In many businesses the incoming invoice is the document with the most stops. It lands in some inbox, someone forwards it, prints it, puts it in a folder. Purchasing checks whether the goods have arrived. The managing director signs it off. Accounts scans it back in and sends it to the tax adviser. It gets paid once all that is done, and sometimes only after a reminder.
Almost all of this route can run digitally. Two things help: e-invoices, which bring their data with them, and AI, which can read invoices that do not. Neither replaces checking whether the invoice is right.
The route of an incoming invoice
On paper or digitally, every incoming invoice goes through the same steps:
- Receipt: the invoice arrives, by email, post or portal.
- Capture: supplier, invoice number, date, amounts and tax rates are recorded.
- Matching: which purchase order, job or cost centre does it belong to?
- Checking: are quantity and price right? Have the goods arrived, has the work been done?
- Approval: someone with authority approves the invoice for payment.
- Posting: the invoice and its data go to accounts or the tax adviser.
- Payment: it is paid on time, ideally with the early-payment discount.
Today almost every one of these steps involves a person retyping, forwarding or searching. The aim is not to take people out. The aim is for them to do only what they are needed for: checking and deciding.
Step 1: one inbox instead of many
As long as invoices land in personal mailboxes, no software will help. The first step is therefore organisational: one central address such as invoices@yourbusiness.com, which you give to all suppliers. Paper invoices are scanned in one place and end up in the same inbox.
Step 2: capture without retyping
This depends on the form in which the invoice arrives.
E-invoice: an XRechnung or ZUGFeRD invoice brings its data in fixed fields. A program reads it directly, without text recognition and without guessing. Since 2025 German businesses have had to be able to receive e-invoices, and from 2027 and 2028 there will be more and more of them. More in e-invoicing in your business.
PDF and scan: here a program has to read the data from the image. This used to be done with text recognition and fixed templates per supplier. Today AI models can also read invoices they have never seen before: they find invoice number, amounts and line items even though every supplier lays out their invoice differently. That usually works well, but not always. So every invoice read this way needs checking before it moves on. How AI handles documents is covered in how AI reads your documents.
Step 3: matching
This saves the most time. An invoice matched automatically to the right purchase order no longer has to be looked for. Two things are needed:
- The order has to be in the system. If you order by phone or by informal email, there is nothing to match the invoice against later.
- The supplier has to be identified clearly. That works through the VAT identification number, bank details or supplier number, provided your master data is kept up to date.
AI can make suggestions here: this invoice probably belongs to that order, this line probably goes to that cost centre. A suggestion stays a suggestion until someone confirms it.
Step 4: checking
For materials and goods, the tried and tested route is matching purchase order, goods receipt and invoice. If quantities and prices agree, the invoice can go to approval without further effort. If something differs, it goes to the person who ordered, with a note of what does not fit.
For services, repairs or consulting there is often no goods receipt. Here the person who ordered confirms that the work was done.
Step 5: approval
Who may approve which invoice? That is a question for management, not for the software. Amount limits are typical: up to a certain amount the person who ordered, above that the head of department as well, and above a higher amount the managing director. The software then applies these rules, sends the invoice to the right person, chases anything left lying and records who approved what and when. How such workflows are set up is covered in digital approvals.
AI approves nothing. It can read, match and point out anything unusual, for instance a supplier’s changed bank details. The decision to transfer money is made by a person. Why it should stay that way is explained in what AI cannot do for your business.
Step 6: handover to accounts
The approved invoice goes with its data to accounts, which for most German businesses means the tax adviser via DATEV. How that handover works is described in handing documents to DATEV. How invoices have to be kept, especially e-invoices in their original form, is something to settle with your tax adviser.
What to use
There are dedicated programs for incoming invoices, and many inventory and ERP systems cover part of this. First check what you already have and simply never set up.
A solution of your own pays off when the route of an invoice in your business is unusual: when invoices are checked against jobs or projects held in another system, when several companies or sites with their own approval rules come together, or when the invoice data has to go into a program with no ready-made connection. Where AI really takes work off your hands and where it does not is best seen on your real invoices. How we approach such questions is described on our AI Consulting page.
Checklist
- Is there one central address for incoming invoices, and do all suppliers know it?
- How many invoices already arrive as e-invoices?
- Are orders in the system before the invoice arrives?
- Are suppliers on file with their VAT identification number and bank details?
- Is it set down in writing who may approve up to which amount?
- Who checks invoices read by AI before they move on?
