[MUSIC PLAYING] Hi, good afternoon, good morning, good evening, wherever you are all dialing in from. My name is Raghu Banda. I'm a Senior Director with SAP Labs based out of Palo Alto in the West Coast. Today, I'll be talking through optimize business processes session, which is predominantly focused on how you do machine learning with SAP S/4HANA. Let us go into quickly looking at the agenda.
I have mainly three parts-- four parts of the agenda, starting with the overview of AI in SAP software. And then we will kind of go into a little bit deeper into the different approaches that we use while doing machine learning with SAP, predominantly SAP S/4HANA. Then in the third section, I'll go very briefly into the methodology, where-- what are the different goals and technologies that we leverage in doing these machine learning. And then finally, we'll wrap it up with a quick overview of the use cases and the best practices that we have used in the context of building these SAP S/4HANA-related business processes. And of course, at the end, I'll quickly give you some references of how you can get more information on this. So without missing much time, we can kind of get into the first session.
Let us now go into a quick overview of AI in the context of SAP software. We all know that there is a lot of things happening around AI in the AI space. Whether it is with the consumer space or the enterprise space, there a lot of things are happening. There are quite a lot of companies thereout who are directly involved with doing artificial intelligence into their business processes. Some of them have performed very well, while some of them have still a lot of issues.
The main difference is that there are different ways of integrating or embedding intelligence into these business processes. While some of the firms have learnt the art of embedding this intelligence and also kind of taking it to the next step of deploying it into their production technologies, some of the business processes or some of the companies are not yet there. So in the context of this session, we will see how we can kind of address some of these, so that the SAP customers and partners can benefit out of it.
This is kind of a repeat of my earlier slide, but I will kind of give a little more details into this. Like I said, many companies are definitely struggling with implementing artificial intelligence. While, as you know, there are 65% to 70% of these companies who do get into this issue of how do we kind of embed intelligence, there are a few companies who are really confident in kind of getting this done.
The problem is twofold. One is that you will have to identify what are the business processes that typically need intelligence that has to be embedded. And then how do you take it to the next step of kind of taking it to the production step and then get the customers on board on that? So these are the main two things that we will have to get into it. Before going into the business processes and what are the different approaches of doing this, let us give a quick background into what is intelligent information management. There are different ways of looking at it, right? Like, you have data coming from different sources.
You have structured data. You have unstructured data. You have streaming data. You have sensor data. These are different kinds of data that's coming in.
So you will have to discover which kind of data is needed, and you have to refine the data. And then it is like a musician who is trying to compose a particular symphony at a particular musical instrument, right? You have-- if you go for a rehearsal or a practice session, you will see too much of chaos. But once all this is put in together, and if you go to the actual performance, you see that it all comes out beautifully. In the same way, you get data from different kinds of sources. How you compose the data and orchestrate the data, that is where-- how well you can manage your information management.
And finally, there are some rules and regulations that you will have to follow, where your data is coming in from, what kind of changes you are doing, and how do you send this data out. So this is the final step of how you govern and comply within the rules and regulations of how do you manage this intelligent information data. Now, let us take this a step further and understand what exactly is this enterprise AI and how do you manage this data, build the models. And then we will get into the next section. So like I've explained earlier, the overall idea is that you have humongous amount of data coming in from different sources. And now, you will have to understand and interpret this data and build the models.
So managing the data is the very first step where you have in the enterprise world or in a consumer world. In the enterprise world, it is much more important, the reason being, enterprises are very much dependent on how their customers behave. So they will have to learn a lot from the data that it's coming in from.
So typically, managing the data which is coming from different sources is the first key important factor. The second step is that, how do you manage the development of these models that you built based out of this data? You will have different development teams. You will have different data scientists, architects, or your business users who are involved when you are building some kind of models. And this is where-- how well you can manage your development and take it to the next step.
And then finally, once your models are built, now, how do you deliver these models or how do you deliver it into your end result of deploying into the system and taking it to the production system, and how your end users or your customers can leverage it? So this is where you have the overall complete end-to-end process of how an enterprise AI would look like. How do you manage your data? And based out of the data, how do you manage your development of building the data models? And how do you finally deliver these models into these applications that you're embedding intelligence into it? So these are the three different steps that we have to take care of when you are talking about enterprise AI. With this, now, we will kind of-- before getting into the next section, I would kind of wrap it up before getting into the next section.
