AI in accounting firms: seven in ten already use it, now the challenge is integrating it

What the industry is saying

On 8 October Wolters Kluwer held its Foro Asesores in Madrid, following the Barcelona edition. AI took up much of the conversation, as MuyPymes reports, and Bas Kniphorst, head of the company's Tax & Accounting division in Europe, summed it up in remarks reported by Negocios del Mundo: firms "are rapidly moving from experimenting with artificial intelligence to integrating it into their daily work".

The figures presented come from the fifth Barómetro de la Asesoría, based on a survey run in March 2026 and published by the company in June:

• 71% of advisers use AI in their daily work, up from 42% the year before.
• The main benefit they see, for 71%, is automation and efficiency. 55% believe it improves client service and 46% that it streamlines their own processes.
• The two biggest concerns are data quality (63%) and lack of in-house experience and training (53%).
• 64% name regulatory pressure as their main challenge, and 95% say new regulations increase their workload.
• 77% of firms have fewer than 10 employees.

In other words: almost every firm has tried AI, most are small and they're swamped by regulation. The question is no longer whether to use it, but how to make it save real hours.

Using AI isn't the same as integrating it

In most firms, "using AI" means someone opens an assistant to summarise a regulation, draft an email or check a document. It's useful, but the savings stay at a few minutes here and there and depend on each person. Integrating it is something else:

The difference isn't the AI model, it's what surrounds it: where the data comes from, what gets validated and who reviews.

The two barriers, and how to overcome them

Data quality

It's the concern of 63% and with good reason: if AI reads a badly scanned invoice or a client is duplicated in the accounting software, the result will be poor however good the model is. What works:

• Validate before posting. A tax ID that doesn't match, a taxable base that doesn't add up or a duplicate invoice should be stopped before it reaches the books, not found at closing.
• A single source of truth. Client and supplier master data in one place, and clean. It's worth spending a few weeks tidying it before automating.
• Flag what's uncertain. When AI isn't sure about a value, it says so and leaves it for a person instead of making it up.

We go into more detail in invoice automation software for accounting firms.

Training

More than half see it as a barrier. The team doesn't need to learn to code, just three things:

• What can be entered, and where. Clear rules on client data: which tools are approved, under which contract and what is never pasted into a personal assistant.
• How to review. Knowing what to check in what AI prepares and when to reject it.
• How to correct. Proposing changes to the process rules when something goes wrong, not just fixing the one case.

On top of that, since February 2025 the AI Act has required companies to take AI literacy measures for the people who use it at work, an obligation the Omnibus softened but didn't remove. We cover it in the EU AI Act after the Digital Omnibus.

What to integrate first

With a small team and a heavy regulatory load, start with what repeats every month, follows clear rules and is done by hand today:

1. Document collection. Asking each client for what's missing, chasing it and filing it when it arrives.
2. Reading and posting invoices. Extracting the data, validating it and proposing the entry.
3. Government notifications. Sorting them, spotting deadlines and alerting whoever is responsible.
4. Repeat client questions. Answering the usual ones with the information in the file and escalating anything that needs judgement.

We go through each one in automating an accounting firm. For the first one, measure how many hours it takes beforehand: without that figure you won't know whether it worked.

Regulation is pushing the same way

The forum also covered what's coming in Spain: mandatory B2B e-invoicing, Verifactu, digital time tracking and pay transparency. On 5 October Spain's Ministry of Finance announced it would delay Verifactu to October 2028 to line it up with e-invoicing, as we explain in our e-invoicing calendar.

The delay buys time, but the direction hasn't changed: every invoice will become structured, traceable data. A firm already working with automated processes and clean data will find that transition far easier than one still keying in PDF invoices.

What isn't delegated

The forum's conclusion was that AI doesn't replace the professional: expert knowledge is still what makes it possible to interpret the information, and AI makes checking information, interpreting results and owning decisions more important. In practice, that means AI prepares and a person decides on anything with consequences for the client: a tax return, a response to a government request, tax advice. How to organise that review without it becoming a bottleneck is covered in supervising AI agents.

How we do it at Amud

With the accounting firms we work with, we don't start with the tool but with the process that eats the most hours. We build an AI agent connected to email, the client portal and the accounting software, with validations before anything is posted, uncertain items flagged for review and every action with consequences waiting for a person's approval. That way AI stops being something each person uses their own way and becomes part of how the firm works.