Use Trial to Extract Information from Standard Documents with Generative AI and SAP Document AI

Before you upload our sample invoice file to SAP Document AI, you’ll create a corresponding schema.

1. Open the SAP Document AI basic UI, as described in the tutorial: Use Trial to Subscribe to the SAP Document AI Basic UI.

2. In the left navigation pane, click Schema Configuration.

   

3. To create your own schema, click Create.

   

4. In the dialog that opens, enter a name for your own schema – for example, invoice_schema. Note that the name can’t include blanks. Next, select Invoice as your Document Type.

5. Click Create to create the schema.

   

6. Your schema now appears in the list. Access the schema by clicking on it.

   

You can also adapt copies of the appropriate preconfigured SAP schemas to extract information from standard document types. However, we won’t cover that approach in this tutorial.

To add your first header field, click Add.

You must enter a field name and data type for each new field. The available data types are string, number, date, discount, currency, country/region, and list of values.

Default extractors are only available for standard documents. See the list of all default extractors in Extracted Header Fields and Extracted Line Items.

You can also optionally add a field label (user-friendly name) and a description.

A description is an optional entry that you add to provide an explanation or additional context for a field. Descriptions are particularly useful with fields that use generative AI to extract information. SAP Document AI uses your descriptions as prompts for the LLM. So, when wording a description, it’s helpful to imagine that you’re explaining what you want to extract to a person with no prior knowledge.

In this tutorial, we’ll use descriptions when creating each of the fields for which extraction is handled using generative AI.

As your first header field, add the total amount of the invoice.

1. Enter the name for your field – for example, total_amount. There’s no need to enter a description here.

2. Select number as the Data Type.

3. Use auto as the Setup Type and grossAmount as the Default Extractor.

   Note that your schema supports various combinations of setup types and extraction methods. When you use the setup type auto with a default extractor, the solution’s pretrained machine learning models are used to extract the information from the document. When you use the setup type auto without a default extractor, generative AI (LLMs) is used instead. The setup type manual supports extraction using a template. For more information, see Setup Types. For further details of this approach, see the tutorial mission Shape Machine Learning to Process Custom Business Documents.

4. Click Save.

The field now appears in your list of header fields, where you can see all the information that you’ve just entered. You can edit or delete the field by clicking the respective icons on the right.

Click Add again to open the Header Field dialog.

1. Enter the name for your second header field – for example, snNumber.

2. Enter the description source number

3. Select string as the Data Type.

4. Use auto as the Setup Type and leave Default Extractor empty.

5. Click Save.

Now, go ahead and add the remaining header fields and line item fields shown in the following tables and images. Pay attention to the different data types and whether you should add a description or a default extractor.

Header fields:

Line item fields:

Note that the SAP Document AI basic UI also includes a feature that allows you to group schema fields by category. To use this feature, you must first activate it under UI Settings. For simplicity’s sake, we haven’t included the feature in this tutorial. If you’d like to find out more about it, see Schema Field Categories.

Once you’ve added the fields, you need to activate the schema so that you can use it to extract information from documents. Right now, the schema has the status DRAFT, indicating that it can’t be used yet.

To activate the schema, click Activate.

Now, the status of your schema changes to ACTIVE. To make changes to your schema, you must first Deactivate it.

Congratulations, you’ve now created and activated your own schema for invoice documents.

1. Access Document from the navigation on the left of the screen, then click + to upload the invoice document.

   

2. On the Select Document screen, choose Invoice for the Document Type.

3. Select the Schema you created (invoice_schema).

4. Right-click on the link, then click Save link as to download the sample invoice document locally.

5. Drag and drop the file directly or click + to upload the sample document.

   

6. Click Confirm.

   The document status changes from PENDING to DONE.

   

7. Access the document by clicking on it. You now see the page preview of the document file you uploaded, and the information extracted from the invoice header fields and line items using LLMs, the solution’s pretrained machine learning models, and the schema that you created.

   

Note that the first time you extract fields using generative AI, the results are in the orange confidence range.

Congratulations, you’ve completed this tutorial.

You’ve now successfully extracted information from an invoice document using the schema configuration feature from SAP Document AI, LLMs, and the solution’s pretrained machine learning models.

Feel free to repeat the steps using your own standard documents.