Introducing Machine Learning in Meter Reading and Billing

Objective

After completing this lesson, you will be able to explain the concept and usage of machine learning in SAP S/4HANA Utilities.

Machine Learning in SAP S/4HANA Utilities

This unit explains how Machine Learning is used in SAP S/4HANA Utilities. It is important for people in any market role and in any country.

SAP S/4HANA Utilities takes advantage of new ideas from the SAP Intelligent Enterprise when they are useful. Machine Learning helps solve problems with meter reading and billing to make processes more automated.

Diagram of SAP Utilities Core structure featuring SAP Intelligent Enterprise and SAP S/4HANA. The top section labeled SAP Intelligent Enterprise includes components: Fiori, Machine Learning, SEC, SAC, MCM, and MaCo. Below, the SAP S/4HANA section contains SAP S/4HANA Utilities with two sub-sections: M2C and U4G. M2C includes modules like BF, DM, EDM, with features such as Meter Reading, Machine Learning, and Billing Invoicing. U4G contains modules such as MDU, IMG, MaCo, BE, EEG, and MDM, illustrating integration in the SAP Utilities Core

Machine Learning as Part of SAP Intelligent Technologies

SAP uses intelligent technologies to automate and optimize business processes to improve the customer experience.

Intelligent technologies provide features like Machine Learning (ML), Internet-of-Things, and Realtime Analytics. These intelligent technologies are incorporated into applications through the SAP Business Technology Platform.

SAP S/4HANA is enriched by intelligent solutions such as SAP SuccessFactors, SAP Fieldglass, SAP Concur and SAP Ariba.

SAP S/4HANA Utilities is enriched by ML to use previous decisions from utilities metering expert agents and utilities billing expert agents to resolve implausible meter reading results and outsorted billing documents either automatically or by making a solution proposal.

Diagram showing relationships between Business Processes and Customer Experience. Three main sections are presented: Capabilities, Integration, and Application. Capabilities include Machine Learning with sub-topics: Internet-of-Things and Analytics. Integration highlights SAP Business Technology Platform. Application focuses on SAP S/4HANA Utilities with Machine Learning, listing SAP SuccessFactors, SAP Fieldglass, SAP Concur, and SAP Ariba. Arrows connect Business Processes and Customer Experience, symbolizing interaction.

Architecture of Machine Learning

Machine Learning scenarios can be categorized as Embedded or Side-by-Side.

Embedded Machine Learning algorithms do not heavily use memory or CPU power. In SAP S/4HANA Utilities, these algorithms help solve problems related to meter reading and billing. Embedded Machine Learning is built into SAP S/4HANA, using HANA ML from the infrastructure provider and application data for training models.

Side-by-Side Machine Learning algorithms need a large amount of data and significant CPU time for training models. To prevent heavy loads on the transactional SAP S/4HANA system, these scenarios are managed on SAP BTP. The trained model data is accessed through remote interfaces, especially for advanced use cases like image recognition, sentiment analysis, or deep learning for natural language processing, which rely on neural networks.

ISLM stands for Intelligent Scenario Lifecycle Management. It facilitates the integration and use of predictive functionality. Acting as a framework, ISLM handles the lifecycle management operations for Machine Learning scenarios. It can train an intelligent scenario and use the trained model to provide accurate inference results.

SAP Joule allows users to interact with SAP S/4HANA through a conversational interface that understands natural language. It uses Machine Learning to gain insights from historical data and past experiences. Additionally, SAP Fiori is incorporated to enhance the user interface with extra visual elements, such as confidence intervals or forecasting charts

Diagram illustrating the integration of machine learning with SAP systems. On the left, SAP S/4HANA features Embedded ML with HANA ML and Application Data, connected through ISLM. On the right, SAP Business Technology Platform includes Side-by-Side ML, AI Services, Training Inference Serving, Data Science Tools, Deep Learning CPU, Monitoring Operating, and Data Storage. The connections between SAP S/4HANA and the platform indicate AI Consumption and Data Model Training Integration. Top section includes SAP Fiori and SAP Joule, bridging the two main systems

Machine Learning Scenarios

SAP S/4HANA Utilities needs a higher level of automation to handle exceptions during the meter reading and billing processes. To achieve this, it uses Embedded Machine Learning to process and resolve unusual reading results and billing documents that have been flagged as needing further review.

