- SAP Managed Tags:
Introduction:
In this blog post, I will highlight the role of Machine learning in the current competitive environment and how ML is helping enterprises to respond to customers, competitors, regulators, and partners faster than their peers in transforming them as intelligent enterprises. According to Gartner (2020), 69% of routine work currently done by managers will be fully automated by 2024 and ML based innovations are the most critical for the enterprises want to transform the business.
How ML works in SAP S/4HANA?
Machine learning algorithms use customer-specific history and exceptions to predict future outcomes and these outcomes can be used to automate business user decisions.
In SAP S/4HANA, we have two kinds of ML capabilities: Embedded ML and Side by Side ML.
Embedded ML is used for simple ML scenario using classic algorithms like regression, clustering, classification, and time-series that requires low CPU & RAM and no external data is required. It is based on the SAP Analytics cloud.
While Side-by-Side ML is used for Complex ML scenarios using deep learning like image or language processing that requires high CPU & RAM and external data is required. It is based on the SAP Business Technology Platform.
Scenarios delivered via ISLM
The purpose of ISLM (Predictive Analytics Integrator - Pai 2.0) is to provide a common interface for the consumption of the ML models independent of the underlying predictive engine to provide predictions and results.
The simplified ISLM interface provides the feature of training the ML model (underlying algorithm) with the data that’s already available in the customer system and publishing the ML model for productive use within SAP Fiori applications.
The models get better with more data accumulated over time. The data is available to the model for training by mapping the CDS views and other data structures.
Here is the list of 24 scenarios delivered via ISLM:
Scenarios not delivered via ISLM
Several of the embedded Machine learning scenarios are delivered via ISLM. Some of the scenarios are not delivered via ISLM which means that those ML scenarios are trained internally by SAP and the customer has no control in terms of training.
Here is the list of 15 ML scenarios that are not yet available via ISLM. (not via Intelligent scenario management app)
How is it different from other intelligent technologies like RPA and Situation handling?
Machine learning is used in Predicting the future outcomes by utilizing the past data and help the business users in decision making. It provides Real-time, predictive insights so users can make faster and better decisions and adjustments.
While RPA is used for automating the repetitive tasks, here the focus is to identify recurring tasks and automate them using RPA bots. It accelerates the digital transformation of business processes by automatically replicating tedious actions that have no added value.
On the other hand like ML, Situation Handling also help in decision making but it is based on Decision flow charts, that are created by the users to take decisions for different kind of situations. Here future outcomes are not predicted to make decisions.
List of ML Scenarios and an aggregated view:
Achieving an end to end process automation using intelligent capabilities:
Some Key terms:
Some of the key terms related to ML model training to understand how effective an ML model is are explained below:
Predictive Power-
Description:
The predictive power of a model is the quality indicator of models generated using the application. This indicator corresponds to the proportion of information contained in the target variable that the explanatory variables can explain. To improve the predictive power of a model, new variables may be added to the training dataset. Explanatory variables may also be combined
Details:
A model with a predictive power of:
Link:
https://help.sap.com/doc/5e8fcd4e118542f8945ebdba4db4ad66/3.3/en-US/AA_userGuideandScenarios_EN.pdf
Predictive Confidence-
Description:
The prediction confidence is the robustness indicator of the models generated using the application. It indicates the capacity of the model to achieve the same performance when it is applied to a new data set exhibiting the same characteristics as the training dataset. To improve the prediction confidence of a model, additional observation rows may be added to the training
Details:
A model with prediction confidence:
Links:
https://help.sap.com/doc/5e8fcd4e118542f8945ebdba4db4ad66/3.3/en-US/AA_userGuideandScenarios_EN.pdf
Conclusion-
In this Blogpost, we have seen how SAP S/4HANA Machine learning models can be used to address typical customer scenarios right from detecting anomalies in financial transactions to predicting the future sales. Nowadays Machine learning based applications are helping all kinds of professionals right from Marketing experts, Financial experts, Demand planners to Procurement managers to take right business decisions. On top of that These machine learning models can be combined with other intelligent solutions like intelligent RPA and Situation handling to achieve end-to-end process automation.
Thanks for reading the Blogpost, I look forward to your comments. Alternatively, you can also post your queries on SAP S/4HANA Cloud Community.
Image Source in the Blogpost: SAP Frontrunner Team
In this blog post, I will highlight the role of Machine learning in the current competitive environment and how ML is helping enterprises to respond to customers, competitors, regulators, and partners faster than their peers in transforming them as intelligent enterprises. According to Gartner (2020), 69% of routine work currently done by managers will be fully automated by 2024 and ML based innovations are the most critical for the enterprises want to transform the business.
