Business Automation with Artificial Intelligence and AI Business Services | SAP TechEd in 2020 ST

Business Automation with Artificial Intelligence and AI Business Services | SAP TechEd in 2020 ST

Show Video

Hello and welcome to the session today, which is all about business automation with artificial intelligence and AI Business Services. My name Jana Richter and I'm the Chief Product Owner for Artificial Intelligence within SAP. And within the next 30 minutes, I would like to spend with you some time on the topic AI in an SAP universe.

So what are we going to do? First of all, I would like to have a quick look with you on the topic artificial intelligence and how does maps to SAP processes and the Intelligent Enterprise. Secondly, what I would like to focus on in today's session is the topic SAP AI Business Services. AI Business Services are really a concept and service that we just started to introduce last year and where I would like to convey what's the concept behind of them.

Highlight some of them combined with customer examples and a quick demo. Let me start with the whole topic artificial intelligence, what is it and how does it map into an SAP world? And to start with the basics I would like to discuss what's artificial intelligence. And I think that's something that meanwhile accompanies everyday life of many people. So every one of you probably has an understanding and there's a variety of definitions out there. My personal favorite one is the one from Patrick Winston, who is a professor from MIT, and he said that today's artificial intelligence is all about new ways of connecting people to computers, people to knowledge, people to the physical world and people to people.

So the basic underlying assumption behind this is that AI today has the main purpose to change our way on how we interact with machines. If you think a couple of decades back this was why probably he even got this kind of a geeky image out there, people really had to think, how is a computer working, how is this machine working? What can I do to achieve a certain result? And meanwhile, it's the other way around. We really try to bring computer software, the whole physical world, into an approach that we understand how humans interact and that machines really adapt to this and that even the thinking and the whole thought process, if you will, of a machine is very much following how humans interact. So what does that mean for typical SAP business processes, SAP software? If you've seen a couple of sessions here, you're are probably pretty familiar with this picture. And I'm not going to give you the general pitch of the Intelligent Enterprise, but I would like to rather point out where does AI fit into this picture and well, as you could probably already tell by the title Intelligent Enterprise this fits into more or less every layer. So if we start in the top with business processes, well SAP really aims to infuse AI into all the business processes with the basic idea to rethink how those processes are run, to give you a higher level of automation, a higher level of efficiency, a whole new experience, new insights that you get unlocked by combining information from various sources and applying machine learning models on top, which even allow you to run your business totally differently to give your customers a different experience.

And this, of course, spans the whole set of processes and eventually is, of course, reflected within the product portfolio, which you can find in the middle. So that's the application layer with the keywords intelligent suite. Intelligent suite meaning that into products like SAP S/4HANA, Concur, SAP SuccessFactors we make machine learning an integral part and infuse artificial intelligence into various components, into various functionalities. And this is done by means of using the Business Technology Platform as the lowest layer underneath which you can see on the technology side. If you look into the Business Technology Platform, then this is split into four different sections, where on the right hand side you can see the intelligent technologies with artificial intelligence being in place.

But as you can imagine, this is not a super clear cut as well, because for any kind of algorithm, the whole topic of having the right data at hand to be able to train the machine learning model is pretty key. So we have a very close relationship to the whole topic of database and data management. And typically it's closely aligned as well with analytics because displaying the data and then applying machine learning models on top that give you additional insights, additional automation is very closely intertwined as well. And all this, of course, should be reflected with application development and integration into the SAP processes and into the SAP product portfolio on an applications layer. So although it's on the right hand side within intelligent technologies and, we can open this up, there you see artificial intelligence and machine learning appearing, of course, it's very much closely connected to the whole Business Technology Platform suite, if you will. Well, with that, I already finalized the generic artificial intelligence pitch and the generic explanation of what AI is and would like to deep dive with you now into the topic AI Business Services.

