There's a scene that plays out in thousands of Spanish accountancy and advisory firms every time a request arrives from the Tax Agency: someone on the team drops everything they were doing, prints out folders of documents, and spends days cross-checking paperwork against spreadsheets. It's necessary work, painstaking — and almost entirely mechanical.
The case: 788 invoices, 971 accounting entries
We took on a real case involving a tax inspection of an SME's fiscal year: 788 PDF invoices had to be assigned order numbers, cross-checked one by one against 971 entries in the accounting spreadsheet. The accountancy firm had quoted 3,000 euros for the job — the honest price of tying up a qualified person for weeks on a repetitive task.
Our AI agent did that same job in hours: it read each invoice, located it in the accounts, assigned the order number and generated a viewer so a professional could review the result before signing it off.
The difference isn't just the cost: it's what your team does during the 21 days the Tax Agency gives you to respond.
How it works, without the hype
An AI agent for accountancy firms isn't a chatbot: it's software that runs a complete process. In this case:
- It reads the documents (PDF invoices, including scanned ones) and extracts the relevant data: issuer, date, amount, description.
- It cross-checks each document against the accounts and flags discrepancies: invoices with no matching entry, entries with no matching invoice, amounts that don't add up.
- It generates the deliverable — the numbering, the index, the report — in the format the request calls for.
- It presents everything in a viewer for human review: the professional validates, corrects anything that needs it, and signs off. Responsibility and judgement remain theirs.
That last point isn't a nicety: it's the correct way to apply AI in regulated work. The agent handles the mechanical part; the expert handles the part that requires judgement.
What else can be turned into an agent at an accountancy firm?
The pattern from the invoices case repeats across almost every document-heavy process in the sector: booking received invoices, bank reconciliations, preparing tax returns, checking payroll against collective agreements, responding to requests. The useful question isn't "can AI do my job?" — it's "what part of my week goes on tasks that don't need my judgement?".
That part can almost always be delegated to an agent. And every hour recovered is an hour of real advisory work for the client — which is where an accountancy firm genuinely sets itself apart.
Where to start
Our advice is to start with a bounded, measurable process, like the one in this case: one type of document, one workflow, one verifiable outcome. Within weeks you'll see whether it works with your real data — not slideware — and you can expand from there. </content>