Automate Digital Invoice Verification and Approval

Digital invoice verification involves three types of checks, which in practice are usually combined into a single step: formal, arithmetic, and substantive checks. This distinction is essential for automation, as the three checks can be performed to varying degrees without human intervention.

Here’s an example from accounts payable: An invoice arrives as a PDF; the amounts are correct, but the order number is missing. The purchasing clerk follows up; two days later, it turns out that only part of the shipment has arrived. The invoice was recognized correctly; what was missing was the link to the order.

The formal and arithmetic checks are now performed entirely by machine. The substantive check requires information that is not included on the invoice itself.

What Digital Invoice Verification Checks

Digital invoice verification involves electronically capturing incoming invoices and automatically verifying their accuracy before they are approved for payment. Each of the three checks answers a different question:

• Formal Review – Does the document meet the legal requirements for an invoice? § 14(4) of the Value-Added Tax Act (UStG) requires ten mandatory items of information. These include the names and addresses of both parties, tax ID number or VAT ID number, a sequential invoice number, and the amount due broken down by tax rate. If any of these items are missing, the right to claim a tax credit is at risk.
• Numerical Review – Do the Numbers Add Up? Line items, subtotals, tax rates, total amount. The software handles all of this arithmetic. Our guide shows how the software also pre-assigns accounts to the individual line items. Practical Test of AI Account Assignment.
• Substantive Review – Does the invoice match the services ordered and delivered? This requires a reconciliation of the purchase order and the goods receipt—the so-called three-way reconciliation.

The first two checks can be performed directly on the document itself. The third check requires two additional documents: The three-way reconciliation compares the purchase order, goods receipt, and invoice side by side. For each line item, it verifies whether the quantity billed matches the quantity delivered, whether the unit price matches the agreed-upon price, and whether the line item total matches the product of the two.

Predefined tolerances for price and quantity variances prevent every rounding difference from triggering a query. For partial deliveries, the system checks against the open order balance; for framework agreements, it checks against the call-off quantity. If the posted goods receipt notification is missing, someone who knows what has arrived is still needed.

How the Process Works from Incoming Invoice to Approval

From the time it is received until it is approved, an invoice goes through six stages:

• Entrance – Invoices arrive as paper documents, PDF attachments, or structured data records. A centralized intake channel is essential for any automation; without it, invoices end up scattered before anyone can review them.
• Recognition – The software reads the fields from the document. For paper and image PDFs, this is handled by the OCR for invoices, at ZUGFeRD or XRechnung The data is already in a structured format.
• Assignment – The invoice is assigned to a supplier, a purchase order, and a cost center. This can lead to follow-up questions, such as when order numbers are missing or supplier master data is ambiguous.
• Exam – The three tests from the previous section, in that order.
• Approval – The authorized person confirms the payment. If a full three-way reconciliation is completed and the amount is below a certain threshold, this step is omitted; in all other cases, it is handled according to delegation rules.
• archiving – According to Section 14b(1) of the Value-Added Tax Act (UStG) Companies must keep all invoices they receive eight years Retain. The retention period begins at the end of the calendar year in which the invoice was issued. An invoice issued in 2026 will therefore remain in the archive until the end of 2034 – Archived in an audit-compliant manner.

Until the end of 2024, the period was ten years; many process descriptions still cite this value. Two exceptions apply to the new retention period. For credit institutions, insurance companies, and securities firms, the shortened period under Section 27(40) of the Value-Added Tax Act applies only to invoices for which the old retention period was still in effect as of January 1, 2026. Section 147(3) of the German Fiscal Code (Abgabenordnung) remains unaffected; other retention requirements may have longer time limits.

Why Automated Invoice Verification Fails in Practice

The percentage of invoices that pass through the automated invoice verification process without human intervention is called Dark Processing Rate. This metric measures how much work is eliminated and thus provides more insight into the process than detection accuracy.