Let me give you a quick definition of what AI Business Services are. Our basic ideas after we run a couple of projects from a technology perspective as well with our SAP applications and our SAP processes, is that we see common patterns across the different processes and across the different SAP solutions where we can apply machine learning models in a similar manner and where we want to give us customers and partners out there, those capabilities at hand as well and say, well, those are typical, let's say, puzzle pieces, really typical capabilities where machine learning can be effective and can help. And we don't think it's effective to each and every time to create a new AI algorithm, to have a new thinking of how to tackle this problem. But we rather say, let's provide those strategic machine learning capabilities that typically automate and optimize processes or as said beforehand and enrich the customer experience that's really provide those capabilities for reuse.

So we decided to provide them as reusable services within SAP Cloud Platform so that you as customers and partners similar to what the SAP application development teams are doing, can use those puzzle pieces as part of your own development project, as part of your environment, which you use to extend SAP standard products and SAP standards processes with your own logic and with your own capabilities. And you can see on the lower side of the slide the business services that we already are providing and we can show a very large extent, cluster them into two bigger groups. The bigger group that you see on the left hand side is the topic business document processing. And business document processing means that we really saw a large driver around automation with artificial intelligence in the whole space of business documents. Well, why is that the case? I mean, in many corporations, meanwhile, the paperless office is reality.

Of course, here and there, we still see written and printed invoices, letters, whatever, coming up. But to a very large extent, things that we used to print 20 years back, meanwhile are digital. But it still means we have those documents. Just the format changed. It's not a physical paper anymore that I'm having anywhere, but it's rather a PDF document, an image where then I can find the information like an invoice, a sales order, payment note. So all those kind of information still exists in a document. But it's a digital document, meanwhile.

But what we see as well is that still a lot of cumbersome, repetitive, labor intensive work is all about handling those documents, understanding what is this, where does it belong? What are the entities that I find on this document? Where do I then place them into a transactional solution like into an S/4 system to start a whole process? So this is really a place where we see a large potential for reuse by looking at the business document processing services, which namely our Document Information Extraction. So it's all about extracting information from unstructured documents, like if you have an invoice to automatically extract the invoice number, the document date and the total amount, text amount and so on. Another one is Document Classification, which is all about classifying unstructured documents and really putting them, for example, into a hierarchy or into different categories by using a machine learning model. Here, the same logic applies. We have a machine learning model which really helps you based on historical data to classify the documents automatically. And it's not a human person who needs to read through a whole document and then understand, for example, let's think of contracts, this is a contract for a certain region, for a certain customer group with certain information inside, but then an automatic service can take care of this.

Maybe one fundamental difference between them - and I don't want to go too deep for document information extraction - we can already provide pre-trained machine learning models for you. This means you can use a machine learning model where we have a large dataset behind and where we can say, well, this is already optimized for document types like payment advices or invoices, for example. And you then have an option to extend this for your own document types as if this is something that SAP didn't provide so far. Document classification in contrast, is a service where we provide the service itself and the machine learning model. But you need to train this with your own data because here in this context, we see a high flexibility and variety of what customers and the different processes as well would like to classify and how the whole hierarchy should look like there.

So this is something that you train with your own data. Business Entity Recognition is a service that helps you to extract really named entities, business entities from unstructured text. So it's not necessarily a PDF document as an input, but for example, email text where you can then locate and classify the relevant information.

And this comes with some pre-trained models and with the option to train your own custom model. So far for the business document processing space. And later on, I will show you document information extraction a bit more in depth with a customer example and with a demo. The other services Data Attribute Recommendation, Service Ticket Intelligence and Invoice Object Recommendation are more on a transactional side. So looking at common SAP business processes and providing some capabilities on really handling data and getting the right insights and for both data Attribute Recommendation and Service Ticket Intelligence, I have some more details later on. The one that I would like to point out because