The benefits of the usage of ML are …

a higher cash flow by exception resolution automation
a higher operational efficiency by manual work support

In addition to the scenarios described in this unit, SAP S/4HANA Utilities provides further ML scenarios such as …

intelligent collections by customer behaviour
matching incoming payments to open items
need-based product recommendation
field service planning optimization
asset maintenance optimization
response generation for agents

Embedded Machine Learning

A utilities company determines the meter reading results during the meter reading process, and these results are then used during the billing process. However, the meter reading results might sometimes be implausible, and the billing documents could be flagged for further review.

Flowchart showing components related to Meter Reading and Billing. Listed items: Intelligent Scenario Lifecycle Management, Core Data Services, Release Confidence, Release Suggestion, and Release Automation. These elements are aligned vertically and pertain to processes within Meter Reading and Billing. The components suggest a focus on data management and automation in billing cycles.

SAP S/4HANA Utilities uses Embedded ML, which covers …

the usage of ISLM to enable predictive scenarios with training- and apply datasets to train and activate customer-specific predictive models
the usage of CDS to access ML relevant data for implausible readings and outsorted billings (see next figure)
the determination of the release confidence value for implausible readings and outsorted billings
the release suggestion or the automatic release of implausible readings and outsorted billings

CDS Views of Machine Learning

The following consumption views enable the usage of ML-relevant data for implausible meter reading results and outsorted billing documents. These are training datasets and apply datasets to determine their release confidence value

Flowchart displaying machine learning datasets for Meter Reading and Billing. Meter Reading includes an ML Training Dataset for Predicting Automatic Release and Implausible Meter Reading Results. Another dataset focuses solely on Implausible Meter Reading Results. Billing features an ML Training Dataset and an ML Apply Dataset for Outsorted Billing Documents. The structure indicates the application of machine learning to improve data accuracy and billing processes.

The ML Training Dataset for Predicting Automatic clearing of improbable Meter Reading Results processes the data to form a forecast about the chances of automatically releasing unsupported meter reading results.

ML Training Dataset for Predicting the Automatic Release of Implausible Meter Reading Results: The ML Training Dataset for Predicting the Automatic Release of Implausible Meter Reading Results prepares the data for the creation of a prediction about the option of releasing implausible meter reading results automatically.
ML Training Dataset for Implausible Meter Reading Results: The ML Training Dataset for Implausible Meter Reading Results prepares the data for the ML training models in the implausible meter reading results environment.

ML Training Dataset for Outsorted Billing Documents: The ML Training Dataset for Outsorted Billing Documents prepares the data for the ML training models in the outsorted billing documents environment.
ML Apply Dataset for Outsorted Billing Documents: The ML Apply Dataset for Outsorted Billing Documents provides ML-relevant data for outsorted billing documents to access the apply dataset.

The characteristics to build the basis for training datasets and apply datasets could be …

segments such as portion or billing class,
amounts such as billing amount or previous billing amount,
statistics such as billing amount average over 2 years.

The following video shows the ML processing for implausible meter reading results in SAP S/4HANA Utilities.

The following video shows the ML processing for outsorted billing documents in SAP S/4HANA Utilities.

Use Machine Learning in Meter Reading and Billing

1. Storyline

1.1 Story

Embark on a journey with SAP and immerse yourself in our in-depth, persona based, benefits driven offline demo. This engaging demo showcases the SAP portfolio, fulfilling your business needs. Our Machine Learning Scenarios and Fiori Applications cater to essential process steps, from meter reading to order creation, ensuring seamless transaction. Explore the Meter Reading application that integrates machine learning to generate release confidences and swiftly resolves implausible readings. Discover the Billing application where machine learning automates the release of outsorted items, thus increasing efficiency. Through this experience, discover how SAP solutions can help you transform your business.

