This update elevates the linking of invoice and PO line items to an entirely new level and transforms a previously time-consuming step into an exceptionally efficient process.
This article is applicable to finway users with access rights based on all role templates.
Until now, users had to manually compare and assign each line item of a purchase order to the corresponding line in an invoice. That manual effort ends today.
With our new AI Auto-Matching engine, finway now takes over this task completely.
This enhancement reduces manual review effort, minimizes errors and accelerates the reconciliation of invoices with purchase orders. The system intelligently interprets item names, descriptions, quantities and prices to create highly accurate assignments, even when document formats differ.
How the new AI linking works
As soon as a purchase order is approved and a corresponding invoice request is generated through the addition of an invoice, or when a purchase order is manually selected within the invoice request via “Edit” → “Select purchase order”, a loading window appears. This animation visualizes the AI process that is currently running.
During this step, finway demonstrates how the AI analyzes and processes the automatic linking of all relevant line items.
Once the AI has completed its work, the familiar item-linking window opens.
The key difference: all invoice line items are already automatically matched to the correct PO items. Manual linking is no longer necessary. If a match should not be correct, it can still be removed easily using the blue button in the center of the screen.
ℹ️ Please note: AI matching is not yet available for delivery notes.
When does automatic item matching occur and when not?
Our new AI-powered matching function compares invoice request line items with the corresponding line items of the purchase order. The system evaluates several criteria in a fixed sequence. A match is only created when all required criteria are met. Optional information may further influence the confidence score, but is not mandatory.
1. Amount comparison (mandatory)
The system first checks the percentage deviation between invoice amount and PO amount.
If the deviation exceeds 25 percent, no match will be created.
If it is within 25 percent, the system proceeds to analyze the item descriptions.
2. Item name comparison (mandatory)
A match is created when the item names are identical, when there is a clear substring relationship or when terms are semantically related.
For example, the term “Travel out & installation” contains “Travel”, and words like “Assembly” and “Installation” may be semantically equivalent.
A match will not be created when the terms refer to completely different products such as “Chair” compared to “Desk”.
3. Consideration of material codes (optional)
If material codes are present both on the invoice and the purchase order, they are included in the evaluation. Matching codes slightly increase the likelihood of a match, while differing codes slightly decrease it.
However, missing codes do not impact the mandatory checks. Thus, a match can still occur even without material codes, provided the amounts and names are consistent.
4. Determining the confidence score
Finally, the AI calculates a confidence score that represents the certainty of the match. The score is influenced by linguistic similarity of the names and the magnitude of the amount deviation.
The smaller the deviation and the stronger the language similarity, the higher the score.
Only results with a confidence score of at least 0.8 are accepted as valid matches. This ensures that all automatic matches are both accurate and reliable.
Example 1: A match is created
Invoice line item
Name: “Travel and installation”
Amount: 95 €
Material code: 1023-A
PO line item
Name: “Travel”
Amount: 100 €
Material code: 1023-A
Analysis according to matching rules
Amount check
– Deviation: 5 percent
– Below 25 percent → validName check
– “Travel and installation” contains “Travel” → clear substring
– Semantically related → validMaterial code
– Identical codes → slight increase in confidenceConfidence score
– High linguistic similarity and low deviation → Score ≥ 0.8
Result: → The AI creates a match.
Example 2: No match is created
Invoice line item
Name: “Office chair ergonomic”
Amount: 250 €
Material code: 8831-Z
PO line item
Name: “Desk wooden”
Amount: 240 €
Material code: 9901-K
Analysis according to matching rules
Amount check
– Deviation ~4 percent
– Below 25 percent → check continuesName check
– Chair vs desk → completely different product types
– No semantic similarity
– No substring
→ Mandatory rule not fulfilled → matching stopsMaterial codes
– Different codes → would further reduce the score
– Even identical codes would not override the failed name check
Result: → The AI does not create a match.