Remittance processing

When your customer sends a remittance advice, Zuora uses a dedicated Remittance AI pipeline to extract invoice-level details and automatically map them to the correct payment.

Remittance matching process

1. Identify remittance information

   • Zuora classifies inbound emails or uploads to detect if they contain remittance information.

   • It supports remittance received via:

     1. Email body (plain text or HTML)

     2. Attachments (PDF, Excel, CSV, TXT)

     3. Scanned lockbox files (TIFF / image-based PDFs)

2. Extract content

   • Digital Attachments: Parsed using structured data extraction.

   • PDFs / Text: Read using pattern recognition and LLM-based table detection.

   • Scanned Images: Processed through OCR to extract visible text (invoice numbers, amounts, payer details).

   • The model attempts to locate invoice identifiers (INV-12345, #1045, etc.) and amounts from the content.

3. Map invoices

   • Once invoice numbers are extracted, Zuora looks them up in your open invoice dataset.

   • Matches are established when:

     1. The invoice exists and belongs to the same customer.

     2. The total payment amount equals (or nearly equals) the sum of the listed invoices.

4. Link the remittance to a payment

   • The system compares payment amount, date, and reference ID with data from your bank statement.

   • If only one payment fits those criteria, the remittance is auto-linked.

   • If multiple potential payments are found, the top three are suggested with confidence scores for user review.

Conditions for reliable remittance matching

Remittance AI performs best when:

1. The remittance lists clearly formatted invoice numbers (e.g., INV-1234, INV-1235).

2. Invoice matching works best, if invoice number in lockbox is same as the one on.Let's say invoice number "INV1234". If remittance contains just "1234". This would not work.

3. Totals match or differ only by known adjustments (TDS, bank charges).

4. The customer's name or domain matches system data.

5. The PDF or Excel is machine-readable (not a low-quality scan).

Custom remittance interpretations

With additional configuration, Remittance AI can support the following interpretations. Alternative invoice formats require explicit instructions for accurate extraction.

1. Alternative interpretation of invoices require explicit instructions to AI to be extracted well.

   1. Interpret non-standard abbreviations like "1234-37" → meaning INV-1234 to INV-1237

   2. Expand compressed patterns like "INV-123, 24, 25"

2. Read and match alternative invoice identifiers such as PO number, order number

Remittance matching limitations

Remittance AI cannot reliably:

• Read handwritten or blurred scanned images.

• Detect invoice numbers printed as watermarks or stylized graphics.

• Link a remittance when invoice numbers do not exist in enterprise resource planning (ERP) data or were recently created but are not synchronized.

• Identify invoices when multiple payments share identical total amounts and dates. These payments require manual linking.

• For invoice line items case, AI will not work reliably.

• All invoices in the remittance should have valid balance due amount. Invoice amount should sum upto the payment amount.

All invoices in the remittance must have valid balance-due amounts. The invoice amounts must sum to the payment amount.

If Remittance AI cannot read a remittance, the payment is still processed. However, the payment remains Unmatched or Partially Matched until a user or an AI correction updates the match.

Smart Match AI without remittance

When no remittance is available, Zuora's Smart Match AI relies entirely on bank data and historical behavior to identify probable invoices. The AI analyses the payment amount, currency, and past payment patterns for the same customer.

Smart Match AI processing

1. Amount-Based Prediction

   • AI searches for open invoices whose total matches or closely approximates the payment amount.

     1. If there are a combination of invoices whose sum total comes to the Payment Amount, the system would be able to predict it confidently.

   • If the difference matches common adjustment patterns (e.g., 1% TDS, bank charges), it adjusts the prediction accordingly.

     1. A variance of upto 10.00 is accomdated in the Amount descrepency.

2. Reference-Based Inference

   • When available, AI parses text from the payment description in bank statements (e.g., "OBI=/RFB/INV-1032").

   • Recognizable invoices are matched directly.

3. Confidence Scoring & Review

   • Each predicted match is assigned a confidence score.

   • Matches above a set threshold are auto-applied.

   • Low-confidence results are flagged for manual confirmation by a CashApp agent.

Smart Match AI limitations

Smart Match AI cannot:

• Distinguish between multiple invoices of the exact same amount for the same customer (manual review required).

• Resolve matches if adjustments and deductions distort amounts beyond the tolerance range.

• Infer invoices if bank description fields contain unrelated free text.

• Guarantee full accuracy when payments cover multiple customers without remittance context.

• Without Remittance, cannot match over 10 invoices.

When AI cannot make a confident prediction, the payment remains "NOT MATCHED." Agents can manually select invoices — and Zuora learns from those corrections to improve future results.