AI software for accountants: how to choose the right tools to automate your firm

Every week a new pitch arrives: "the AI software that will revolutionize your firm." An accountant who already has practice software, digital archiving and three different portals rightly wonders where this new piece is supposed to fit. This guide brings order: which families of tools exist, what they really automate, how they integrate with what the firm already uses, and what to ask before signing. If you want an overview of where AI helps an accounting firm, start with AI for accountants.

Florence's Catasto of 1427: a mountain of paper turned into data

In 1427 the Republic of Florence, short of money for its wars, introduced a new tax and, to calculate it, asked every head of household to file a written declaration: the portata. It listed houses, land, credits, investments, debts and dependents. The Catasto officials received these declarations, checked them and summarized them in orderly registers, the campioni. Most survive in the State Archives of Florence, and historians David Herlihy and Christiane Klapisch-Zuber turned them into a database still available online, which allowed scholars to study an entire Tuscan society.

You may ask: what does a fifteenth-century tax have to do with my firm? The mechanism is the same. On one side, the taxpayer bringing messy papers written their own way. On the other, an office that must turn them into comparable data, check them and produce a number. The Florentine officials worked by hand, patiently, with a standard format. The value lay in the passage from portata to campione.

Your firm today is that office. AI software, when it works, speeds up exactly that passage: from the client's paper to clean data in your practice software. When it doesn't, it adds one more place where documents get lost.

What really needs automating (and what already is)

Part of the work is already structured. In Italy, e-invoices pass through the government exchange system (SdI) in XML and practice software imports them without any AI "reading." The room for AI is elsewhere:

• unstructured documents: receipts, contracts, medical expenses for tax returns, foreign documents, PDFs of every kind;
• bank statements and reconciliation: matching transactions to invoices when descriptions are poorly written;
• client communication: reminders, requests for missing documents, answers to recurring questions;
• reading rules and guidance: summarizing circulars and rulings, outlining what to study;
• reports and commentary: draft comments on financial statements or quarterly trends.

The five families of tools

No family solves everything alone. In a typical scenario, an eight-person firm with three hundred bookkeeping clients gets more from a well-integrated document portal and a few automations than from the most brilliant assistant on the market.

How to choose: questions for the vendor

1. Does it integrate with my practice software? How exactly? Ask for a live example, not a slide.
2. Which documents does it handle well, and which not? Ask what share needs manual correction, and test it on your worst documents.
3. Where is data processed and who sees it? EU servers or adequate safeguards, a GDPR Art. 28 processing agreement, no use of your data for training.
4. What is logged? Who posted what, what the system suggested, what the operator corrected.
5. What does it really cost? Fee, per-document or per-client cost, setup, training. Do the math on peak-month volumes.
6. How do I leave? Export of data and documents in open formats.

Client disclosure and the national council's clause

Since 10 October 2025, Italy's Law 132/2025 requires professionals to inform clients about the AI systems they use (Art. 13). On 9 October 2025 the national council of Italian accountants released a model contract clause for engagement letters, as part of its guide series on AI for accountants. Start from that model and adapt it to the tools you actually use.

A four-stage path

1. Map the flows (one week). How many documents arrive, from where, in which format, who touches them and how many times.
2. Pick a pilot flow. For example medical expenses for tax returns, or bank statements for small bookkeeping clients.
3. Test a full cycle. A month of bookkeeping or a reduced tax-return campaign. Measure time before and after, and the correction rate.
4. Integrate or drop. If the pilot works, extend it and write the procedure. If not, understand why before changing tools.

Mistakes to avoid

• Buying the AI module because "everyone has it." Without a pilot you'll never know if it works.
• Adding a portal no client uses. If clients keep sending photos on WhatsApp, the change must be accompanied.
• Ignoring staff. The person who has posted entries for twenty years knows the odd cases better than any vendor.
• Forgetting spot checks. Automatic extraction rarely errs, but it does err.

Going further

This guide is part of our dossier on AI for professional firms. If you also handle payroll or work with a labor consultant, read AI for labor consultants.

The Catasto officials had no software, but they had a format and a method. AI software helps when the method is already there. If you want to build it together, see how we work with accountants, our services, and book a free intro call.