PartialRamp
Requirement evaluated: For each of the 12,000 invoices processed monthly in Oracle NetSuite, the AP automation system must extract and present structured line-item data from every invoice line, not just header-level fields such as vendor, date, and amount. This is the prerequisite for any meaningful dimension-level coding: if the tool can only parse header data, all downstream coding attempts are limited to a single row per invoice regardless of how many line splits the organization requires.
For a buyer running 12,000 NetSuite invoices per month with dozens of coding fields, Ramp's Bill Pay OCR (Smart OCR, available via Ramp Plus) parses uploaded or forwarded invoice PDFs and extracts each invoice row as a discrete, structured record containing description, amount, quantity, unit price, line type (expense or inventory item), and tax rate rather than collapsing the invoice into a single header-level total. Smart OCR extracts and auto-fills bill details (invoice number, date, due date, description), amounts (bill total, invoice currency), and line items (description, amount, quantity, unit price, type, tax rate). …
Limitations: The auto-coding agent has a documented behavioral boundary: Ramp auto-codes any field your business uses for all transactions, but it does not auto-code fields like Customer or Project if they apply only to some expenses. …
PartialSage AP Automation
Requirement evaluated: The system's AI or rules-based coding model must improve coding suggestions over time using the buyer's own transaction history, so that recurring vendor-to-dimension patterns (e.g., a specific vendor always coded to a given project and department) are learned and applied without manual rekeying. Vendors must describe the learning mechanism, the measured accuracy lift curve, and whether the model is trained per-customer or shared across tenants.
This multi-entity SaaS company codes every invoice across location, department, project, class, and multiple custom dimensions. Sage AP Automation's ML engine operates as a hybrid: a shared cross-tenant base model trained on what Sage describes as 'millions of bills submitted to many companies' via the Sage Network, layered with a per-customer feedback loop where each correction an AP user makes to a draft bill is transmitted back to the Sage Network within 24 hours and used to update the model for that specific tenant. …
Limitations: The buyer's coding model explicitly includes class and several custom dimensions, and the official Sage Intacct AP Automation FAQ confirms that AI/ML coding covers only GL account, location, department, and project: class and all user-defined custom dimensions require manual entry on every draft, leaving the buyer's mo …