WhatsApp Payment-Receipt Verification

“I highly recommend him. He's very attentive and dedicated to his work, very responsible, and always ready to help. I give him 100% for his work. 100% recommended.”

Delivered for a digital-products retailer (client not named). Customers send payment-receipt screenshots on WhatsApp; the workflow extracts the details, checks them against an order ledger and replies.

Staff were checking every receipt image by hand, matching reference numbers to records. It was slow and error-prone at peak times.

A person opened each screenshot, compared amounts and references, then approved or rejected the order.

1. Channel

   WhatsApp Business number connected through Evolution API.

2. Extraction

   OCR reads the amount and reference from the receipt image.

3. Verification

   Details are cross-checked against a Google Sheets order ledger; duplicate-submission logic catches repeats.

4. Routing

   Standard cases get an automatic reply. Ambiguous cases are sent to staff with context.

5. Staff panel

   A chat inbox with a human-mode switch so staff can take over a conversation and send messages manually.

1. A customer sends a receipt image.
2. The workflow extracts the amount and reference.
3. It checks the ledger and looks for a duplicate of the same transaction.
4. Clear matches are confirmed or rejected automatically.
5. Anything unclear is routed to staff with the full context.

• n8n (69 nodes)
• Evolution API
• OCR
• Google Sheets API
• PostgreSQL
• Next.js
• Docker / Traefik

• Edge cases go to a person rather than being forced into an automatic decision.
• Duplicate detection runs before any confirmation is sent.
• Documentation and handover were delivered in the client's working language (Spanish).

• Duplicate-submission checks.
• Human-mode gate so staff can pause automation on any conversation.
• Conversation logging in PostgreSQL for review.

• Delivered with a written handover guide and workflow documentation.

Outcome: Delivered, handed over and accepted by the client. No time-saved or accuracy figures are published.

Limitations: Receipt formats vary, so OCR results can be uncertain; that is why uncertain cases go to staff.