Accounting · Accounting firms and administrations
AI automation for accounting firms
An automated accounting firm receives its clients' documents over WhatsApp or email, reads them with OCR and artificial intelligence, validates them against the tax authority and bank movements, files them by client and period, and assembles the tax return data without anyone typing a figure.
What this service covers
The problem isn't the accounting software
Most firms already have good accounting software. The problem sits upstream: getting the documents, reading them, checking them, and entering them. That work, which needs no professional judgement but eats most of the month, is what can be automated.
The workflow, from document to tax return
- The client sends the invoice over WhatsApp, as a photo or PDF, or it arrives by email.
- A conversational agent receives it, recognises it, and records it without the firm stepping in.
- OCR extracts supplier, amount, date, and document type.
- The system checks with the tax authority whether that document actually belongs to that client.
- It cross-checks what's in Drive against what's registered and flags the documents that are missing.
- A banking agent processes statements and reconciles movements.
- Everything is filed by client and period, named and foldered by the firm's own convention.
- At close, the tax return data is assembled: VAT debit, VAT credit, gross income, and debits.
The reminders a person sends today
Chasing documents is one of the most thankless tasks in a firm. The system sends reminders on the 5th, 10th, and 15th, and bank statement reminders a day later, shifting the date when it falls on a holiday or weekend. Clients who haven't sent anything get the nudge; those who have, don't.
What happens with what the system can't read
Nothing is recorded by force. A badly photographed document, ambiguous data, or an out-of-range amount is flagged as an exception and sent to a review queue, with the image and the proposed extraction side by side. The goal isn't one hundred percent automatic: it's for the team to review exceptions instead of reviewing everything.
Professional confidentiality and data protection
- Documentation stays on the infrastructure the firm defines, its own or its cloud.
- Per-user access: each person sees only the folders that concern them.
- Auditable record of what was processed, when, and with what result.
- Client documentation is not used to train third-party models.
What changes in the firm's operation
The team stops typing and moves to reviewing. Missing items surface during the month instead of at closing. And the firm can take on more clients without adding administrative staff, which is where the real return shows: not in shrinking the structure, but in being able to grow without expanding it.
Accounting software, standalone OCR, and an agent workflow
These three get confused in proposals all the time. They solve different parts of the same problem and none replaces the others.
| Criterion | Accounting software | Standalone OCR | Agent workflow |
|---|---|---|---|
| Records the accounting | Yes | No | No, it feeds yours |
| Receives the client's document | No | No | Yes, over WhatsApp or email |
| Reads the document's data | No | Yes | Yes |
| Validates against the tax authority | Partly | No | Yes |
| Detects missing documents | No | No | Yes, by cross-checking |
| Chases the client for what's missing | No | No | Yes, with automatic reminders |
| Reconciles bank statements | Partly | No | Yes |
The agent workflow doesn't replace the accounting software: it feeds it. It integrates by API or import files with whatever the firm already uses.
Real use cases
Implementation process
- 1
Diagnostic
We analyze your process and tell you what to automate, how, and what return to expect.
- 2
Implementation
We build the solution connected to your systems, with working deliveries and real tests.
- 3
Operation and improvement
We leave it running, train your team, and measure results.
The full detail, with deliverables per stage, is in how we work.
Frequently asked questions
Do I have to change my accounting software?
No. The workflow integrates with the one you already use, by API or import files. If your system is closed and allows neither, that surfaces during the mapping stage and alternatives are assessed before committing to anything.
Do my clients have to learn something new?
No. They keep sending documents as always, over WhatsApp or email. The change happens on the firm's side, not the client's, which is what keeps the rollout from hitting the usual resistance.
How does it validate against the tax authority?
The system connects to verify whether each document actually belongs to that client, and cross-checks what's registered there against what's on file to detect what's missing.
What about professional confidentiality?
Documentation stays on the infrastructure the firm defines, with per-user access and an auditable record. It is not used to train third-party models.
Does it suit a small firm?
What defines the case is the volume of documents processed per month, not the number of professionals. A three-person firm with many small clients can have more repetitive volume than a ten-person firm with a few large ones.
How long does implementation take?
A focused workflow over one document type can be running in weeks. In the published case, the full system with three agents was in production one month after the deposit.
How accurate is the reading?
It depends on the type and quality of the documents you receive, which is why it's measured with a real batch of yours during implementation rather than promising a percentage upfront.
What does it cost?
It's quoted by hours and resources according to complexity. Every project includes one month of warranty, the code stays on the firm's server, and later maintenance is optional.
How many hours a month go into chasing documents?
Tell us how your clients' documentation arrives today and we'll tell you which part of the workflow can be automated, and in what order.