How ML works in SAP S/4HANA?
Machine learning algorithms use customer-specific history and exceptions to predict future outcomes and these outcomes can be used to automate business user decisions.
In SAP S/4HANA, we have two kinds of ML capabilities: Embedded ML and Side by Side ML.
Embedded ML is used for simple ML scenario using classic algorithms like regression, clustering, classification, and time-series that requires low CPU & RAM and no external data is required. It is based on the SAP Analytics cloud.
While Side-by-Side ML is used for Complex ML scenarios using deep learning like image or language processing that requires high CPU & RAM and external data is required. It is based on the SAP Business Technology Platform.
Scenarios delivered via ISLM
The purpose of ISLM (Predictive Analytics Integrator - Pai 2.0) is to provide a common interface for the consumption of the ML models independent of the underlying predictive engine to provide predictions and results.
The simplified ISLM interface provides the feature of training the ML model (underlying algorithm) with the data that’s already available in the customer system and publishing the ML model for productive use within SAP Fiori applications.
The models get better with more data accumulated over time. The data is available to the model for training by mapping the CDS views and other data structures.
Here is the list of 24 scenarios delivered via ISLM:
| Scenario | Intelligent scenario name | Line of Business |
| Predict Damage Code | PREDICTDAMAGECODE | Asset and Service Management |
| Predict Object Part | PREDICTMAINTOBJECTPART | Asset and Service Management |
| Late Payment Prediction - Behavioral Insights | BEI_LATE_PAYMENT | Public Services |
| Receivership Prediction - Behavioral Insights | BEI-RECEIVERSHIP | Public Services |
| Late Filing Prediction | BEI_LATE_FILING | Public Services |
| Sales - Sales Order Automation | SO_AUTO_EXTRACTION | Sales |
| Check Assigned Liquidity Items version1 | FCLM_RDT_CALI_V1 | Finance |
| Predictive Scenario for Bank Reconciliation | FI_EPIC_BKRECNCLN | Finance |
| Demand driven Replenishment - Predicted Individual Lead Time for Stock Transfer | INDIV_LEADTIME_STO2 | Manufacturing |
| Stock in Transit material overdue | MATERIAL_OVERDUE_SIT | Supply Chain |
| Predictive Scenario for Blocked Invoices with Quantity Variance | MMIV_INVC_QTY_BLCK | Procurement |
| Consumption Data for Slow or Non-Moving Materials | MMSLO_CONSUMPTION_02 | Procurement |
| Stock Level Data for Slow or Non-Moving Materials | MMSLO_STOCK_LEVEL_02 | Procurement |
| Monitors the predicted delay of the planned delivery creation of a sales order item | PRDTDDELIVCRTNDELAY | Sales |
| Monitors the predicted delay of the planned delivery processing of a sales order item | PRDTDDELIVPROCGDELAY | Sales |
| Project cost prediction | PREDICT_PROJECT_COST | R&D |
| QM – Defect Code Proposal Model | QM_DEF_CODE_PRPSL | QM |
| Quantity Contract Consumption | QTY_CONTRACT_CNSMPN | Procurement |
| Sales Performance - Prediction | SALESVOLUME001 | Sales |
| Sales Quotation Conversion Rate | SLSQTANPREDICTION | Sales |
| Supplier Delivery Prediction | SUPLRDELIVPREDICT | Supply Chain |
| Determine Release Confidence | UTI_BI_OUTSORTED | Industry Solution - Utilities |
| Process Implausible Meter Reading Results | UTI_MR_IMPLAUSIBLE | Industry Solution - Utilities |
Scenarios not delivered via ISLM
Several of the embedded Machine learning scenarios are delivered via ISLM. Some of the scenarios are not delivered via ISLM which means that those ML scenarios are trained internally by SAP and the customer has no control in terms of training.
Here is the list of 15 ML scenarios that are not yet available via ISLM. (not via Intelligent scenario management app)
| Scenario | Line of business |
| Receivables Line-Item Matching | Finance |
| Remittance Advice / Payment Advice Extraction | Finance |
| Tax Compliance Smart Automation / GRC | Finance |
| Financial account reconciliation | Finance |
| Lockbox | Finance |
| SAP Cash Application, add-on for contract accounting (Account Classification) | Finance |
| Demand-Driven Replenishment: Dynamic Buffer Level Adjustment | Manufacturing |
| Reduce Contract-Off spent | Procurement |
| Material Group Proposal | Procurement |
| Catalog item proposal | Procurement |
| Intelligent Approval of Purchase Requisitions | Procurement |
| Integrated Digital Content Processing for Content Mgt. | R&D Engineering |
| Business Rule Mining | Master data management |
| Intelligent Staffing & Resource Matching | Professional Services |
How is it different from other intelligent technologies like RPA and Situation handling?
