Invoice Categorization Back Office

The billing team received a large number of vendor documents in a financial management system that had to be checked before any further accounting work could start. Each item could contain one or more attachments, and those attachments might be invoices, credit-card receipts or documents that should not be treated as invoices at all. Retrieving them was repetitive: staff logged in, located pending billing records and opened each document individually, which slowed the start of the billing process. Every attachment had to be checked for presence and type before processing, because an unsupported or missing file would otherwise break extraction. Invoice data such as vendor, invoice number, dates, total amount, tax, subtotal, payment terms and purchase-order number was read and typed in by hand, which is slow and vulnerable to typing errors. The business needed to distinguish a genuine invoice from a credit-card receipt and from a non-invoice document, because incorrect categorization sends a document into the wrong accounting process. For anything invoice-related the extracted vendor had to be checked against the FMS vendor list, and a missing or unmatched vendor needs a different action from a known vendor. Results and errors also had to be tracked somewhere the team could review them.

Five steps turn pending billing documents into reviewable records. The process collects billing records that are waiting for review and keeps the information needed to identify each document. Each attachment is then validated before processing: valid PDFs continue, while missing, unsupported or failed files are recorded for staff review rather than silently dropped. Valid content is extracted and combined with the original billing metadata, and the document is categorized as an invoice, a credit-card receipt or a non-invoice. When it is an invoice, the important billing details are collected as structured fields. The vendor name is compared against the existing FMS vendor list, with known vendors continuing normally and missing vendors marked for review. Successful results are saved to the main tracking sheet, and file and processing problems are written to a separate error sheet so the two concerns never mix.

Pending billing records are retrieved from the FMS after successful authentication, and each billing attachment is validated and processed separately rather than as an undifferentiated batch. Valid PDF content is extracted and combined with the original billing metadata, and AI classification assigns each document as an invoice, a credit-card receipt or a non-invoice. Structured invoice details including vendor, invoice number, dates, amounts, terms and purchase-order information are captured when available. The FMS vendor-search status is recorded together with the appropriate next action, so a known vendor and a missing vendor produce different outcomes. Invoice, receipt, non-invoice and missing-vendor records are appended to the Google Sheets classification table, and document, file, download and extraction failures are recorded in a separate Google Sheets error log. An internal processing-summary record is generated from the connected result branch. The practical output is the classification sheet and the error log, which replaces manual PDF reading, data entry, categorization and vendor searching.

Client retains ownership of the FMS credentials and vendor list, the Google Sheets classification table and error log, the AI classification rules, the document category definitions, and every workflow and authentication configuration.