In our previous blog, we introduced the SAP and DataRobot partnership, along with an updated solution architecture and roadmap. In this post, we will dive deeper into how SAP Datasphere and DataRobot collaborate in a real business scenario, highlighting the potential outcomes of this integration.
We will look at a retail inventory management case, where understanding inventory is crucial to avoid stockouts. This becomes challenging at scale, especially for large companies managing thousands of SKUs across multiple countries with varying purchasing patterns, holidays, and seasonal factors. Each store’s performance also varies, adding complexity. The main challenge is handling this vast data scale while ensuring fast and accurate forecasts. Additionally, companies need to maintain forecast reliability over time, quickly identifying and addressing issues like data inaccuracies or performance drops before they affect operations.
For this demonstration, we are using SAP Datasphere and SAP Analytics Cloud business content, which includes pre-configured information models and templates designed to simplify the development of analytics and reporting for specific business scenarios. For SAP Datasphere, this involves predefined models, ready-to-use configurations, and templates that showcase various features. These content packages are tailored to specific industries or lines of business (LoB). Similarly, SAP Analytics Cloud (SAC) offers predefined stories, dashboards, and data models built around existing SAP data sources, also customized for specific industries and LoBs.
For this walk-through, we will use a dataset provided by SAP, which we will upload to an SAP Datasphere instance for reporting and modeling exercises, as shown in Figure 1. This dataset will serve as the basis for developing a forecasting model to predict sales volumes at both the organizational and monthly levels.
An analytical model is created from historical data to provide insights for business leaders. SAC will then directly consume this model from SAP Datasphere, enabling business insights to be visualized through an SAC dashboard. Figure 2 illustrates the SAC dashboard, which displays historical revenue across various regions and other factors influencing enterprise revenue.
Figure 1 (left): Data Builder in SAP Datasphere | Figure 2 (right): Sales Dashboard in SAP Analytics Cloud, based on historical data.
After reviewing historical revenue values, business leaders often ask how to forecast revenue for the upcoming year. An accurate forecast would be invaluable for enterprise planning.
To support informed decision-making, the SAC visualization dashboard can be expanded beyond basic BI reporting to incorporate predictive AI-driven sales forecasting models. While several tools are available for building such models, DataRobot stands out for its robust technical and functional capabilities in developing and deploying accurate forecasts.
In the following sections, we will explore key features of DataRobot that align with a business leader’s goal of integrating accurate, timely forecasts into the decision-making process, highlighting how DataRobot addresses the predictive AI needs of SAP customers who want to rapidly build both predictive and generative AI models.
Before we dive into the steps a data scientist would take to implement the business leaders’ vision, here’s a brief overview of DataRobot. DataRobot offers features for both low-code and code-first data scientists, which we briefly covered in our previous blog. The following screenshot illustrates DataRobot’s intuitive interface, which consolidates AI model building (Workbench), governance (Registry), and operations (Console) within a single application.
Figure 3: Home screen view in DataRobot.
The AI model creation process starts with retrieving the necessary training dataset. DataRobot provides integrated connectors to industry-standard data warehouses, including SAP Datasphere, as shown in the following screenshot. In this case, a data scientist can use the SAP Datasphere connector to access historical sales data stored in SAP Datasphere for modeling. By entering the required credentials and instance-specific information, data scientists can connect directly to the data views within SAP Datasphere used for modeling.
Figure 4: DataRobot connector for connecting to SAP Datasphere for data read/write operations.
Once the modeling dataset is acquired, the next step is typically data wrangling to prepare the data for modeling or to meet specific business requirements. DataRobot offers an intuitive data wrangling feature that allows data scientists to perform complex manipulations without extensive SQL knowledge. In the screenshot in Figure 5, the data wrangling task demonstrates the creation of new features, removal of redundant ones, and aggregation of data at both the organizational and monthly levels. The monthly aggregation was requested by the business leader to support forecasting revenue across all organizations by month for the next year.
Figure 5: Data wrangling framework enables SQL-like data modification tasks without requiring SQL knowledge.