I don't have any in depth material for the short session here today is in Invoice Object Recommendation, which is really there to automatically recommend general ledger accounts, cost centers for invoices without order reference. So whenever you have an invoice where you don't have an order up front that you can very easily identify, what's the cost center in the GL account that I need to book this to based on your own machine learning model? again, because this is, of course, something which is very customer individual which cost centers standards you have, which GL accounts are based on which historical data, the recommendations can be done. To come back a bit to the bigger picture, I gave you now the whole picture of AI Business Services and what are they but how do they fit into from a technology perspective, into the different AI technologies that are available within the SAP portfolio? Let's start on the topmost layer, which is artificial intelligence integrated into SAP solutions, or what I refer to in previous slides as Intelligent suite, Intelligent Enterprise on an application layer. So this means really things like Cash Applearning recommendations in a SuccessFactors context. So meaning here you already get the services not as services themselves, but already baked into the SAP solutions.

And if it's really well done to some large extent, this is even a challenge from a positioning/ product management perspective, you might not even notice that this is just happening with artificial intelligence, but you just see recommendations that fit for your use case. You see a higher automation and this is powered by machine learning, but it's making the things so smooth and giving you the additional insights. It could be artificial intelligence, but it could be something else as well, right? Well, as we just said, the layer that we carved out of those projects is really the AI Business Services layer with the reusable services. And of course, our business services are built on a technology platform itself as well. So the two technology platforms that SAP primarily has is, first of all, data intelligence and the whole topic, Machine Learning Foundation, AI Foundation, which is all about really on both sides, bringing data together, running machine learning models, providing you with options like training inference.

But this is if you run an AI Business Service, more or less the underlying technology, but not directly exposed to you. But of course, you also have this platform layer in case you, with your data scientists, your own use cases, would like to provide own machine learning models. If but next to that is the whole topic of conversational AI, which is all about providing chatbots, typically in an SAP context. And for the AI Business Services that we we're focusing on today,

really our target persona and our target audience is the developer community, meaning you can use those services without necessarily having data scientists within your company working with them, but we really encapsulated the machine learning functionality into a service provided in Cloud Platform that you can consume and your developers can use those machine learning services without extensive data science know-how. But of course, with the impact that we started to a defined use case and with some expected input that we commonly see in this context. And here the machine learning models are already optimized for this use case. So meaning we have everything optimized for the data type, the data size, the right preprocessing logic in place and so on. And all the pipelines that are running behind the scenes are really optimized for the business outcome. Meaning, for example, when this is suitable, we provide pre trained models that already help you to get started fast on common datasets that we typically see across the SAP customer base, multiple pipelines combined if this is needed to achieve a certain outcome.

And what we do because it is a cloud service, is that the whole deployment, monitoring ongoing operations, support and so on is of course provided by SAP. And you can interact with the stable APIs that are available for training and inference, including all the documentation and assets that you get around. So to make this now a bit more practical, what are we talking about? We have a couple of pretty cool cases that I already mentioned beforehand I want to start now with Service Ticket Intelligence. For Service Ticket Intelligence, the typical common business challenges is that many companies have a high volume of service tickets that need to be processed and customers are waiting for their requests, like where's my shipment? What's the status of my payment? And people need to respond to those. And there's high expectations, meanwhile, from customers, first of all, about a quick response and timely response not only should be there, but we want to have the right resolution, and not somebody asking some very basic questions. So really a good interaction experience and a high efficiency around us. And Service Ticket Intelligence really helps

you with machine learning models to automate this customer experience. We provide a machine learning model that helps you to automatically classify tickets. So meaning you already know, what is this about? Who would be the right processor or the right processing group? And we can provide to those processors in the customer service side similar resolutions or similar tickets. So if there's been similar inquiries in the past, you can find related inquiries and the resolutions and thus probably have a quicker, timely response to the customers that lead them to the result that they want to see from the service agents. And that really helps you to reduce repetitive tasks of manually classifying who's the right processor for a ticket. We accelerate with this the response times and increase the customer satisfaction because they get responses quick and with the high accuracy, they get a response that helps them and then eventually, of course, saves cost for your customers' service, but helps you to really engage with your customers and to have a good experience for them so that they will be happy and come back to you and not be annoyed by long processing or customer tickets.

