Case · Inbox

Invoices from the inbox, processed automatically.

Invoices arrive as attachments, as links or as plain emails and disappear among everything else. An AI-driven scanner pulls them out, reads them and enters them in the bookkeeping. The email is only archived once the bookkeeping shows the document.
results · measured
anonymous
88%
of the real invoices recognised
174/176
amounts read correctly
11 → 2
unreadable PDFs after OCR
Glazen envelop waaruit documenten in geordende lichtbanen stromen
Starting point

The starting point

The business owner wanted an empty inbox without ever losing an invoice. But invoices arrive in every form: as a PDF attachment, as a scanned document without text, as a link to a portal or as text in the email itself. Sorting them out by hand took time, and the odd one slipped through.
What was wrong
Invoices in different formats and from varying senders
Scanned PDFs without readable text
Invoices among ordinary email and newsletters
No certainty that everything was in the bookkeeping
Solution

How it works

01

Scanning the inbox

A scanner reads the Gmail inbox and judges for each email whether it is an invoice: as an attachment, as an email notification or forwarded.
02

Reading with OCR where needed

If a PDF has no text layer, OCR reads the document. Limits prevent oversized files from slowing down the process.
03

Linking senders to rules

Known senders follow a rule. Unknown senders become a proposal, never an automatic rule.
04

To the bookkeeping

The invoice goes to Moneybird. The email is only archived once Moneybird shows a matching document; if that does not happen, a notification follows.
05

Exceptions as an attention list

Anything uncertain lands in a list with alerts and buttons: Process, Always do this or Ignore.
06

Every change tested first

Before every change, a check runs on the real email history without changing anything. That way you see in advance whether it improves.
Results

What was measured

88%
invoices recognised
against a baseline of 73.6%, on ~1,028 real invoices
174/176
amounts correct
tested on 207 real invoices
11 → 2
unreadable PDFs
after adding OCR
8
errors found in advance
through a critical review before going live, of which 1 critical
The recognition rate measures what share of the manually processed invoices the system recognises as an invoice. It is not a percentage that runs entirely without a person: doubtful cases deliberately go to the attention list.
Guardrails

Secure by design.

The checks are part of the design. That way the system can work independently without you losing control.
Archive only after confirmation by the accounts team
Unknown senders become a proposal, not a rule
Notification if a document does not appear within the waiting window
Every change tested first on the real email history, read-only
Limits on file size and processing
More about AI process automation
Frequently asked questions

Questions about this case.

Ask your own question

Does this also work with Outlook or Microsoft 365?

This case runs on Gmail. The principle also works with other email environments that offer a secure integration. In the quick scan we look at what suits you.

What if an invoice is not recognised?

That is why we measure recognition on real email and keep a watch list. Anything uncertain goes to a person instead of disappearing.

Can I see what happens to my email?

Yes. Every step is logged and exceptions appear in an overview with notifications. Also read how the bookkeeping itself was automated.

Free AI quick scan

Something like this for your business?

In a 45-minute conversation we map out together which tasks you can automate, what it delivers and what you are better off not starting. No obligation, and you keep the outcome.