Advanced Document Processing with SAP Enterprise Service Management

[Akshay_shripathi]

SAP Enterprise Service Management introduced new capabilities that harness Generative AI to help service agents process documents faster, reduce manual workload, and ensure more accurate handling of cases. Key features include extracting information from the subject, body, and attachments of a case to enable actions like backend document lookups. Additionally, these capabilities support converting attached documents into structured text, making it possible to post documents seamlessly to a backend system.

In this blog, I will walk you through the steps to set up the new capabilities in SAP Enterprise Service Management for advanced document processing.

The primary example I will use to walk through these steps is a supplier invoice inquiry process. In this scenario, the supplier contact emails to inquire about the status of their invoice. The service agent will then gather additional information about the invoice from the backend system, such as S/4 HANA, before responding to the supplier.

Step 1 – Defining Input Sources:

Each case will include a subject (mandatory), a description (optional), and attachments (optional). For a given process, first determine which of these elements will serve as the input sources for information extraction.

In the supplier invoice inquiry example, the invoice number could be mentioned in the subject or body of the email, or the invoice may be attached. Therefore, the input sources for this case would be the subject, description, and attachments. In contrast, for scenarios where documents such as payment advices or sales orders need to be posted to a backend system, these documents are typically provided as attachments, making attachments the sole input source.

Note: Cases can be created through various channels, including the self-service portal, email, or manual data entry from interactions like phone calls, chat, or in-person exchanges.

Step 2 - Configure a Scenario under Business Information Extraction:

Note: This step is applicable only if the subject or description of the case is one of the input sources for the scenario.

To extract information from the subject or description, create a scenario that defines relevant elements. These elements provide context to the language model, guiding it on what to search for in the incoming text.

For detailed steps, please refer to the associated help document.

For example, in the supplier invoice inquiry scenario, create an element called "Invoice ID" with the following details:

Next, create the scenario with the following description and include "Invoice ID" as an element:

Scenario Description: The supplier contact sends an email to inquire about the status of their invoice. Extract Invoice IDs from the email.

Tip: I recommend experimenting with different descriptions, short samples, and sample values. These variations primarily affect the prompts sent to the large language model and can significantly impact the accuracy and relevance of the output. Experimenting can help fine-tune the extraction process for better results.

Note: Business Information Extraction also supports document classification. To enable this, create classes and add them to the scenarios.

Step 3 - Configure a Schema under Document Extraction:

Note: This step applies only if attachments are one of the input sources for the scenario.

To extract information from attachments, create a schema for the new document. The contents of this schema enables the large language model to better understand the document and identify the fields that need to be extracted.

For detailed steps, refer to the associated help document.

For example, while Invoice is a standard schema, let’s consider extracting information from a handwritten check. Use the following description for the schema:

Extract key information from a scanned image of a handwritten check, including the date, payee name, and amount, while identifying the layout and structure of the document.

Additionally, define header fields and provide an appropriate description for each field as shown below:

Tip: As with step 2, I recommend experimenting with different variations for the schema and field descriptions. These variations primarily influence the prompts sent to the large language model, and as a result, they will impact the accuracy and relevance of the output. Trying out different descriptions can help fine-tune the extraction process for better results.

Step 4 - Create a Business Document Type:

Business Document Types represent backend documents in a semantic format on the case details page.

For your scenario, create a new Business Document Type and define columns to capture the required data, along with the corresponding input options.

For detailed steps, refer to the associated help document.

For example, in the supplier invoice inquiry scenario, define columns such as Invoice ID, Fiscal Year, Document Number, Document Status and Posting Date. These columns should encompass not only the fields extracted in Steps 2 and 3 but also additional fields populated after making API calls to the backend system.

The input options for these columns would include Manual Entry, Document Extraction and Business Information Extraction. Document Extraction and Business Information Extraction can be set-up to execute automatically.

Step 5 - Configure Actions for a Business Document Type:

Actions are configured as part of the business document type setup to facilitate communication with a backend system. These actions may include standard options like Post, Lookup, or custom actions such as Validate.

For detailed steps, refer to the associated help document.

For example, in the supplier invoice inquiry scenario, configure a Lookup action that uses the Invoice ID and supplier details to retrieve information from S/4 HANA, such as Fiscal Year, Document Number, Document Status and Posting Date.

Note: If Document Extraction is used as an input source for the business document type, all extracted header and line-item fields are stored as a JSON attachment to the case. When a Post action is configured, this JSON file is automatically sent to SAP Cloud Integration, which can then relay it to the backend system.

Step 6 – Add Business Document to a Case Type:

For the Business Document table to show up as a tab within the case details page, link the Business Document to a Case Type.

For detailed steps, refer to the associated help document.

For example, in the supplier invoice inquiry scenario, create relevant case type and link the newly created business document to it.

Try implementing these steps in your SAP Enterprise Service Management environment and unlock a new level of efficiency for document processing. Have questions? Let’s discuss in the comments!