1.2 Acting Persona

Sarah Miller, customer service representative

Note the following needs:

A thorough understanding of the company's products or services to effectively respond to customer questions and concerns
Strong problem-solving abilities to manage complaints and resolve customer issues efficiently
Patience and empathetic listening skills

Note the following challenges:

Handling certain customer complaints
Staying informed about the continuously evolving product line or services provided by the company
Maintaining a balance between delivering high-quality service and working efficiently

1.3 Value Drivers

Reduce meter reading and data administration cost
Reduce meter reading and data administration cost by streamlining and automating meter related collection and processing functions
Reduce inaccurate utility bills
Reduce inaccurate utility bills by applying sophisticated quality checks and algorithms to handle implausible meter readings

2.2. Key Take Aways

The demo highlights Fiori Applications optimizing essential business processes. The Meter Reading app uses machine learning to efficiently manage missing orders and implausible readings. Billing and invoicing are streamlined with automated creation and printing of documents, reducing manual intervention and enhancing data integrity.

3. Master Data

1. Master Data Creation

Screen for creating master data

Create one complete set of business data by following the below steps:

Navigate to the Master Data Utilities page.
Launch the app, Create Master Data Res (Z_DATA_ML), and populate the below fields:
- Select Country as US.
- Select the checkbox for Electricity Contract.
- Enter Move/in_Date as Jan 1st of 2 years earlier. (As machine learning will require meter reading and billing data for the last 2 years, the move-in date should be updated accordingly.)
- Enter data in the First Name field.
- Enter data in the Last Name field.
- Select the MRU_ML checkboxes.
- Choose Enter and click on Execute.

3.2. Master Data and Additional Information

1. Create Billing History

Interface showing dates for billing, invoicing, and payment with a contract account number

Use tool ZMETERCYCLE to create the billing history. Use billing dates 01/01/2023 to the 5th of one month prior to the current month.

In this case, the move-in date is 01.01.2023. Scheduled MR date is 31.01.2025. The Billing date is the Scheduled MR date plus 5 days for generic usage, which is 02/05/2025. This will create billing history until Jan 31st.

