The workflow
How the AI accountant works
Kontír AI carries every document from arrival to a ready-to-approve posting suggestion — with private AI, on EU hardware. The accountant holds the final word on every entry. Here is the process, step by step.
Five steps
From document to ledger — step by step.
Together, the five steps cover the most labour-intensive, repetitive process in an accounting firm. Every one of them runs automatically — except the decision at the end.
Document ingestion
Invoices, receipts and bank statements arrive as PDFs or images, and Kontír AI processes them automatically. No pre-sorting, renaming or manual keying: the system works from the document itself — a photographed receipt is as valid an input as a machine-generated PDF invoice.
What this means in practice
- Time spent on manual data entry is freed up: documents enter processing the moment they arrive.
- One channel for every document type — invoices, receipts, bank statements.
Data extraction
The system's own AI model extracts the data bookkeeping needs — vendor, amount, VAT rate, date. There is no external AI service in the chain: processing runs on private EU hardware, and financial data never leaves the country.
What this means in practice
- Client data never reaches a public AI provider — not for processing, not for training.
- GDPR, a data processing agreement and a full audit trail support compliance.
NAV reconciliation
Extracted data is automatically matched against NAV Online Invoice records. What matches moves on through the process; what differs — amount, VAT rate, a missing document — is flagged and placed in front of the accountant.
What this means in practice
- The monthly manual NAV cross-check disappears — only flagged discrepancies need attention.
- The system flags discrepancies; the accountant decides what to do about them.
Posting suggestion
For verified entries, the system proposes postings per the Hungarian chart of accounts — not from a generic template, but from the firm's own habits. The same vendor can map to different ledger accounts for different clients: the system learns this per client.
What this means in practice
- Suggestions follow the firm's own practice — its chart-of-accounts usage, its logic.
- The more the firm works with it, the more precise the suggestions become over time.
Approval queue
Every entry lands in an approval queue: the accountant sees the document, the extracted data, the NAV reconciliation result and the posting suggestion in one place. They review, correct where needed and approve — control and responsibility stay with the firm.
What this means in practice
- No black box: behind every suggestion sit the document and the extracted data.
- Nothing is finalised without a decision — the accountant has the final word on every entry.
Division of labour
The AI suggests. The accountant decides.
This is not caution for its own sake — it is the correct division of labour. Machines excel at what is repetitive and high-volume; accountants excel at what requires judgement, expertise and accountability.
Responsibility cannot be delegated
Bookkeeping is framed by legal and professional responsibility — that cannot be offloaded onto an algorithm. Kontír AI therefore prepares but does not decide: the sign-off, and the responsibility, remain with the firm.
Aligned with professional standards
Chamber and professional standards assume human control behind accounting decisions. The approval queue makes that control fast — not optional.
Client trust stays with the firm
Clients trust their accountant, not a piece of software. When the firm approves every entry, that trust is untouched: Kontír AI works in the background, and the client relationship belongs to the firm.
Learns per client
It learns your firm's logic — client by client.
No two accounting firms post the same way — and within a firm, no two clients are alike. The same expense type may belong on one ledger account for one client and on another for the next, because the business, the engagement and the established practice differ.
Kontír AI therefore does not force a generic model onto everyone: it shapes its suggestions around the firm's past decisions, separately for each client. When an accountant corrects a suggestion in the approval queue, they are also teaching the system — so over time, the suggestions track the firm's practice ever more closely.
Read more: What is an AI accountant? →- Learns per client — not a shared average, but each client's established posting practice.
- Follows the firm's chart-of-accounts usage, not a generic template.
- Approvals and corrections make the suggestions more precise over time.
The platform keeps growing
Posting is just the beginning.
Kontír AI expands on the same private AI platform as the firm's needs grow — and every module follows the same principle: the AI prepares, the professional decides.
Payroll automation
Preparing payroll's repetitive administration — under the same approval principle.
VAT return assistant
Return preparation built on entries already processed and NAV-reconciled.
Year-end statement preparation
Taking the most labour-intensive preparatory steps of year-end closing off the firm.
Cash-flow forecasting
A forward-looking view of clients' cash flow, drawn from the books.
Client portal
An orderly channel between the firm and its clients for documents and questions.
See it in action.
Show us how your firm works today — in a demo, we will show you what Kontír AI takes over, and how.