You can automate the invoices you receive by email by connecting the inbox to a system that saves the file, extracts the data, checks for issues, and prepares a draft in your software. Administration receives the document with the fields already filled in so they can review them.
The workflow we suggest to get started is this:
Email received → PDF saved → data extracted → checks → draft for review.
The key part is the checks: what happens if the supplier sends the same invoice again, a page is missing, or the destination software stops responding. Here you’ll find a practical design for tackling that process.
Define where it starts and where it ends
Before choosing tools, agree on three things with the person who handles administration:
- Input: an inbox, a folder, or a label that gathers the invoices received.
- Output: a draft in the management software, a review inbox, or an import file.
- Owner: who reviews doubtful documents and confirms the entries.
Keep the first pilot to one document type and one recipient company. For example, supplier invoices in PDF that arrive in an email folder. Quotes, delivery notes, and receipts need their own rules.
If you’re still deciding which task to tackle, our guide on how to choose the first process to automate helps you sort out that decision.
The five steps in the flow
1. Receive the email and keep the original
The flow detects the messages in the chosen folder and collects their attachments. Save the PDF with a reference to the source email, the sender, and the date it was received. That way, you can go back to the document whenever a question comes up.
It’s worth logging each attachment separately: one email can include several invoices or an invoice plus another document. An unreadable file should show up as an issue, with a person responsible for solving it.
2. Extract the fields you actually need
First, define the columns your software accepts. An initial list may include supplier, tax ID, invoice number, date, currency, subtotal, taxes, and total. Add line items only if you’re going to use them.
There are tools built for this kind of reading. For example, Google Document AI’s Invoice Parser extracts header fields and invoice lines and supports Spanish. That lets you get structured data from the document; connecting it to your software is a separate step in the project.
Save the extracted value alongside the original. When a field is missing, the system should leave it pending instead of filling it in by guesswork.
3. Check the data and spot possible duplicates
This stage combines business rules with document review. As a starting point, check:
- That the recipient matches your company.
- That the supplier can be matched to an existing record or needs to be created.
- That the required fields for import are present.
- That the amounts and currency are consistent with the document, taking into account discounts, surcharges, or withholding tax when they appear.
- That there is no previous record with the same combination of supplier, invoice number, and recipient company.
Comparing the file itself also helps identify an identical attachment. But a resend can include the same document under a different name: the PDF name alone is not enough as a control.
The Microsoft AI Builder invoice model provides confidence scores for different fields. Use them as a signal to decide what to review. A high score does not replace checking the supplier, the recipient, or the total.
4. Prepare the record in the work tool
Check whether your software lets you create drafts through an integration or import a file with the defined fields. This is one of the first project tests: reading an invoice well is of little use if you still have to type it in again by hand.
If the software only allows final records, place the review before that step. An intermediate inbox can show the PDF on one side and the data on the other, with the reason for each issue.
When logging the result, keep the reference returned by the destination software. That confirmation lets you know what has been created and avoids repeating the operation blindly when there’s a connection failure.
5. Review, approve, and close
At the beginning, review all drafts. Correct the necessary fields, confirm the destination, and record who approved each document.
Keep the statuses clear: received, pending review, approved, recorded, or issue. The success notice should match a verified result. “Data extracted” and “record created” are two different moments.
This first flow can end with the invoice prepared and validated. Payment approval requires a specific design and permissions.
An example: an invoice arrives and then is sent again
Imagine an invoice from a known supplier with number F-1042 and a total of 1,452 euros. This is an illustrative example, meant to check how the flow behaves.
| What happens | What the system should do |
|---|---|
| The PDF arrives for the first time | Save the original, extract the fields, and prepare a draft. |
| The supplier sends the same invoice again | Detect the match and link the existing record for review. |
| The total is read with doubt | Flag the field and keep the document pending. |
| The software does not respond when saving | Mark the operation as pending confirmation and check whether it was created before trying again. |
| A person approves the draft | Record the approval and confirm the result in the destination software. |
This test teaches more than a demo where every file goes through smoothly. Include a document that must stop as well: the team needs to see how an exception is handled.
What tools you need and what to check first
The project has three parts: connecting to email, reading the document, and connecting to the destination software. The exact tool depends on what you already use and on the permissions and integrations available.
If you work in the Microsoft environment, their documentation explains how to use AI Builder invoice extraction inside Power Automate. It’s an option for the reading stage, which must be completed with your process’s inputs, validations, and outputs.
Also test the less convenient files. Microsoft warns that, when selecting pages to extract an invoice, the range must contain only one: mixing pages from two invoices can return partial data from both. If a supplier sends several documents together, they’ll need to be separated or reviewed.
Before signing up, ask for a proposal that specifies:
- Which inboxes, formats, and software are included.
- Who can view the documents and where they are stored.
- What happens with duplicates, errors, and unreadable documents.
- Which part of the cost corresponds to implementation, licenses, usage, and maintenance.
- How system access is handed over and who handles an issue.
How to know if it’s worth it
Measure the full job before and during the pilot: finding the email, opening the file, entering the data, filing it, and fixing errors. Then measure how long the draft review takes and how much time issues require.
You can start with a varied sample of documents: regular suppliers, new formats, multiple pages, a duplicate, and a hard-to-read file. That sample helps uncover problems; ongoing operation will show whether the time saved is real.
Watch three outcomes: total handling time, corrected fields, and unresolved pending documents. If the review still costs almost as much as the manual process, it’s worth adjusting the scope or the reading step before scaling up.
The next step for your company
Gather the name of your management software, how the invoices arrive, and an anonymized sample of the usual formats. With that, you can check the input and output connection, and map out the first flow with administration.
At Digency, we design custom AI automations for your tools. Tell us how you receive and register invoices and we’ll define which steps to automate, where to keep the review, and how to check the result.
Cover: image generated with AI.
