Overview

How are predictions performed?

The first step in making the best predictions possible is to learn from the data. When you capture a Vendor invoice, you will review the captured results, make any necessary changes, and then approve it. The next time you capture a Vendor invoice for the same Vendor, we will look at all previously approved invoice captures, identify the closest match, and return the predicted fields.

What data is the AI being trained on? 

AI is being trained on all previously approved invoice captures for the same Vendor.

What is the difference between captured and predicted fields?

Captured fields are fields that are captured from the document. Some examples of captured fields include the Issue Date, Due Date, and Invoice Number. Predicted fields are fields that most likely do not exist on the Vendor invoice, such as GL Expense Account and GL Variables. The Ledger and Vendor fields are captured first, and then predicted. Below is a list of fields with the related classifications:

• Captured
  • Issue Date
  • Due Date
  • Discount Amount
  • Payee Reference
  • Currency
  • Invoice Comment
  • Quantity
  • Unit Cost
  • Date
• Both Captured and Predicted
  • Ledger
  • Vendor
• Predicted
• Expense GL Account
  • GL Variable 1
  • GL Variable 2
  • GL Variable 3
  • GL Variable 4
  • Project
  • Project Task
  • Product
  • Payable

Does AI Invoice Capture only work with Email to Payable?

No. AI Invoice Capture does not need to be used with Email to Payable.

If you do not want to utilize the Email to Payable feature, you can create a Payable record, and upload the document(s) to the Payable record. If there are multiple documents uploaded, click the Set File As Default button for the file that should be captured. Then, click the Invoice Capture Review button in the Related Documents carousel to edit and/or review the captured fields.

Is the information used to train the model protected from being shared outside our Salesforce organization? 

Yes. Each company has a separate model, so each company only has access to their own data. Customer information and data is not shared outside their own Salesforce organization.

Where in Salesforce does the model look for learning or training purposes?

The model looks to the Payable for learning and training purposes.

What file formats are supported?

The following file types can be included as a part of the AI Invoice Capture process:

• • PNG
  • JPG
  • JPEG
  • PDF
  • Excel (.xls/.xlsx) [Should be formatted similarly to a standard invoice format]
  • Word (.doc/.docx)

When sending the batch for approval, may we send it to two people for review? Will the invoice scans be attached for their reference?

Yes, 2 or more people can be named in an approval process. You cannot directly attach a file to an approval process itself, but you can attach files to the record that is being submitted for approval, such as the Payable record. These files are then accessible to approvers as part of their review process.

How does the model learn to populate GL Accounting Variables (GLAVs)?

The model learns and predicts based on historical data from the Invoice Comment value. The model will look for posted Payables for that particular vendor, and suggest the GLAV values based on the GLAVs from the closest matching Invoice Comment value.

Can the model read multiple currencies?

Yes. The model captures the currency present in the scanned document. It will populate the CurrencyIsoCode if the org is configured for multi-currency. The currency must be included in the allowed list of currencies for the org.