SAP AI and ML Tools – when to use what, insights 2022 edition

[leojmfrancia]

After being an SAP technical consultant for more than 10 years, serving various industries; and having done presales for over 3 years, I explored Artificial Intelligence and Machine Learning (AI/ML). Having been an AI/ML Catalyst in SAP’s 2022 cohort during my time there, I had access to various training materials as well as product and subject matter experts globally within SAP.

In my previous LinkedIn post, I shared a simple decision tree on when to use which SAP tool for AI/ML purposes (included here for convenience).

This blog might be useful if any of these apply to you:

1. You are a data/business analyst or an SAP practitioner looking at ways to improve and automate decision making


2. You have acceptable data quality levels, e.g., based on business and technical team assessments


3. You have your data sources ready, with teams agreeing on who will do what in which tool to make the data ready for consumption. I recommend checking out pbaumann08's blog series on data architectures starting here.


4. You have a business problem or use case in mind


5. You are deciding which SAP AI/ML tool to apply to data with (presumably) a large SAP footprint

Now, onto the table. Having used these tools in the last 2 years, there are areas that I have not explored comprehensively yet. Eager to hear your thoughts and experience.

Tool Name / Attribute	What the attribute means	SAP Data Intelligence	SAP Al Core and Launchpad	SAP Al Business Services	SAP Analytics Cloud Predictive Scenarios	SAP Embedded Intelligent Scenario Lifecycle Management (ISLM)	Hyperscaler ML platforms (Azure ML, AWS Sagemaker, GCP ML, Databricks, etc.)
What the tool is for	-	Big Data orchestration and pipelining	Base AI/ML integration and operations platform	For different use cases, see here	Analytics tool with predictive capabilities	Operate machine learning scenarios within S/4HANA	Big Data orchestration and pipelining + Machine Learning
Suggested link for initial reading	-	SAP Data Intelligence: Self-learning resources	SAP AI Core & Launchpad Introduction	SAP AI Business Services	Your first Predictive Scenario in SAC	Brief Introduction to ML capabilities in S/4HANA	Federated Machine Learning Libraries for Hyperscalers
Development environment	Where you perform the creation of data pipelines, models, etc.	SAP Data Intelligence, Jupyter Notebook	SAP BTP	SAP BTP	SAP Analytics Cloud	SAP S/4HANA, SAP BTP, 3P libraries (e g. via Jupyter Notebook)	Jupyter Notebook, platform-specific
3P libraries	Third-party libraries like pandas, NumPy, sci-kit learn, etc.	Yes	Yes	No	No	No	Yes
Container support	Allows build, test, and deployment of packaged images that runs in any environment	Yes	Yes	No	No	No	Yes
Native SAP integration	Preconfigured option to connect to SAP data sources	Yes	Yes	Yes	Yes	Yes	No
Job scheduling or orchestration	Automation and monitoring of data flow, from transformation to final output	Yes	Yes	No	Yes	No	Yes
Pipelining	Moving data from one place to another	Yes	Yes	No	No	Not applicable	Yes
Pre-processing	Manipulation or transformation of data	Yes	Yes (using libraries)	No	No (limited)	Not applicable	Yes
Model training	Using new data for an algorithm to learn and make predictions	Yes (using libraries)	Yes (using libraries)	Yes	Yes	Yes	Yes
Classification	Ordering or categorization of data into one or more "classes" (e.g., Spam vs Not Spam)	Yes (using libraries)	Yes (using libraries)	Yes	Yes	Yes	Yes
Regression	Predicts continuous values based on variable relationships	Yes (using libraries)	Yes (using libraries)	Yes	Yes	Yes	Yes
Clustering	Grouping of unlabelled examples	Yes (using libraries)	Yes (using libraries)	Yes	No	No	Yes
Natural Language Processing	Breaking down and interpretation of human language	Yes (using libraries)	Yes (using libraries)	Yes	No	No	Yes
Neural network	Processing of data like the human brain, learning through trial-and-error	Yes (using libraries)	Yes (using libraries)	Yes	No	No	Yes
AutoML	Automation of repetitive tasks when building an ML model	Yes (using libraries)	Yes (using libraries)	Yes	Yes	No	Yes
Batch model serving (inference)	Making ML functions available via reports	Yes (using libraries)	Yes	Yes	Yes	Yes	Yes
Online model serving (inference)	Making ML functions available via API	Yes (using libraries)	Yes	Yes	No	Yes	Yes
MLOps	Efficient and reliable deployment and maintenance of ML models	Yes	Yes	Yes	Yes	No	Yes

Note: “Using libraries” means the feature is not native and you will use libraries in tools such as Jupyter Notebook to make the feature available in your deployment. The table above does not include details on image classification capabilities as of now, as I have not yet personally come across use cases beyond structured and unstructured text.

Once you have identified the most suitable tool for your use case, I suggest considering the steps below before proceeding with a full-blown project (mark.muir has a blog about this):

1. Review the tool prerequisites


2. Review the licensing


3. Integrate the AI/ML tool into your pipeline


4. Design your AI/ML experiment and test approach

In summary, there are different approaches you can consider on how to deploy AI/ML in your SAP landscape depending on your requirements. I hope it gives you a starting point in exploring the AI/ML tools available. The applicability of the information in the table above will likely change (and quickly) over time. To stay up to date, you can follow pages such as:

• SAP AI Core environment topic page


• SAP AI Core questions


• Other posts on SAP AI core


• And my profile, leojmfrancia, for upcoming posts on Data Science, AI, and ML

Would be great to hear about your thoughts and experience using these tools in the comments section.

 

Invariably stochastically yours,

Leo