Reading supplier invoices with AI

The problem

Every morning, the same pile. Supplier invoices arriving by e-mail, as PDFs, as scanned letters — each with its own layout, its own language, its own references. They have to be opened one by one: identify the branch concerned, check whether it is an invoice or a credit note, note the date and the due date, understand what was purchased, then decide which account to post it to. Repetitive retyping that swallows entire days of the accounting department — while the genuinely tricky cases wait.

Our approach: the AI reads, the human decides

We developed a supplier invoice reading module for a client, integrated directly into their existing accounting process.

1. Reading the document: the module extracts and structures the text of the invoice, whether a native PDF or a scan. The supplier's layout does not matter.
2. Understanding with an LLM model: a language model (LLM) interprets the document the way an accountant would. It identifies the issuer, the destination branch, the nature of the document — invoice or credit note —, the date, the due date, the subject and the detailed content of the services billed.
3. Proposed accounting entry: the module proposes the account or accounts to post to, combining two sources — the history of past entries for this supplier and this type of service, and the client's chart of accounts.
4. Systematic human control: nothing is posted automatically. The employee sees the document, the extracted values and the proposed entry side by side, and validates, corrects or completes them.
5. Learning through use: every correction enriches the history, and the following proposals become more accurate.

One principle: the human keeps control

This is the point on which we do not compromise. The AI does not replace the accountant, it prepares the work. No entry reaches the books without human approval, every proposed value stays visible and editable, and the employee can take the data entry fully back by hand at any time. Accounting responsibility stays where it belongs.

What the client gains

• Hours recovered every week on repetitive data entry
• Time freed for complex cases: disputes, rebilling, splits between branches — where human expertise really counts
• Fewer retyping errors: dates, amounts and due dates are no longer keyed in again
• Consistent posting quality, regardless of who handles the pile that day
• Shorter processing times: fewer invoices paid late

Who is it for?

Any organisation processing a regular volume of supplier invoices, especially one with several branches, entities or cost centres: multi-site groups, trustee firms, distributors, construction companies, institutions. The same principle — the AI prepares, the human validates — applies to other incoming documents: orders, delivery notes, statements, contracts.