Machine learning is used in Predicting the future outcomes by utilizing the past data and help the business users in decision making. It provides Real-time, predictive insights so users can make faster and better decisions and adjustments.
While RPA is used for automating the repetitive tasks, here the focus is to identify recurring tasks and automate them using RPA bots. It accelerates the digital transformation of business processes by automatically replicating tedious actions that have no added value.
On the other hand like ML, Situation Handling also help in decision making but it is based on Decision flow charts, that are created by the users to take decisions for different kind of situations. Here future outcomes are not predicted to make decisions.
List of ML Scenarios and an aggregated view:
| Scenario Name | Line of Business | Functionality | Type | Delivery |
| Receivables Line-Item Matching (CashApp) | Finance | This scenario Includes automating the mapping of incoming bank statement items to open receivables items or accounts. The historical clearing information is automatically sent from S/4HANA to the Machine Learning for Cash Application cloud service to train the model and derive matching criteria. | Side by Side | Non-ISLM |
| Remittance Advice / Payment Advice Extraction | Finance | In this Scenario, Payment advice documents received from the payer can be uploaded to the Manage Payment Advice Fiori app for automatic processing. The payment advice extraction feature within SAP Cash Application will accept PDFs including, unstructured scans, and use computer vision technology to read and extract information into the ERP. Once these payment advice documents are converted into structured information, they are used to enhance payment clearing. | Side by Side | Non-ISLM |
| Payables Line Item Matching | Finance | This Scenario helps in clearing the outgoing Payments, which the customer makes to their vendors. The system usually is unable to clear these payments automatically due to the minimum information stated on the bank statement. With Machine Learning, it will identify and propose the matching payables and automatically clear them. | Side by Side | |
| Lockbox | Finance | In this case, Accounts Receivable jobs can be scheduled to receive Machine Learning proposals to match open receivables to incoming lockbox items from checks and to automatically clear them. Machine Learning algorithms use trained data models with a large set of historical data of successfully cleared lockbox items. | Side by Side | Non-ISLM |
| SAP Cash Application, add-on for contract accounting (Account Classification) | Finance | In this scenario, the machine learning model in Cash Application is used to find matching proposals for all items which cannot be cleared based on the configured clearing rules in FI-CA. The Machine Learning approach can capture much richer detail of customer- and country-specific behavior, without the costs of manually defining detailed rules. Proposals are returned to S/4HANA and those that meet the configurable confidence threshold are automatically cleared for full automation. Proposals below a specific threshold are presented to the accountant within the standard Fiori app for further verification. | Side by Side | Non-ISLM |
| Tax Compliance Smart Automation / GRC | Finance | In this scenario, Using predictive analytics and machine learning, tax-checking processes can be automated to ensure compliance and Minimize risk and liability from non-compliance | Embedded | Non-ISLM |
| Financial account reconciliation | Finance | With this scenario, ML is used to automate and simplify the processing of purchase order items, the goods receipt, and the matching of the payables invoice in financial accounting. This accelerates and simplifies the process of reconciling goods and invoice receipts accounts and thus saves manual efforts to automate certain tasks in accounting and the financial close. | Side by Side | Non-ISLM |
| Business Integrity Screening / GRC | Finance | This scenario includes increasing the efficiency of Business Integrity Screening with Predictive Analytics. Predictive Analytics helps to lower the number of false-positive alerts and on the other hand, it provides insights into the parameters influencing fraud cases. | Embedded | |
| Detect Abnormal Liquidity Items | Finance | In this scenario, Historical Cashflow items and liquidity items configuration is used to detect abnormal and predict corresponding liquidity items. With this manual effort required for adjustment for liquidity items can be reduced and correct reporting within cash operations can be ensured. | Embedded | |
| Demand-Driven Replenishment: Dynamic Buffer Level Adjustment | Manufacturing | In DDR, Using Predictive ML Algorithms, a dedicated app helps planners manage safety stock, reorder point, and maximum stock through the buffer level proposals | Embedded | Non-ISLM |
| Demand-driven Replenishment - Predicted Individual Lead Time for Stock Transfer | Manufacturing | This scenario enables the users to predict the lead time for stock transfer products, based on historical data and using SAP Predictive Analytics Integrator (PAI). | Embedded | ISLM |
| QM – Defect Code Proposal Model | Manufacturing | Defect codes can be proposed using an analytical predictive model. The defect description and the detailed description are considered for the determination of the proposals for defect codes. If the system determines proposals for defect codes when you enter a detailed description, the field for defect codes is highlighted accordingly. The proposal function for defect codes is based on the Predictive Analytics Integrator (PAi), which integrates predictive functions in business applications. | Embedded | ISLM |