After preparing the training dataset through data wrangling, a time series forecasting model can be developed using DataRobot’s intuitive interface. DataRobot offers a wide range of algorithms – both open-source and proprietary – essential for building business-focused AI models. Supported modeling approaches include supervised and unsupervised learning, clustering, segmentation, and time series forecasting. The screenshot in Figure 6 showcases the time series model creation interface, allowing data scientists to fine-tune key aspects of the model, such as the historical data range for feature generation, the forecast horizon, and the selection of relevant accuracy measures.
Figure 6: Time series modeling interface in DataRobot.
Once the model hyperparameters are set, DataRobot automatically derives the necessary features from the original dataset to support the creation of an optimal model. It generates a model using both original and derived features, through either a fully automated or manual process. During this phase, DataRobot selects the most appropriate modeling blueprints based on the available feature sets. A blueprint represents the end-to-end process for fitting the model, including preprocessing, modeling, and post-processing steps.
After completing the modeling process, DataRobot presents a list of tested blueprints, highlighting the one that delivered the most accurate results, which is recommended for deployment. The selected model is then packaged for production deployment to generate forecasts at the desired frequency or manually if needed.
The screenshot in Figure 7 shows the leaderboard at the end of the model creation process, along with various artifacts automatically generated by DataRobot, including blueprint documentation, feature impact, and model accuracy.
Figure 7: Model leaderboard showing tested blueprints, highest-ranked model, and feature impact.
After selecting a model for deployment, users can deploy it to a production environment with a single click. DataRobot offers multiple deployment options, including one-click deployment to SAP AI Core, as shown in Figure 8.
Figure 8: Final output of the model creation process, highlighting the leaderboard and deployment options.
Once the selected model is deployed in the production environment, DataRobot provides a robust framework for monitoring its performance and maintaining prediction accuracy over time. Through this framework, users can assess data drift, prediction accuracy, and overall deployment health, as shown in Figure 9.
Figure 9: Interface for monitoring model health and data drift during operation in the business environment.
Figure 10 shows how a scoring dataset can be accessed directly from an SAP Datasphere data view to generate forecasts. The generated predictions can also be written back into SAP Datasphere for further use within SAC.
Figure 10: Scoring dataset from SAP Datasphere consumed by DataRobot to generate and store forecasts back in SAP Datasphere.
The forecasts generated by DataRobot are integrated into an SAP Datasphere data flow, becoming part of an SAP Datasphere analytical model. These forecasts can then be accessed by SAP Analytics Cloud, enabling visualizations that combine forecasted and historical sales data. For this tutorial, the generated forecasts are available in an SAP Datasphere data view, as shown in Figure 11.
Once the forecast data is generated in DataRobot and integrated into an SAP Datasphere analytical model, it can be accessed by SAC for visualization via SAP Datasphere connectivity. The forecast page, added to the original SAC dashboard, is shown in Figure 12. This new dashboard page displays the generated forecasts and highlights other key factors of interest. For example, the bottom section of Figure 12 showcases features impacting the forecast, allowing users to perform ad-hoc what-if analysis using SAC’s drill-down functionality.
Figure 11 (left): SAP Datasphere data view for consuming generated forecasts | Figure 12 (right): Additional page added to the existing SAP SAC BI dashboard to help business leaders understand and make decisions based on next year’s sales forecast.
SummaryÂ
In this blog, we have demonstrated how existing SAP Datasphere Business Contents can be extended to address specific business needs. DataRobot and SAP Datasphere work together to help customers unlock the full value of their business data, providing the tools necessary to manage both scale and granularity while tackling complex use cases. This integration empowers businesses to make more informed decisions, improve forecasting accuracy, and drive better outcomes.
Acknowledgments
I would like to thank Farooq Azam from DataRobot for his valuable contributions to this blog. His expertise on the SAP and DataRobot integration was instrumental in shaping the content for this blog.