So as you see in this slide, you will notice that there are different aspects of intelligent technologies. And you will-- we are trying to connect these different aspects of the human-- intelligent technologies in a human way. Thinking and decision making is one kind of a thing where we leverage machine learning and decision making. And then listening and speaking is the other aspect of human intelligence, which is where you listen and speak into the system. That is where we leverage SAP Conversational AI. And then acting based out of repeatable steps, that is where you leverage your robotic process automation.
And then understanding and the experience-- based out of your experience, how do you complete the customer feedback loop? That is where you have your experience management coming in. And at the heart of all this is SAP S/4HANA. So this slide, before I kind of get into the next section, I want to kind of give an understanding of where SAP S/4HANA is kind of standing in there. So that is the reason I'm kind of giving a big picture of how this listening and conversational AI aspect of it, and then how do you think and act, and how do you kind of understand the experience of the customer, complete the customer feedback loop. All right, so let us now continue to the next section, where we will focus on doing-- understanding the different approaches while doing machine learning and predictive analytics with the SAP S/4HANA.
Now, in this slide, in the next few slides, we'll focus on this. In this slide, let us first understand what are the different approaches of doing predictive analytics and machine learning with SAP S/4HANA. On the top left hand side, you will see, basically, we will be discussing about three different ways of doing it. One is embedding into SAP S/4HANA. The second approach is consuming the services that are built on a SAP Cloud Platform or SAP Data Intelligence. And the third is how can you leverage or do additional approaches with SAP Analytics Cloud.
These are three different ways of doing this. Later on, in the fourth-- I would say it is not a fourth approach, but it's a different way of how you can enhance and extend already released models that are available with these business cases or the business processes. So as you understand, in this slide, we have seen all these four different ways of approaches that are leveraged with predictive analytics and machine learning. Now, let us continue into the next slide.
I will walk you through a quick decision tree kind of an approach, where you can understand how you define-- or how you decide which particular approach you would want to leverage when you are building some particular use case. So depending on the use case that you have, if it is really a simplistic use case which will need simple machine learning or simple algorithms like classification or regression, or where you want to do some kind of grouping or some kind of classification of these different things, then you directly go into the middle box, where we are talking about the ML Library, which uses the embedded mechanism, where you have two different ways of doing it. You can leverage the HANA-based PAL algorithms or HANA-based APL algorithms. That is one of these approaches.
Then you have, on the right hand side box-- which is where you have some complex requirements or complex use cases, where there is high usage of the CPU or there is, for example, text recognition, or voice recognition, or image recognition kind of techniques that are needed. That is where we will have to get into the complex use cases, and we get into the deep learning techniques. And this where on the right hand side, you will leverage one of these approaches.
You build these machine learning services on the SAP Cloud Platform, leveraging either the AI Foundation Layer, which is already available, or leveraging the SAP Data Intelligence, where you can build new models. And then these services are now available, and these can be enhanced. These can be consumed by the SAP S/4HANA business processes. So you have seen how you embed-- for the simpler use cases, how you can take an embed into the central part of it, where you embed using HANA APL and PAL. On the right hand side, you will see that for the complex use cases, where you need the deep learning techniques, you use not only the SAP algorithms but also the non-SAP algorithms, like your TensorFlow, or Python libraries, or scikit-learn. So the complicated use cases, you pick up these and build these scenarios or these services outside of SAP S/4HANA.
But these are, again, consumed by S/4HANA. And on the third side, on the third approach on the leftmost side, you will notice that there are, again, a different requirement where you would not want to realistically embed into a particular business process, or at the same time, it's not a very complex requirement. But you would want to provide a quick understanding for the business user. Like, you have a particular business process running in SAP S/4HANA.
Now you want to leverage this business process. There is already a CDS view available for this, a whitelisted CDS view. So you could take this whitelisted CDS view. You can create a custom CDS view on top of that, and you can put it into SAP Analytics Cloud using the configuration mechanism. And then you can build additional-- using the APL algorithms in SAP Analytics Cloud, you can build quick predictive visualizations of simulations using the modeling techniques there. And I could either-- you can quickly leverage the dashboards and the BI screens available in a SAP Analytics Cloud, or you could also publish it back into SAP S/4HANA.
