A framework for evaluating and implementing Machine Learning in S/4HANA
There is an increasing pressure on the enterprises to do more with less, provide superior customer/employee experience and continuously adapt to the changing business models. Technologies such as Cloud, Artificial Intelligence (AI), Robotic Process Automation (RPA) etc plays a pivotal role in enabling enterprises to meet these challenges.
SAP, a leader in cloud based ERP solutions for its part is embedding AI in the applications (S/4HANA and other apps) by making the business process ‘Intelligent’ and enables process automation, combining Intelligent Robotic Process Automation (SAP Intelligent RPA), SAP Conversational AI and SAP AI Business Services through its Business Technology Platform (https://www.sap.com/australia/products/business-technology-platform.html)
Most recent advances in AI have been achieved by applying Machine Learning (ML) to large data sets. Machine Learning algorithms (models) detect patterns and learn how to make predictions and recommendations by processing data rather than by explicit programming logic.
In this blog post, you will learn couple of simple frameworks to assess business problems in the context of Machine Learning and SAP and select a right approach to solve the problem. This blog will help answer the below questions:
- What are the business problems should and should not be solved by using ML?
- For the problems to be solved by ML, what is the right approach to be used?
To summarise, if your business problem has no clear business logic to solve (or if the logic involves very complex business rules) with historic data, it is a good candidate to solve using ML (E.g Predicting delivery time or segmenting prospects) On the contrary if there is a clear business logic to solve your business problem, you can use programming logic to solve your business problem (E.g calculating inventory)
Today with the rise of public cloud platforms and many independent data analytics tools, it is easy to implement ML in any organisation. SAP for its part is directly embedding ML into its business applications and enabling citizen developers / data scientists through its Business Technology Platform (BTP) to develop ML models and consume in the business process.
In general, there are two ways of implementing ML in SAP – Embedded and Side-by-Side.
The above figure illustrates a simple version of ML deployment options available for the S/4HANA landscape. Whilst, SAP is embedding ML directly in the business process in S/4HANA and other applications, it also offers a plug and play version of reusable AI services and a ML platform via Data Intelligence to develop custom models in Business Technology Platform. With these options available, naturally the question comes – “When should I use which option?” To answer this question, let us consider the “Three laws of Cloud” which was made famous by the current Intel CEO Pat Gelsinger.
- Laws of Physics – Enterprises need to have their computing workloads (read ML models) closer to the data if they need a near instant response time for processing (prediction)
- Laws of Economics – Enterprises with on-premise infrastructure are turning to public cloud services due to rising cost of storage, network and bandwidth as they scale their business
- Laws of land – Due to data sovereignty rules, enterprise have to move workloads out of public cloud to on-premise or private secure cloud
All the above laws indicate one underlying principle – “Bring your models closer to your data” to avoid data latency and expensive data transfer costs. Let us use these principles in putting together a decision matrix to answer our second question – “For the problems to be solved by ML, what is the right approach to be used in SAP?”
The key decision criteria are the location of the data / business logic and the type of use case (and its impact on computing requirements) If your core business process uses S/4HANA and other LOB/Industry SAP apps, it makes sense to leverage SAP ML tools and at the same time, if you are a matured organisation with highly dispersed applications footprint (best of breed), then you can opt for a hybrid option. SAP and leading hyperscale cloud providers have reference architectures for hybrid deployment options to use best of both worlds (SAP and Hyperscale)
This blog is to give a high-level overview of the ML deployment options available and a simple framework to select the appropriate option. A detailed assessment depending on the customer context (Business / Technology Strategy, Architecture/Cloud Strategy and Guardrails, Data Analytics Strategy and Maturity, Applications Landscape, Executive commitment to enterprise AI etc) is required before recommending appropriate ML deployment option.
Thanks for your interest in the topic and I hope this blog post would have answered some common questions on implementing Machine Learning in S/4HANA and various deployment options. Please share your feedback in the comments section.
If you have any questions about Machine Learning in S/4HANA, I would encourage to ask questions in SAP Community using this Q&A tag link for Machine Learning: https://answers.sap.com/questions/ask.html?primaryTagId=240174591523510321507492941674121
Hi Vasanth Kandaswamy!
Thank you for this blog which points out on a high level when ML in S/4HANA makes sense or not.
Would be interested which kind of people and roles you see here as I have some discussions around this. Do you see more SAP or Non-SAP guys implementing this in general (probably deprendent on the scenario) and more SAP functional consultants, SAP developers or maybe Data Scientists interested in implementing S/4HANA ML? Especially if we talk about the embedded scenario.
Thanks for your question Peter. I see a blend of team working. The Data Engineers / Data scientists mostly work on the core data engineering / machine learning tasks and SAP consultants used to frame the problem statements, data validation, testing and implementation of the use case. In future, we will see consolidation in the workforce. i.e A SAP consultants/ developers building end to end ML pipelines with little or no support from data scientists.