Further linksÂ
SAP and DataRobot Partnership Announcement at Data Unleashed 2023
Previous Blog Post on Partnership Updates in 2024
DataRobot
DataRobot Resource LibraryÂ
Â
Â
​ In our previous blog, we introduced the SAP and DataRobot partnership, along with an updated solution architecture and roadmap. In this post, we will dive deeper into how SAP Datasphere and DataRobot collaborate in a real business scenario, highlighting the potential outcomes of this integration.We will look at a retail inventory management case, where understanding inventory is crucial to avoid stockouts. This becomes challenging at scale, especially for large companies managing thousands of SKUs across multiple countries with varying purchasing patterns, holidays, and seasonal factors. Each store’s performance also varies, adding complexity. The main challenge is handling this vast data scale while ensuring fast and accurate forecasts. Additionally, companies need to maintain forecast reliability over time, quickly identifying and addressing issues like data inaccuracies or performance drops before they affect operations.For this demonstration, we are using SAP Datasphere and SAP Analytics Cloud business content, which includes pre-configured information models and templates designed to simplify the development of analytics and reporting for specific business scenarios. For SAP Datasphere, this involves predefined models, ready-to-use configurations, and templates that showcase various features. These content packages are tailored to specific industries or lines of business (LoB). Similarly, SAP Analytics Cloud (SAC) offers predefined stories, dashboards, and data models built around existing SAP data sources, also customized for specific industries and LoBs.For this walk-through, we will use a dataset provided by SAP, which we will upload to an SAP Datasphere instance for reporting and modeling exercises, as shown in Figure 1. This dataset will serve as the basis for developing a forecasting model to predict sales volumes at both the organizational and monthly levels.An analytical model is created from historical data to provide insights for business leaders. SAC will then directly consume this model from SAP Datasphere, enabling business insights to be visualized through an SAC dashboard. Figure 2 illustrates the SAC dashboard, which displays historical revenue across various regions and other factors influencing enterprise revenue.Figure 1 (left): Data Builder in SAP Datasphere | Figure 2 (right): Sales Dashboard in SAP Analytics Cloud, based on historical data.After reviewing historical revenue values, business leaders often ask how to forecast revenue for the upcoming year. An accurate forecast would be invaluable for enterprise planning.To support informed decision-making, the SAC visualization dashboard can be expanded beyond basic BI reporting to incorporate predictive AI-driven sales forecasting models. While several tools are available for building such models, DataRobot stands out for its robust technical and functional capabilities in developing and deploying accurate forecasts.In the following sections, we will explore key features of DataRobot that align with a business leader’s goal of integrating accurate, timely forecasts into the decision-making process, highlighting how DataRobot addresses the predictive AI needs of SAP customers who want to rapidly build both predictive and generative AI models.Before we dive into the steps a data scientist would take to implement the business leaders’ vision, here’s a brief overview of DataRobot. DataRobot offers features for both low-code and code-first data scientists, which we briefly covered in our previous blog. The following screenshot illustrates DataRobot’s intuitive interface, which consolidates AI model building (Workbench), governance (Registry), and operations (Console) within a single application.Figure 3: Home screen view in DataRobot.The AI model creation process starts with retrieving the necessary training dataset. DataRobot provides integrated connectors to industry-standard data warehouses, including SAP Datasphere, as shown in the following screenshot. In this case, a data scientist can use the SAP Datasphere connector to access historical sales data stored in SAP Datasphere for modeling. By entering the required credentials and instance-specific information, data scientists can connect directly to the data views within SAP Datasphere used for modeling.Figure 4: DataRobot connector for connecting to SAP Datasphere for data read/write operations.Once the modeling dataset is acquired, the next step is typically data wrangling to prepare the data for modeling or to meet specific business requirements. DataRobot offers an intuitive data wrangling feature that allows data scientists to perform complex manipulations without extensive SQL knowledge. In the screenshot in Figure 5, the data wrangling task demonstrates the creation of new features, removal of redundant ones, and aggregation of data at both the organizational and monthly levels. The monthly aggregation was requested by the business leader to support forecasting revenue across all organizations by month for the