So the idea is that for quick understanding for the business users, you might leverage SAP Analytics Cloud using APL for quick run scenarios. And then this is more like a template, but if it becomes much more useful for the customer, or the end user, or the partner, then you could try to embed it using the first approach, embed it into the business process itself. So that is the overall decision tree at a high level, how we are kind of trying to address these different ways of doing predictive analytics and machine learning. The next slide, this is the slide. It is an extension of how we are doing these three different ways of building predictive analytics and machine learning. You could see, like I've explained earlier, you could embed-- you start with the use case idea, and once you have this use case idea, you can take the data, you can collect the data, prepare this data.
And then you could design these different machine learning models using the different algorithms. And it depends whether you are embedding into the SAP S/4HANA, or you are consuming from SAP Cloud Platform. So depending on that, you will design your machine learning model. And based on that, you will kind of-- then the next step would be, once you design your machine learning model, you will go to, how do you train this predictive models and increase, depending on the confidence levels and the-- of the model that you have delivered. And then finally, your use case is delivered.
As you can see here in this screen, there are four different screens we are showing. One is talking about the predictive stock in transit scenario, where you are trying to predict if there is a delivery delay for a particular stock which is being moved from one location to the other location. This is the top-- on the top of the screen that you have.
So in this scenario, what happens is that we are leveraging an embedded scenario. The predictive algorithms are built into the business process directly. Then the next use case and the next-- you have seen the invoice payment forecasting. This is a different use case where what we are doing is that we are leveraging the SAP Analytics Cloud APL algorithms and trying to provide the business user a quick understanding of how this would look. So this is using a different technique.
Whereas the bottom two use cases, the recommendation and IRIC corporate, and then the other use case here, we are-- for extending the contracts, we are leveraging the different mechanism of using the consumption of the machine learning. So in that way, you could-- this particular slide, you can explain-- you understand how do we start with a use case, how do we create the data and prepare the data, and how we design the machine learning models, and then later on, how do we train the models, and then how do your end results look like. This is an extension of the earlier slide, with the technologies that are needed behind-- that are used behind this process. So in the first model, when we have moderate AI requirements-- let us now go into the use case. Now, you have a particular use case.
You know there is an existing machine learning solution or a service available for this particular use case. The model, you can adapt the model, and you can do the proper configuration. And as you see on the top, you can directly train your model. And then you can do the-- based on your confidence levels and the model fine-tuning, you can deliver the results.
So this is, again, consumed in an ML-enabled application. That is for a existing machine learning use case or a particular machine learning functionality you want to leverage and that is already built. Now, there are scenarios or situations where you want to create new machine learning models.
So on the bottom, you will see the second level. You could also create a new ML scenario, and the service can be developed. So what we'll do, you create the new ML scenario. You could create it as an embedded model into the SAP S/4HANA application, or you could create it as a side-by-side model, or we call it a sidecar approach, where you can create the model outside of SAP S/4HANA in the SAP Data Intelligence.
Whether you leverage-- whether you build a model in the SAP S/4HANA system or use a sidecar approach, we are leveraging the ISLM technology, which is called Integrated Scenario Lifecycle Management. When you're embedding the models into SAP S/4HANA, you are leveraging the HANA-based APL or PAL algorithms, and this is, like I've explained earlier, for moderate AI requirements. For complex AI requirements, you will obviously have to leverage the data intelligence platform or the AI foundation platform that you have and build this models using the algorithms which are maybe a mix of SAP algorithms and then non-SAP algorithms from TensorFlow or Python library or scikit-learn. And then the corresponding ABAP classes or the corresponding ABAP CDS views are built. And then you train the models, and then the application gets-- the ML service gets consumed by the application, which is sitting either in SAP S/4HANA, or it can be a third party application running on SAP Cloud Platform. So the overall idea here is that wherever the model sits, you can leverage the ISLM technology in building this.
All right, let us now continue to the next section, where we will briefly discuss the methodology, the tools, and technologies that are behind these different models that we have so far talked about, the different approaches. Before going there, we want to kind of understand the technology in detail. We have-- kind of looking into these three different ways of doing predictive analytics and machine learning. So let us do a quick comparison of these approaches, the side-by-side, the sidecar approach, versus the embedded approach.
In the embedded approach, let us look into the capabilities. In the embedded approach, what is basically happening is that your application obviously is sitting in SAP S/4HANA. Your machine learning algorithms are sitting in SAP S/4HANA, and your application logic also sits in the embedded mode in SAP S/4HANA. So overall, the CPU time is much lesser.