| Early Detection of Slow / Non-moving Stocks | Supply Chain | In this scenario, the information about material movements plus historical and current trends in supply and demand will be combined to predict which stock units will potentially become slow/non-moving. Based on these results you can react immediately with follow-on activities such as scrapping or stock transfers. | Embedded | ISLM |
| Stock in Transit | Supply Chain | This Scenario includes predicting delays in stock transfers. Users can see things like the shipping duration, posting date, forecasted delivery date, etc. | Embedded | ISLM |
| Reduce Contract-Off spent | Procurement | ML algorithms are used for the Proposal of options for Materials without Purchase Contract to reduce the contract-off spent | Side by Side | Non-ISLM |
| Material Group Proposal | Procurement | In this scenario using Machine learning algorithms, Free text data of the purchase orders is trained to predict the group code for a material. It helps the Purchaser to select from the proposed material group while creating purchase requisitions or purchase orders. | Side by Side | Non-ISLM |
| Catalog item proposal | Procurement | In this scenario, Using Machine Learning for proposing the creation of Catalog Items, SAP S/4HANA will help procurement organizations in a significant way to reduce their Free text Orders and therefore their process costs. | Side by Side | Non-ISLM |
| Supplier Delivery Prediction | Procurement | This scenario includes Predicting the PO delivery date for Purchase order items. With this information, you get a forecast based on your company’s empirical data if a goods receipt can be successfully completed in time. | Embedded | ISLM |
| Images-based purchasing | Procurement | In this scenario, machine learning algorithms are used to learn the required features and propose images of the catalog items for creating new Purchase Requisitions. The application will allow the user to search the catalog items based on a text search or image search, the user can then proceed to create a Purchase Requisition with the same | Side by Side | |
| Intelligent Approval of Purchase Requisitions | Procurement | ML scenario is used to analyze the approval pattern history for the purchase requisitions based on price, source of supply, material group, approvers, attachments, and so on, and provide recommendations for a mass automated approval. | Side by Side | Non-ISLM |
| Predictive Scenario for Blocked Invoices with Quantity Variance | Procurement | In case of blocked invoice items that are relevant for a cash discount, the system predicts in the Payment Block section. Based on data records of past goods receipts from the supplier, the system indicates whether the goods receipt can probably be expected on time concerning the cash discount due date. The prediction may help you to take the appropriate measures if the cash discount is at risk. In this case, you can, for example, send a reminder to your supplier. | Embedded | ISLM |
| Contract Consumption | Procurement | This Scenario includes a prediction of contract consumption date to Minimize or eliminate production delays due to stock-outs from expired contracts. Also provides Real-time data analysis with drill-down capabilities. | Embedded | ISLM |
| Integrated Digital Content Processing for Content Mgt. | R&D Engineering | Side by Side | Non-ISLM | |
| EPPM: Project cost forecast based on historical data | R&D Engineering | It enables the users to predict the overall project cost for new or running projects based on historical project data and using a K-nearest neighbors (KNN) algorithm. | Embedded | ISLM |
| Sales Performance Prediction | Sales, Services | Using this scenario, a Sales manager can gain predictive insights into the current sales performance in his area. SAP helps identify gaps between planned and predicted performance and takes action by expediting the execution of the sales plan and easily converting a quotation into a Sales order using Co-Pilot. | Embedded | ISLM |
| Quotation Conversion Probability Rate | Sales, Services | Quotation conversion rate indicates the percentage of net values converted from quotations into sales orders. As an internal sales representative or a sales manager, you can use quotation conversion rates to track to what extent your quotations are being converted to sales orders before expiring. By leveraging machine learning capabilities, you can gain predictive insights into quotation conversion by comparing actual and predicted results | Embedded | ISLM |
| Delivery Performance / Delivery in Time | Sales, Services | In this scenario, the system can predict the delivery delay by using past data to compare the planned delivery creation date from the confirmed schedule line of delivered sales order items with the actual delivery date of the corresponding delivery. | Embedded | ISLM |
| Service Ticket Intelligence | Sales, Services | In This scenario, ML Algorithms are used in the ticket classification feature of Service Ticket Intelligence, and it can work in combination with C4Service routing capabilities to shorten case closure times. | Side by Side | Non-ISLM |
| Automatic Creation of Sales order | Sales, Services | In this scenario, ML algorithms are used to create sales orders from the extraction of PDF files. After a purchase order file in PDF format is uploaded, the system automatically extracts data from the file and determines master data (for example, the sold-to party) from the extracted data. | Side by Side | ISLM |