next year.Figure 5: Data wrangling framework enables SQL-like data modification tasks without requiring SQL knowledge.After preparing the training dataset through data wrangling, a time series forecasting model can be developed using DataRobot’s intuitive interface. DataRobot offers a wide range of algorithms – both open-source and proprietary – essential for building business-focused AI models. Supported modeling approaches include supervised and unsupervised learning, clustering, segmentation, and time series forecasting. The screenshot in Figure 6 showcases the time series model creation interface, allowing data scientists to fine-tune key aspects of the model, such as the historical data range for feature generation, the forecast horizon, and the selection of relevant accuracy measures.Figure 6: Time series modeling interface in DataRobot.Once the model hyperparameters are set, DataRobot automatically derives the necessary features from the original dataset to support the creation of an optimal model. It generates a model using both original and derived features, through either a fully automated or manual process. During this phase, DataRobot selects the most appropriate modeling blueprints based on the available feature sets. A blueprint represents the end-to-end process for fitting the model, including preprocessing, modeling, and post-processing steps.After completing the modeling process, DataRobot presents a list of tested blueprints, highlighting the one that delivered the most accurate results, which is recommended for deployment. The selected model is then packaged for production deployment to generate forecasts at the desired frequency or manually if needed.The screenshot in Figure 7 shows the leaderboard at the end of the model creation process, along with various artifacts automatically generated by DataRobot, including blueprint documentation, feature impact, and model accuracy.Figure 7: Model leaderboard showing tested blueprints, highest-ranked model, and feature impact.After selecting a model for deployment, users can deploy it to a production environment with a single click. DataRobot offers multiple deployment options, including one-click deployment to SAP AI Core, as shown in Figure 8.Figure 8: Final output of the model creation process, highlighting the leaderboard and deployment options.Once the selected model is deployed in the production environment, DataRobot provides a robust framework for monitoring its performance and maintaining prediction accuracy over time. Through this framework, users can assess data drift, prediction accuracy, and overall deployment health, as shown in Figure 9.Figure 9: Interface for monitoring model health and data drift during operation in the business environment.Figure 10 shows how a scoring dataset can be accessed directly from an SAP Datasphere data view to generate forecasts. The generated predictions can also be written back into SAP Datasphere for further use within SAC.Figure 10: Scoring dataset from SAP Datasphere consumed by DataRobot to generate and store forecasts back in SAP Datasphere.The forecasts generated by DataRobot are integrated into an SAP Datasphere data flow, becoming part of an SAP Datasphere analytical model. These forecasts can then be accessed by SAP Analytics Cloud, enabling visualizations that combine forecasted and historical sales data. For this tutorial, the generated forecasts are available in an SAP Datasphere data view, as shown in Figure 11.Once the forecast data is generated in DataRobot and integrated into an SAP Datasphere analytical model, it can be accessed by SAC for visualization via SAP Datasphere connectivity. The forecast page, added to the original SAC dashboard, is shown in Figure 12. This new dashboard page displays the generated forecasts and highlights other key factors of interest. For example, the bottom section of Figure 12 showcases features impacting the forecast, allowing users to perform ad-hoc what-if analysis using SAC’s drill-down functionality.Figure 11 (left): SAP Datasphere data view for consuming generated forecasts | Figure 12 (right): Additional page added to the existing SAP SAC BI dashboard to help business leaders understand and make decisions based on next year’s sales forecast.Summary In this blog, we have demonstrated how existing SAP Datasphere Business Contents can be extended to address specific business needs. DataRobot and SAP Datasphere work together to help customers unlock the full value of their business data, providing the tools necessary to manage both scale and granularity while tackling complex use cases. This integration empowers businesses to make more informed decisions, improve forecasting accuracy, and drive better outcomes.AcknowledgmentsI would like to thank Farooq Azam from DataRobot for his valuable contributions to this blog. His expertise on the SAP and DataRobot integration was instrumental in shaping the content for this blog.Further links SAP and DataRobot Partnership Announcement at Data Unleashed 2023Previous Blog Post on Partnership Updates in 2024DataRobotDataRobot Resource Library     Read More Technology Blogs by SAP articlesÂ
#SAP
#SAPTechnologyblog