It is not very data-intensive, and the use cases are literally embedded into the business process. So in this approach, predominantly, what we are doing is that to make the business process end-to-end run much more smoothly, we are leveraging the HANA-based APL or the PAL algorithms. And there are about 90 to 95 or more than about 90 flavors of these algorithms which are available, which you could leverage in building these. Now, the main difference when you go from an embedded mode into a side-by-side, the sidecar approach, is that in the sidecar approach, what happens is that your application, again, can sit either in the SAP S/4HANA box, or it can be running somewhere on your SAP Cloud Platform. But the idea is that the business logic and your data and your algorithms, they are sitting on-- most likely on the SAP Cloud Platform. And in there, it might be sitting in the SAP Data Intelligence layer or in your AI foundation layer.
And here, what is happening is that we kind of-- depending on the requirement of the use case, when you have to use some deep learning techniques, when you're running through your image recognition or voice recognition kind of scenarios-- for example, there is one of those use cases, where you take a picture. A procurement specialist could take a picture, and based on this image search, he can identify a proper match. And he can create it a purchase requisition and a purchase order in the system.
So for that, what happens is that you are leveraging the image search mechanism, where you have to-- the system has to kind of run through all the images in the system, and then map to the particular system image, and then kind of come up with that. So for that, what happens is that you will have to leverage some of the-- the data that is needed is much more larger, and it is much more CPU-intensive. And then, now, the data sets are also huge. And then the system also learns more, and you have more amount of data. So for these kind of approaches, we are leveraging the sidecar approach, which is where your data resides in a SAP Cloud Platform. And then you could also push additional data into the SAP Cloud Platform and then address these things.
And your application, your machine learning algorithm is also sitting on SAP Cloud Platform, either in the Data Intelligence platform or on the AI foundation platform. Now, what happens is that you will have to build the additional application logic. You build the machine learning algorithm. Using the machine learning algorithm, you build the application logic.
And this application logic can sit either in the SAP Cloud Platform or in the SAP S/4HANA, depending on where you're leveraging this use case. And then between SAP S/4HANA and SAP Cloud Platform or the SAP Data Intelligence, which is sitting on SAP Cloud Platform, you will have to do the basic configuration between your SAP S/4HANA system and your SAP Cloud Platform system. So that happens.
And then you will handle it. The ML service is built, and then the consumption of this ML service, it is consumed by your SAP S/4HANA business process. And now, in the third approach, the sidecar approach, where you leverage SAP Analytics Cloud, which, again, runs on SAP Cloud Platform, this is where you're leveraging only the APL algorithms, which are, again, a quick way-- the HANA APL algorithms, which are available as part of SAP Analytics Cloud. Here, what we are doing is that we are not embedding the predictions into the business process.
But we are just quickly taking a business process, take the output of this or take a user story from there. And now, you can add additional simulations or predictions using that by creating a virtual data model in your SAP Analytics Cloud and then build the predictions. Here, in addition, the advantage, the additional advantage with this sidecar approach using SAP Analytics Cloud predictive modeling, using APL, is that you could also leverage data sets, not only from SAP S/4HANA but also from non-S/4HANA sources, like your SuccessFactors, or Ariba, or other peripheral networks, or even from outside data sets. So this is-- again, it helps the business user to quickly create a data set from these different data sources, create a virtual data model, and then create some prediction.
And these predictions can be leveraged and understood, and then you can either push it back into S4, or you can leave it that way for analysis. So these are the different ways of the technologies that are here. Now that we looked into the technology behind these three different approaches, let us quickly see how these things will look like, these different approaches in the system. Since we are doing a virtual TechEd webinar, I do not want to get into a demo kind of a thing, where it might be a little more complicated to show all this. So that is the reason I wanted to kind of give a quick idea about how it looks in the system. In the embedded predictive approach, your predictive analytics algorithms and the machine learning algorithms are embedded into your intelligent business processes into S/4HANA, where on the top, you see-- you log into the system.
And then you could see that you have your Fiori launchpad. You click on one of these apps, the Intelligent Scenarios app, create an intelligent scenario. And once the intelligent scenario is built, you build-- you leverage the Intelligent Scenario Management, and then you embed it into the business application. It is different, where in the sidecar approach, again, you get into the system.