In the projects we support, the same three causes keep coming up:

• Missing Order References – No three-way comparison without an order number.
• Inconsistent Supplier Master Data – The same supplier with multiple spellings prevents matching, even when all fields are correctly recognized.
• Framework Agreements and Partial Deliveries – An invoice for a partial shipment can only be verified against a total order if someone has set up the rule for that.

Recognition accuracy is considered a bottleneck and is specified in requests for proposals and demonstrated in demos. In the projects we see, however, automation fails less often due to an incorrectly read amount than due to a process that does not provide the necessary reference data.

This determines the order of steps for an automation project: first, clarify master data and ordering requirements; then, select the software.

In this process, Konfuzio handles the recognition, mapping, and comparison against order data, and forwards the verified fields—along with a confidence score and reference—to the leading system.

This article discusses which tool is best suited for this purpose—invoicing software, a document management system, an IDP platform, or an ERP module—and how to identify flexible interfaces. Tools for Digital Invoice Approval.

How the e-invoicing requirement is changing things

Definition and Scope

Section 14(1), sentence 3, of the German Value-Added Tax Act (UStG) defines an electronic invoice as an invoice that is „issued, transmitted, and received in a structured electronic format and allows for electronic processing.“ The format must either comply with the European standard set forth in Directive 2014/55/EU or must be agreed upon between the sender and the recipient. This second option applies only if the required information can be extracted from it into a standard-compliant or interoperable format. It leaves the door open for established EDI procedures.

According to paragraph 2, sentence 2, item 1, this obligation applies to transactions between businesses. A prerequisite is that both parties are established in Germany or in one of the territories specified in § 1, paragraph 3 of the Value-Added Tax Act (UStG).

Transition periods for issuance

Companies are already required to receive such invoices. When it comes to issuing them, the rules vary § 27, paragraph 38 of the Value-Added Tax Act (UStG) The transition periods:

• Through December 31, 2026 – Paper and, with the recipient’s consent, non-compliant electronic formats will remain permissible for transactions after December 31, 2024, for all businesses and without any transaction limit.
• Through December 31, 2027, with prior-year revenue not exceeding 800,000 euros – The exemption is extended for exhibitors whose total revenue in the previous calendar year did not exceed this limit.
• Through December 31, 2027, at EDI – The exception also applies to submissions via electronic data interchange (EDI), regardless of revenue.

Implications for the exam

Little will change in terms of verification. XRechnung is a pure XML format. ZUGFeRD is a hybrid format: a PDF file with embedded XML that is readable by both humans and machines. The XML contains a varying number of fields depending on the profile; according to the Explanatory Notes from the Federal Ministry of Finance Starting with version 2.0.1, XRechnung and ZUGFeRD comply with the EN 16931 standard—with the exception of the ZUGFeRD profiles MINIMUM and BASIC-WL.

The fields are pre-structured, so no recognition is required. Validation remains: The software checks whether the XML conforms to the schema and the standard’s business rules—including the totals. The formats also include a separate field for the order number—the most common reason for unprocessed invoices. Whether the service was delivered is not indicated there either; a factual review remains necessary.

In addition, the incoming mail remains a mix. This affects transactions between domestic businesses; foreign invoices and tax-exempt transactions under Section 4, items 8 through 29 of the Value-Added Tax Act (UStG) continue to be received as PDFs or on paper. Anyone who Digitization of Paper Invoices ...because everything will be organized from now on, this requirement is too broad.

Conclusion

The formal and computational checks are performed automatically. The level of automation depends on the substantive check—and that, in turn, depends on order data and master data, not on the quality of the recognition. As the final step in this process, digital invoice approval is only as fast as the preceding check.

An automation project therefore begins by measuring the dark-processing rate and identifying why the remaining invoices get stuck. These causes recur and can be identified before a software solution is selected.

Would you like to know what percentage of your incoming invoices can be processed without manual intervention? Please feel free to contact us. Our experts will provide you with a no-obligation initial assessment.