Automating document processing: invoices, forms and requests that process themselves

Automating document processing means that incoming documents, such as invoices, application forms, packing slips and email attachments, are automatically read, checked and passed on to the right system. In many organisations this is still manual work. Someone opens a PDF, looks for the order number, retypes the amount and the supplier into another system and puts the document in a folder. It is repetitive, error-prone and it keeps good people from better work. With today's smart tools, much of it can be automated. In this article we explain how that works and what to look out for.

What is document processing?

Document processing is the process of receiving, reading, checking and recording information from documents. It is the opposite of document generation: instead of creating documents from data, you extract data from documents.

Those documents come in through email, upload forms, scans or integrations with other parties. They are often unstructured: every supplier has a different invoice, every customer fills in a form slightly differently, and attachments can be PDFs, photos or Word files.

How does automated document processing work?

Automated document processing works in four steps: receiving, reading, checking and passing on. The technology for reading has improved greatly in recent years, which means documents without a fixed layout can also be processed well.

Receiving

Documents arrive at a fixed place: a separate email address, an upload form on your website or a folder that is monitored. The system picks up every new document automatically.

Reading

Text recognition converts scans and photos into text. A language model then recognises what kind of document it is and extracts the relevant details: supplier, invoice number, date, lines, total, or for a request the name, the address and the service requested. The difference from older systems is that you no longer need to create a template per supplier. The model understands what an invoice number is, wherever it appears.

Checking

The extracted details are checked against what you already know. Does this supplier exist? Is the order number correct? Do the lines add up to the total? Has this document not already been processed? What is correct goes through. What deviates or is doubtful goes to an employee.

Passing on

Approved details are recorded in the right system: your accounting software, CRM, ERP or a custom database. The original document is linked and archived, so you can always see where a piece of data came from.

Which documents are suitable for automation?

Documents that come in often, contain largely the same kind of information and are currently retyped are the most suitable. Examples we come across:

• purchase invoices that need to be booked in the accounts;
• application forms and intake documents from customers or clients;
• packing slips and delivery documents that need to be matched with orders;
• CVs and cover letters that need to go into a recruitment system;
• certificates and inspection reports that belong to a file.

Why does a person need to keep an eye on it?

A person needs to keep an eye on it because no system processes one hundred percent of documents without errors, and because some decisions require responsibility. The goal is not to automate the employee away, but to shift their work from retyping to checking and deciding.

A well-designed system indicates per document how certain it is. High certainty and everything adds up: through automatically. Low certainty, deviating amounts or an unknown sender: onto a review list. That way most of it goes through automatically and your team keeps control of the exceptions.

What should you look out for when automating document processing?

Pay attention to privacy, traceability and the quality of your reference data.

• Privacy. Documents often contain personal data. Choose services that process data within the EU where necessary, and set out what happens to documents after processing. Especially in healthcare, that is a hard requirement.
• Traceability. Log what the system has read and decided. If something goes wrong, you want to be able to see where.
• Reference data. Checking only works if the data you check against is correct. A messy supplier list produces unnecessary exceptions.
• Start small. Start with one type of document, measure how much goes through automatically and only expand once that runs well.

Which tools do we use to build document processing?

We build document processing with a combination of automation platforms, language models and a custom database. For simple flows we use Make as the orchestrator that ties receiving, reading and passing on together. For larger volumes or more sensitive data we build the logic in a custom backend on Supabase or Xano, with a dedicated review screen for the exceptions.

What do the costs depend on?

The costs depend on the number of document types, the volume, the number of checks, the systems the data needs to go to and the requirements around privacy and logging. In addition, there are running costs per processed document for text recognition and language models. These are usually small compared with the time you save, but do factor them in.

How do you start?

Choose the type of document that causes the most retyping. Count how many come in per week, how long processing takes and which errors occur. That gives you a baseline to measure the effect of automation.

Read more about workflow automation or about automatically processing customer requests, or get in touch.