You understand you are now building the machine learning models. You select whether it is an embedded approach or a side-by-side approach. When you click on the Side-By-Side Approach, it takes you to the screen here on the bottom that shows the launchpad, the Data Intelligence launchpad. And from there, you can create your models. And then you can-- since your systems are already configured, you can continue building that.
In the third approach, leveraging the SAP Analytics Cloud, what we do here is that you log into the SAP Analytics Cloud system, and again, your configuration between-- like in the earlier model, where you have configuration to be done between SAP Data Intelligence and SAP S/4HANA. In the same way, in this approach as well, you will have to make sure that the basic configuration between SAP Analytics Cloud and SAP S/4HANA is already-- the prerequisites are already entered. So these are configurations that are already performed. And then based on that, now you can come back.
You can leverage your custom CDS views that are available in S/4HANA, take these custom CDS views and build a predictive scenario. And then the data sets are also acquired from SAP S/4HANA, or in some scenarios, you could also do a live connection. And then you can build your predictive scenario, picking up one of these algorithms.
And then you can visualize the end results in SAP Analytics Cloud. And there are also situations where you can publish it back into S4. So with this, what we have done so far is that we have understood-- we have given a very high level overview about AI in the context of SAP software. We understood what are the things that are needed in an enterprise AI and how we can address the customers and partners' problems. And in the second section, what we have done is that we understood at a high level what are the different approaches that needed in doing this predictive analytics and machine learning.
And we've also briefly discussed about the decision tree, about which particular approach you use at what instance, in that particular section. And we also looked into the process behind that. And in the third section, we kind of understood, or we kind of done a little bit of a deep dive into these tools and methodologies behind these different approaches. So again, there, we started with a little more technical details about the different comparisons or how we compared these different approaches. And then we kind of went into the next-- we kind of gave a quick glimpse into the system, how we start in creating these scenarios.
Of course, you will have a lot more resources available. And we are also building up a best practices package, which is where we will briefly touch base on that and what are the different use cases that are available. So now that we finished those three sections, now, in the fourth section, what I am trying to explain is that there are some use cases and best practices we are still creating. But there are some use cases that are already built, leveraging these three different approaches. Predominantly, we have leveraged the option one, embedding into S/4HANA, leveraging the HANA PAL and APL algorithms.
Earlier, we used to use the Predictive Analytics Integrator framework, which is now upgraded to a ISLM framework, which is Integrated Scenario Lifecycle Management, leveraging HANA, PAL, and APL. And then we have used some of the other use cases on the AI foundation layer, which is where the ML services are built on a SAP Cloud Platform. And these are now used by-- consumed by the SAP S/4HANA business processes.
And then there are a couple of use cases where we are also-- we briefly talked about how you could leverage SAP Analytics Cloud and the HANA APL features in that. So there are these different use cases that we have leveraged in the context of how you can build. We kind of do not have the use cases yet on leveraging data intelligence on SAP Cloud Platform for the consumption of these business service, machine learning services. But we leveraged the AI foundation layer on a SAP Cloud Platform when we talked about the sidecar approach.
And there are a lot more things coming up wherein you could do a lot more things on how you could embed predictive analytics or machine learning in SAP S/4HANA or building it outside on SAP Cloud Platform, or SAP Analytics Cloud, or Data Intelligence. We have also-- that is the reason, I know, in the context of this virtual TechEd, we could not leverage a lot more in explaining the customers and the partners. But there are a few beautiful blogs that we have put across.
There is this blog series which talks about all these different things that are available. You could follow the blog series, which talks about how you leverage predictive intelligence, and there are other blogs. Along with that, me and my colleague, we also have put a book which is released last month, September 24, which predominantly talks about how do you implement machine learning with SAP S/4HANA. And that is the book which gives you not only the architecture details or implementation details but also the business implementation and all that in the context of what we talked about. Now, we are also coming up with a best practices package which helps the customers and the partners understand all these in the context of how you could leverage these different machine learning approaches with SAP S/4HANA.
So this will give the customers and the partners a very good understanding of where we stand. There is also a expert panel session where me and Siar will be taking some questions. Thank you, everybody, for attending this session.
We really appreciate your time during these strange times, during these strange pandemic times. I know it takes a lot of effort to get onto the meetings or get onto the Zoom calls to attend these. We would really appreciate that. We could-- if you have any questions, feel free to reach out to me or my colleague, Siar, and we will gladly address that. So you could reach out to us on the blog series or through these lectures. [MUSIC PLAYING]
2020-12-21