Before
Previously, a recipient likely had to identify invoice-related emails, inspect and extract their details, search supplier records, and compare the company and invoice date manually. When no match was found, the recipient also had to notify others, wait for a decision, delete confirmed fraudulent messages, and communicate the outcome through separate updates.
How it works
- Screen and sanitize each incoming email, then identify invoice-related subjects and extract the relevant invoice fields.
- Search supplier invoice records by amount and compare returned records against the extracted supplier name and invoice date.
- Route unmatched invoice emails for approval or denial; record approvals or automatically delete denied messages and notify security.
Workflow details
- Email intake and privacy protection: The workflow starts when an email is received and extracts the sender, subject, date, headers, and recipient details. Before analysis, it removes spaces from the message content and masks banking identifiers and phone numbers to reduce exposure of personal or financial information.
- Invoice relevance screening: The subject is split into individual words and checked for invoice-related terms such as "facture" or "invoice." Messages that do not appear to concern invoices are stopped with a result indicating that the email is likely unrelated to invoicing.
- Automated invoice extraction: For relevant messages, the workflow extracts the issuing company, total amount, pre-tax amount when available, currency, invoice number, invoice date, and payment deadline from the sanitized content. It then searches supplier invoice records using the extracted amount.
- Record validation and human review: Returned invoices are evaluated one by one by comparing the supplier name and invoice date with the extracted email details. If a matching invoice is found, the workflow records that result; if no match is found, it sends the recipient a structured review message containing the sender, date, company, amount, currency, and subject, with approval and denial choices.
- Decision-based remediation: An approval updates the review message and records who approved the email. A denial updates the incident as a confirmed fraud attempt, moves the email to the trash, and posts a follow-up notification stating that security was notified and the email was deleted.
What changed
Each relevant email now receives consistent privacy masking, invoice extraction, record lookup, and supplier/date validation before human involvement. Only unmatched invoice emails require an approval or denial decision, after which the workflow updates the status and carries out the corresponding notification or deletion steps.
Work removed
Manually reviewing invoice-like subjects, masking banking and phone information, extracting invoice fields, searching supplier records by amount, comparing supplier names and dates, sending review notifications, moving denied emails to trash, and posting remediation updates.
Expected outcome
Conservatively, the workflow can remove about four recurring manual actions per flagged email, including field extraction, record lookup, cross-checking, and follow-up or remediation, while keeping the final approval decision with the recipient.
