AI Invoice Routing Agent

How Cassidy automates using AI

Step 1: Trigger on invoice arrival

The Workflow activates when an invoice hits your AP inbox, supplier portal, or upload folder—capturing PDFs, scanned images, and e-invoicing formats like cXML or PEPPOL.

Step 2: Extract and normalize invoice data

Cassidy's AI extracts header and line-level fields—vendor name, invoice number, amounts, PO references, tax details—and normalizes everything to a canonical schema with confidence scores for each field.

Step 3: Match vendor and detect duplicates

The Workflow cross-references your vendor master using fuzzy matching on legal names, tax IDs, and remit-to addresses, while flagging potential duplicates based on invoice number, amount, and date heuristics.

Step 4: Apply GL coding and distributions

Cassidy applies vendor defaults and keyword-based rules, then layers ML suggestions learned from historical coding patterns. High-confidence codes auto-apply; lower-confidence items route for human confirmation with suggested options.

Step 5: Execute PO and receipt matching

For PO-backed invoices, Cassidy performs 2-way or 3-way matching against purchase orders and receiving documents, applying your tolerance thresholds. Clean matches flow through; variances queue as exceptions with linked context.

Step 6: Route for approval

The Workflow dynamically routes invoices through your approval matrix based on amount thresholds, GL account, department, and vendor risk—supporting sequential or parallel approvals, escalations, and mobile sign-off.

Step 7: Post to ERP with full audit trail

Once approved, Cassidy creates the vendor bill or journal entry in your accounting system—QuickBooks, Xero, NetSuite, or Sage Intacct—with all coding, dimensions, and an immutable log of extraction values, rule evaluations, and approver actions.