Before a system is allowed to read the data of customers and employees, the business needs to know where the data end up, who processes them and with what safeguards, to put the supplier’s commitments in writing and, for the riskiest uses, to assess the impact in advance. No decision that affects a person should be left to the machine alone.

Those who automate without looking at these aspects expose themselves to challenges that cost more than the time saved.

What we do

Our firm applies these methods to itself first. For bookkeeping entries, bank reconciliations, deadline schedules, payroll checks and reports we use automated tools as a support. The work, the judgement and the responsibility remain with the professional, who checks before anything leaves the firm.

We have already brought the same methods into several client businesses: management accounts updated automatically, dashboards for each job, document flows that no longer pass from hand to hand. We do so starting from the figures and the rules we follow every day: which data are needed, which checks must be kept, which obligations have deadlines and consequences.

The process is simple.

1. Together we look at how your accounts office works today and measure where the hours go.
2. We choose two or three processes to start with, the ones that are easiest to measure and to check.
3. We run them alongside the current method until the figures match.
4. We put the rules in writing with you: who checks, what stays with people and, together with whoever looks after privacy, personnel and systems in your business, how data are handled and what must be communicated to employees.

We do not promise businesses that run themselves. We say something more modest and more useful: the hours that today go into repetitive work can go back to those who know how to use them better.

If you want to understand where to start in your business, write to us. The first meeting is for looking at the processes, not for selling a program.

Sources

Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (EUR-Lex, consolidated text of 27 July 2026)

Italian Law No. 132 of 23 September 2025, Articles 11 and 13 (Normattiva, text in force on 10 November 2026)

Article 1-bis of Legislative Decree No. 152 of 26 May 1997 (Normattiva, text in force on 10 November 2026)

Article 4 of Law No. 300 of 20 May 1970 (Normattiva, text in force on 10 November 2026)

Regulation (EU) 2016/679, Articles 13, 22, 28, 35 and 44 (EUR-Lex)