Demo: AI General Ledger Coding in ERP AP Automation

• 0:06 — Introduction and agenda — Greg Zinda introduces himself as director of product management at Pairsoft, noting about eight years with the company and two prior years as a PaperSave client. He outlines the agenda: current AI GL coding capabilities, a demo, what’s coming next, and Q&A.
• 1:17 — How AI GL coding works — The feature draws on the history of items already processed through PaperSave invoice automation workflows to suggest GL codes. Existing ERP vendor defaults can still be respected, but the model also considers other fields.
• 2:13 — Fields used for suggestions — Beyond vendor, the model looks at fields such as invoice amount, description, and configuration-specific fields like entity in multi-entity setups or department. It combines these to suggest the top three codes, and multiple codes can be selected.
• 3:07 — Q&A instructions — Attendees are asked to submit questions through the Zoom Q&A panel, with answers to be provided there and some addressed live at the end.
• 3:33 — Demo: invoice and AI icon — An invoice already added to PaperSave is shown with OCR-extracted fields including vendor and invoice number, and the invoice date set to today. The familiar distribution field is unchanged apart from a new AI icon that opens a dropdown of suggested account numbers with optional extra columns such as description or class.
• 4:50 — Selecting accounts and limitations — Selecting a suggested account automatically populates the distribution field. Multiple accounts can be selected, but the system cannot yet determine how to break distributions out across them, so that must be done manually; single-account invoices are the quickest case.
• 5:38 — Configuring fields and workflow placement — The demo environment uses vendor, invoice amount, and description as inputs, and configurations can be changed to use other fields. A second example shows selecting a code and submitting the item; the step can occur anywhere in the workflow as long as the input fields are populated before the AI dropdown is clicked.
• 7:05 — Deployment process — The feature is currently being deployed to a handful of customers. Deployment requires little customer involvement — Pairsoft reviews customer data to determine useful fields, though customer insight on relevant fields is helpful — after which the icon is deployed to the environment.
• 8:01 — Planned future use cases — Additional AI use cases under consideration include suggesting which approvers should review an invoice and PO line-level matching of invoice lines to PO lines in the ERP. Customer feedback is invited to help set priorities, and the capability is described as something that will be continuously expanded.
• 9:20 — Q&A: GP defaults and account details — For GP integrations with default GL codes on the vendor card, the default that comes in can be removed and replaced by the AI selection. The dropdown currently shows account numbers, but account description and other columns can be added.
• 10:56 — Q&A: Projects and intercompany — Applying suggestions to other values such as project fields — a column available for Blackbaud customers, with other columns for GP clients — is described as a possible future use case. With intercompany turned on, the destination company column in the distribution table would also be suggested, though the scenario has not come up often yet.
• 12:25 — Q&A: Pricing, smart forms, and version requirements — The feature is sold separately from the core product with additional subscription costs, and a free trial period is offered; interested customers should contact their account manager. It is not currently in smart forms, that is being investigated, and it is available only in PaperSave Cloud version 7 — not version 6 Service Pack 11 in the cloud.
• 14:17 — Q&A: Users, permissions, and locations — Taking the submitting user into account as a signal is possible but not currently implemented; the feature follows existing permissions for editing the distribution table rather than restricting access to specific users. For a customer with 300 locations where location is part of the GL segment, a location field would need to be available so recommendations could be based on the selected location.
• 15:59 — Q&A: Many-account splits and invoice reading — Invoices split across more than three GL accounts are not a good fit yet, since only three accounts are suggested; expanding to suggest multiple accounts and their breakout based on history is being considered. Cases like a cell phone bill split according to the bill’s contents cannot be handled because the system does not read the invoice image, only historical coding patterns.
• 17:06 — Q&A: Availability, ERPs, and closing — The feature can be deployed now and is actively being rolled out to customers. Oracle Fusion is not supported because a PaperSave integration for it does not exist yet, while a NetSuite integration is described as coming soon. Two distributions per invoice are possible but amounts must be entered manually, invoice reading is being explored for PO line-level matching, and unanswered questions will be followed up individually.

What does the AI GL coding feature actually do?

It suggests general ledger account codes for an invoice being processed in PaperSave. It looks at the history of items a customer has already put through PaperSave invoice automation workflows and, based on that history, suggests the top three GL codes it believes the invoice should be coded to. The user clicks an AI icon next to the distribution field, picks a code from a dropdown, and it populates the distribution field automatically.

Which fields does the model use to make its suggestions?

In the environment shown, the model used the vendor, the invoice amount, and the description. Configurations can use other fields that are specific to a customer’s setup, such as entity in a multi-entity ERP scenario, a department field, or a location field where location is part of the GL segment. Pairsoft reviews a customer’s data during deployment to determine which fields are useful, and customer input on that is welcomed.

Does the AI read the invoice image itself?

No. The suggestions are based only on what has historically happened with a customer’s invoices, not on the content of the invoice image. This means cases like a cell phone bill that is split according to the line items printed on the bill cannot be handled. Reading the invoice is being investigated, primarily because PO line-level matching would not be possible without it.

Can it split an invoice across multiple GL accounts?

Multiple accounts can be selected from the suggestion dropdown, but the system does not yet calculate how to break the amounts out across them — that must be done manually by editing the distribution rows. Invoices that are coded to more than three accounts are not a good fit currently, since only up to three accounts are suggested. Suggesting multiple accounts along with the breakout based on history is being considered.

Is this included in the core product, and what does it cost?

It is not part of the core product and is sold separately, with additional subscription costs. Additional costs were described as applying to the subscription rather than to implementation and setup. It is not deployed to everyone automatically — customers have to sign up — and a free trial period is offered for those who want to try it before buying. Interested customers are directed to contact their account manager.

What are the version and ERP requirements?

The functionality is only available in PaperSave Cloud version 7. Customers still on version 6 Service Pack 11 in the cloud cannot use it. Oracle Fusion is not supported because a PaperSave integration for it does not yet exist, though it is being looked into. A NetSuite integration was described as coming out very soon and as the next ERP that will work with PaperSave. GP and Blackbaud integrations were both referenced during the session.

How does it interact with existing ERP vendor defaults and user permissions?

Existing ERP defaults set up by vendor can still be respected. For GP integrations where default GL codes come from the vendor card, the default that comes in can be removed and replaced with the code selected from the AI suggestions, so defaults can still be used where they make sense. Access is not restricted to specific users; the feature follows the same permissions that govern who can edit the distribution table, so a user without rights to edit distributions cannot use it.

Where in a workflow does the AI coding step happen, and what is required to deploy it?

It can be placed at any point in a workflow configuration — the beginning, the middle, or wherever invoices are normally coded. The only requirement is that the fields feeding the recommendation are filled in by the time the user clicks the AI dropdown. Deployment requires very little customer involvement: Pairsoft reviews the customer’s data to determine which fields are helpful and then deploys the icon into the customer’s environment.