Document intelligence

Teams that receive invoices, delivery notes, contracts, forms or other documents and read and type them in by hand: administration, purchasing, logistics, finance or advisory firms. It works when they arrive in volume or in different formats and errors are found late.

• Someone reads every document to copy its data into another system.
• Typing errors show up weeks later, in a reconciliation or a complaint.
• Documents arrive by email, through portals or on paper, and each supplier uses its own format.
• There is a backlog of pending items that grows at peak times.
• Exceptions, such as amounts that do not match or missing data, are found only when recording.

• An inventory of document types and of the rules that validate them.
• Classification and extraction of the fields that matter, with a confidence level per field.
• Validation against your systems and your rules: amounts, suppliers, duplicates and totals.
• A human review loop for whatever fails validation.
• Recording in the target system once validated.

• A catalogue of documents and agreed validation rules.
• A working flow for intake, extraction, validation and recording.
• A review queue for exceptions.
• A measurement of extraction quality on a sample of your documents.
• Documentation and training for the team.

They depend on where your documents arrive and where their data is recorded: email, document manager, ERP or accounting. They are defined in the diagnosis and limited to what is necessary. Before using real documents, we agree with the client which data is processed, where, and who has access to it.

• AI proposes the data; rules validate it and the system calculates amounts and totals.
• A person reviews whatever fails validation, before it reaches your systems.
• Every recorded value keeps its source document.
• Extraction is measured on an agreed sample, not taken for granted.
• Failures raise an alert and are logged.

• Handwritten or very low-quality documents, or formats that change every time: we assess them with tests in the diagnosis.
• Legal or tax interpretation of the documents.
• Universal accuracy guarantees: accuracy is measured on your documents.
• Approving or paying without human oversight.

We measure extraction and validation quality on a sample of your documents, along with typing time, errors caught before recording and pending items, against the agreed baseline.

Tell us which documents arrive and what is done with them. A first conversation is to understand the process and assess whether we are a good fit; if we are not the right option, we will say so.

Contact us →See how we work

• Process automation
• Integration and data
• AI assistants and agents