We have a variety of customers here already, and one that I would like to point out is DuluxGroup. So they are a marketer and manufacturer of premium brand paints and coatings, and they had exactly the goal that I mentioned beforehand, they wanted to streamline the customer support process and meet the high expectations of their customers. And they implemented Service Ticket Intelligence and can see now 70 percent automatic ticketed classification in a five times improvement to the first response time. And this, of course, goes hand in hand with a faster onboarding for the contact center representatives because they get more information at hand that helps them. So I think this is a really cool service that helps you on the customer support front. Another service that I would like to introduce to you is Data Attribute Recommendation and actually Data Attribute Recommendation is one of our superstars in the portfolio because this is a very flexible service that is very efficient for a variety of problems and can really help you there.

Basically, it classifies entities such as products, stores, users into multiple classes using pretax numbers and categories as input. And you can see one of the use cases where they can attribute recommendation is very efficient here, which is the whole automatic prediction of missing master data fields for new products. We see this quite frequently that customers have a challenge with new product releases to the market because you might lack the holistic view of the master data and the different business functions. And over time, people entered information that might be very inconsistently represented across the entire organization.

And Data Attribute Recommendations provides the master data field prediction to the business user. So we have here a retrainable machine learning model as well. So again, this is something that you train with your own data, targeted them for your own use cases, and it can be trained very frequently. So in case you have changes in your organization, you will see them reflected.

And this really helps you to predict custom fields, work centers, asset maintenance, and thus really allow that people do less mistakes. So it's a reduced manual maintenance effort. It's one voice of master data because it's consistently maintained and that overall then improves the quality of your master data. And even here, a pretty cool scenario and use case is from Severstal. Severstal is the second largest steel producer in Russia and number 15 in the world. So this is all about steel and steel related mining.

And they had a height or a significant challenge that they had high effort for business users to order new spare parts, a high number of returns because things were chosen incorrectly. And it's been quite time consuming to manually create master data and to manage this. And by implementing data attribute recommendation, they decrease the effort for new material creation by 20 percent, suggested appropriate fields to help ensure that the users really feel this properly and not put in something which is very inconsistent with what they had.

And with this, they reduced the master data management efforts as well to a very large extent by using machine learning. Good. The last service that I would like to show is Document Information Extraction. So this is all about business document processing.

And one of the use cases that is very prominent in this context is accounts payables invoice extraction, that can be - I already explained - quite consuming to handle incoming invoices, Really understand what's the entities that we have in there put this into a transactional context and do this accurate. Right. So they might be quite some turnaround times here and productivity losses. If this is not done properly, and when using Document Information Extraction, this unstructured data from invoices can be extracted, can be structured, typically enriched as well, to give you the right information that fits the business process. It might have to be enriched.

And this really helps you to reduce manual work, speed up the processing and improve overall the accuracy and optimize even payment strategies. So this would be then if you have an invoice reconciliation process at the end. Here I decided not to take again a customer example, but rather say this is one of the examples which is used, for example, within Concur invoice. So this is really one of the SAP solutions in the context of travel and expense, primarily for invoices where we consume Document Information Extraction service behind the scenes made this an integral part of the offering. And thus 3000 Concur customers process more than two million invoices per month. So I have to say, this year during the current COVID times, probably the number is slightly lower.

But this is like the average of 3000 customers processing a really large amount of invoices and really reduces your average cost of processing the invoices by 80 percent because you can just upload the invoice and get the entities extracted instead of manually filling a lot of fields in addition to the document that you uploaded anyways. As promised, let's do a quick demo here and let's look at Document Information Extraction. And I want to show you one of the recent highlights that we provided for the service within Cloud Platform that you not only work as a developer with APIs, so APIs are there since quite a long time already, but we now added the user interface that makes the interaction quite some fun when you explore the service and see what it's doing. Let's get started within Cloud Platform trial. So everyone one of you who worked with Cloud Platform probably knows the cockpit and might have seen a trial account already. So this is exactly the environment where we get started as well and now enter our trial account and let's go into the global account of our choice.