| Most likely damage code for a maintenance notification | Asset Management | In this scenario, ML algorithms are used to propose the most likely damage code for a maintenance notification | Embedded | ISLM |
| Object Part Code in Notifications to Maintenance Technicians | Asset Management | In this scenario, Machine Learning-Based Suggestions are given for Object Part Code in Notifications to Maintenance Technicians | Embedded | ISLM |
| Business Rule Mining | Master Data Management | This Scenario includes Rule mining, Rule mining applies machine learning to discover rules in existing master data. The rule repository of MDG allows you to manage your rules for master data quality in one single place. It provides you with a repository to catalog and define data-quality rules, including comprehensive descriptions of rules, business aspects, and contact information. | Embedded | Non-ISLM |
| Late Payment Prediction - Behavioral Insights | Public Services | Use of Machine learning to government tax and collection data to get customer behavioral insights in order to identify business partners and/or contract accounts with a high propensity to miss the upcoming due date in open invoices and bills | Embedded | ISLM |
| Receivership Prediction - Behavioral Insights | Public Services | Use of Machine learning to government tax and collection data in order to get customer behavioral insights in order identify business partners and/or contract accounts with a high propensity to file for bankruptcy (or any user custom defined status change) | Embedded | ISLM |
| Late Filing Prediction | Public Services | Use of Machine learning to government tax and collection data in order to get customer behavioral insights in order identify business partners and/or contract accounts with a high propensity to miss the upcoming due date in tax returns | Embedded | ISLM |
| Determine Release Confidence | Industry Solution - Utilities | In This case, API objects are used in the Pai predictive scenario to determine the release confidence value for outsorted documents. | ISLM | |
| Process Implausible Meter Reading Results | Industry Solution - Utilities | Here Machine learning can be used to process an implausible meter reading results. Release confidence values for implausible meter reading results can be determined and Automatic release of implausible meter reading results, based on the release confidence against a given threshold is possible. | ISLM |
Achieving an end to end process automation using intelligent capabilities:
Fig 1: Managing delivery delays
Fig 2: Invoice to pay process
Some Key terms:
Some of the key terms related to ML model training to understand how effective an ML model is are explained below:
Predictive Power-
Description:
The predictive power of a model is the quality indicator of models generated using the application. This indicator corresponds to the proportion of information contained in the target variable that the explanatory variables can explain. To improve the predictive power of a model, new variables may be added to the training dataset. Explanatory variables may also be combined
Details:
A model with a predictive power of:
- "0.79" can explain 79% of the information contained in the target variable using the explanatory variables contained in the dataset analyzed.
- "1" is a hypothetical perfect model, capable of explaining 100% of the target variable using the explanatory variables contained in the dataset analyzed. In practice, such a predictive power would generally indicate that an explanatory variable 100% correlated with the target variable was not excluded from the dataset analyzed.
- "0" is a purely random model
Link:
https://help.sap.com/doc/5e8fcd4e118542f8945ebdba4db4ad66/3.3/en-US/AA_userGuideandScenarios_EN.pdf
Predictive Confidence-
Description:
The prediction confidence is the robustness indicator of the models generated using the application. It indicates the capacity of the model to achieve the same performance when it is applied to a new data set exhibiting the same characteristics as the training dataset. To improve the prediction confidence of a model, additional observation rows may be added to the training
Details:
A model with prediction confidence:
- Equal to or greater than "0.98" is very robust. It has a high capacity for generalization.
- Less than "0.95" must be considered with caution. Applying it to a new dataset will incur the risk of generating unreliable results.
Links:
https://help.sap.com/doc/5e8fcd4e118542f8945ebdba4db4ad66/3.3/en-US/AA_userGuideandScenarios_EN.pdf
Conclusion-
In this Blogpost, we have seen how SAP S/4HANA Machine learning models can be used to address typical customer scenarios right from detecting anomalies in financial transactions to predicting the future sales. Nowadays Machine learning based applications are helping all kinds of professionals right from Marketing experts, Financial experts, Demand planners to Procurement managers to take right business decisions. On top of that These machine learning models can be combined with other intelligent solutions like intelligent RPA and Situation handling to achieve end-to-end process automation.
Thanks for reading the Blogpost, I look forward to your comments. Alternatively, you can also post your queries on SAP S/4HANA Cloud Community.
Image Source in the Blogpost: SAP Frontrunner Team
2 Comments