Steps

4.1. Metering Process
This section focuses on the below process steps by using Fiori Applications:
The Meter Reading Fiori application will provide a list of missing orders, allowing the creation and update of meter reading results. In case those are implausible, machine learning will help to release them.
4.1.1 Metering Process
Note the following benefits:
- Increased efficiency and speed
- Real-time monitoring and reporting
- Improved accuracy
Persona: Sarah Miller, customer service representative
4.1.1 Create Meter Reading Order
The Customer Service Representative initiates meter reading by requesting (order) for the specific scheduled date to collect the meter read. CSR will review the missing meter reading orders and able to create the meter reading.
1. To navigate to the Utilities Billing space, click on the Utilities Billing tab.
  Create Meter Read Order:
  Use the Periodic Billing Overview Fiori Application to review the status of the billing for a specific billing cycle (i.e. Portion). From the application, CSR will drill from the charts and tables, to details of the billing. The list of installations/contracts where meter billing orders are missing are displayed using Missing Billing Order on the tile, Unbilled Contracts. It calls the Process Missing Billing Orders Fiori application, where the meter reading order and billing order could be created.
  As an alternative, the Process Missing Billing Orders Fiori application could be called from the Fiori launch pad.
2. Click on the Periodic Billing Overview tile to access the Periodic Billing Overview app.
3. Click on the icon (F4) to search for portion and scheduled billing date.
4. Search Portion.
  1. Enter Portion as P-ML.
  2. Click on Go.
5. Select Portion:
  Select portion P-ML. Scheduled billing date is the one which is the next scheduled billing date after the last meter reading was recorded in billing history. For example, in this case, select the last date of the previous month.
6. Fetch:
  Click on Go.
7. Overview:
  1. The Periodic Billing Overview app will provide figures and graphics for the selected portion and scheduled billing date.
  2. Click on a row listed by the Missing Billing Orders tile, which covers the meter reading unit MRU-ML of installation.
8. Click on the icon to select the Portion from the list.
9. The Select Portion popup opens. Select the portion P-ML record.
10. Click on Adapt Filters.
  In case too many records are displayed and cannot find your data, you include additional search criteria. To do so, use the button, Adapt Filters on the top right. If it is not visible, extend the search criteria section by using the down arrow. On the Adapt Filters popup, click on the search criteria you need, for example, Business Partner and use the OK button. Then provide the filter value, for example, the number of business partner and chooseGo.
11. Add Business Partner in Filter.
  1. On the Adapt Filters popup, select the check box for Business Partner.
  2. Click on OK.
12. Fetch data: Click on Go.
13. Create Order:
  1. Missing billing orders are listed based on the selection criteria from the Periodic Billing Overview app. Additional selection criteria is provided to narrow list. As an alternative, the Process Missing Billing Orders app could be called from the Fiori Launchpad.
  2. Click for the row which belongs to contract/installation of the billing order to be created.
  3. Click on Create Order.
14. The screen displays a confirmation message for the billing order creation. Click Exit to close the popup.
15. Refresh:
  As it is displayed on an additional page, close the page to continue from Missing Billing Orders.
  Click on Go.
16. Navigate to the home page:
  1. The missing billing order record is removed as the order is created successfully.
  2. Click on theSAP logo to navigate to the home page.
4.1.1Create Implausible Meter Reading Results
The customer service representative creates implausible meter reads.
1. To navigate to the Utilities Billing space from the home page, click on the Utilities Billing tab.
  Meter reading results can be updated. It is also possible to use estimation. Validation for meter reading results will be executed. In case of implausible meter readings, those could be released immediately or could be saved as implausible.
2. Navigate to Periodic Billing Overview:
  From the tiles, select the Periodic Billing Overview app.
3. Unbilled Contracts:
  1. Figures and graphics are updated. Click on Incomplete Meter Reading.
  2. The list provided by the Incomplete Meter Reading Results tile could also be used. Click on the row which is for the meter reading unit of installation.
4. Click on Incomplete Meter Reading.
5. Implausible Meter Reads:
  1. To be able to create the implausible meter reading and outsorted bill doc, enter a consumption which is above threshold. To make sure it will be outsorted during billing, enter a consumption as below on top of the Prev MR Result available on the screen. electricity: 60000 kWh gas: 60000 m3 water: 1000 m3 To find the upper limit, click MR and enter a meter reading result above Upper limit of MR.
  2. Press the Enter key.
6. Click Save.
7. An Information popup appears with a confirmation message:
  1. Click Exit.
  2. As it is displayed on an additional page, close the existing page to continue from Display Outstanding Billing Items.
8. Click the back arrow to navigate to the Periodic Billing Overview screen.
9. Click on the SAP logo to navigate to the home page.
4.1.1Release Confidence for Implausible Reads:
Th Customer Service Representative validates the release confidence determined automatically by the machine learning model for the implausible meter reads.
1. To navigate to Utilities Meter Reading, click on the Utilities Meter Reading tab.
2. Click on the Resolve Implausible Meter Readings tile to launch the app.
3. On theResolve Implausible Meter Readings screen, click on Adapt Filters.
4. Adapt Filters:
  1. On the Adapt Filters popup, select the Installation checkbox.
  2. Click OK.
5. Selection Screen:
  1. Enter Meter Reading Unit as MRU-ML.
  2. Click Go.
6. Validate Release Confidence:
  Machine Learning reports will be executed automatically in the background as soon as implausible reads are created to determine the release confidence.
  REML_FILL_EABLCONS HIST
  REML_MR_IMPLSBL_RELPRED
  The Release confidence column is updated.
  Note
  If the Release Confidence column is not showing, then please add it by clicking on the settings icon.
7. Edit the meter read document.
  If you prefer to release without going into detail, the line for the relevant meter reading result can be selected and can be released using the Release button on the top right.
  Click Edit.
8. Release and Save.
  1. The meter reading result can be manually changed or estimated.
    Note
    That could be released with high consumption to create the outsorted billing document.
  2. Click on the Release and Save button.
9. Confirmation:
  Click Exit.