And we can see in the section entitlements, the variety of services, and you can see Document Information Extraction trial and the according user interface already within the entitlements. And now going to the subscriptions, we see this user interface and the application that we already subscribe to and that we want to work with. And here we can see a UI and we learn over the time that people really like to interact in a way that they upload the documents, for example, by a drag and drop from their local desktop. So something that you're, of course, familiar with from various other products.

So let's upload one of our demo invoices. This is an invoice. So we pick this document type and in the next step we can decide which of the header fields and then which one of the line item fields we would like to service to extract from the document that we are uploading. Having done so, the process will start when I click on confirm and maybe to note this is an asynchronous process. So what we will see is that the document is pending for a few seconds. And once it's ready, you see in the user interface as well, that the document is now marked as ready and we can review what it has done. And we get started with the PDF document and we already see the blue bounding boxes where the service really extracted and found entities like here, the sender address, or you see on the right hand side things like invoice, date, purchase, order number.

So really all the entities that it was able to extract successfully and detect. And you not only see the entity marked, but of course the structured result as well with the OCR result. And we not only have the view on the PDF document, but we can see the same in a tabular format on the right hand side for both the header fields as well as the line item fields. And something that I would like to show you, as well as an option to correct the output.

So if we see, for example, material number of units would be something that I'm interested in, but for whatever reason, it was not extracted properly, we can edit this and you as a user can then decide to draw some of those bounding boxes yourself. So first of all, everything that you see greyed out has been extracted by OCR. So the characters were recognized. And now I can really say which fields belong to which part that I'm marking here. So you see, this is super convenient. To just pick the entry, see what OCR extracts in terms of text characters and then mark which entity this represents. You can confirm this change.

And once you've done so, this is really the result that you have from the service. Well, with that, of course, I would like to say you've seen the UI part now in a very realistic scenario, meaning you are not only getting started with the service, you are not exploring the service or looking into a correction because you see certain things are probably supposed to be corrected. In a typical scenario, of course, the interaction will work on an API level then. Right, because if you as a developer stitch this into your business processes, into your own logic, you probably like this UI to explore the service and provide this to business users. But as a developer, you work with the APIs and make this part of your overall process, similar to the picture that we've seen beforehand.

With that, we are already close to the end of the session. So I hope you enjoyed the roughly last half hour and I hope that you really learned about the benefits of SAP AI Business Services so that you have a fast time to value with enterprise ready services that could instantly show you which value you can gain in your processes from artificial intelligence, benefit from the innovations that SAP brings to the market, and reinvent your own business processes and your own custom development logic. And with that, have an easy way to enter the world of machine learning that gives you all the benefits of automating, optimizing processes, really digitally transforming and thus really having a significant impact on your top line. With the basic idea that we want to encourage general reusability and usability so that you can solve concrete business problems for various business processes with services and puzzle pieces that you get out-of-the-box and that integrate into your own system landscape by using the services and best practices that we share and an easy commercial model and even administration model via Cloud Platform and Cloud Platform enterprise agreement.

With that having said, I hope you already see that artificial intelligence is really tailored within SAP for your key processes and entities. We provide something which is also ready for developers and not necessarily involves the data scientist on your side and flexible integration into your landscape with standard SAP tool. If you're are interested in more information, that's really a nice lineup up of TechEd sessions ranging from the roadmap down to real hands-on sessions where you can try out business services like business document processing services, Document Information Extraction, our Data Attribute Recommendation service and information around offerings around like the Shared Service Center and of course, the communities, the help pages are always advisable as well. And in Discovery Center of Cloud Platform, you can find a lot of related information too. Overall, from TechEd continuing your learning experience by joining the SAP Learning Hub, where you can really deepen and validate your SAP solution skills. Here's a quick link as well with the QR codes to the different services that I quickly touched upon today.

Thanks a lot, everyone, for attending the session. And have a nice day. Bye bye.

2020-12-20 09:09

Show Video

Other news