Machine Learning a “Top Priority” at SAP

Their CEO Bill McDermott expects intelligent applications to fundamentally change the way we work.

They are already changing things inside the software giant it seems.

Dr. Markus Noga is vice president of Machine Learning at SAP. His global team focuses on improving SAP products and services through leading research projects and innovative solutions of startups. Markus has been in charge as VP for SAP’s portfolio as well as leading the company’s corporate strategy group. He spoke exclusively to us about the company’s vision for an AI-charged back office of the future.

TechNative: How much is a focus on machine learning dictating your R&D and the next wave of solutions from SAP?

Markus Noga: This is one of SAP’s top priorities for the years to come. Obviously, the new developments since 2012/2013, have been accelerating real world usefulness of machine learning and just creating a next-phase of value creation for our customers. So, with everything we do, we look at how can we make solutions more intelligent and drive more of the action and traffic more to the work itself, rather than routing a workflow to a person. And, how can we extend the benefit and the reach of this into areas where it was currently not economical to have a person do the task and thereby the extend the coverage, the reach and the business impact of solutions.

SAP’s Dr. Markus L. Noga

I think that over time, it is the now the new normal for enterprise applications and people simply expect transactional tasks to be progressively automated. I was in a meeting with a European Union Representatives last week and one of their leaders for education actually shared an education vision for Europe. We will need to prepare our workforce for a future where unstructured problem solving is the most important skill because we can count of transactional tasks being progressively automated over time. I think that’s a bold vision and one where I applaud our European leaders for their farsightedness. And, it’s one that we, as enterprise software vendors, also have to keep in mind in the way we build software.

TN: What’s different about SAP’s vision for machine learning?

MN: We’re pursuing an application lead strategy that’s driving by real-world business use cases. We’re trying to bring value to our customers, not with a tool box of building blocks that can be adapted to any particular scenario, but by leading with business solutions for the horizontal cross-industry cases that drive the most value for the most customers. We’re pursuing this in a technology-agnostic way. From our point of view, the value really comes from applying available technology to solve real world business problems, not from trying to be two or three percentage points better on any underlying fundamental tech. Current trends in the industry, like Google’s bold move of open sourcing their TensorFlow machine learning framework entirely, and even needing startups doing the same for their framework, clearly points to a direction where some of the underlying technology is moving in an open-source direction very rapidly.

Our approach of solving the business problem is one that’s part of the overall vision and mission of the innovation center network that my team and I belong to. The innovation center network is about pursuing adjacent new markets as well as disruptive technologies that we can bring into SAP’s core. And, with machine learning, we’re certainly facing such a disruptive technology. Integration center number is based in locations on almost every continent and we’re hiring actively and if you are a machine learning expert or passionate about bringing business applications based on machine learning to the next level, we’d love to hear from. Visit our page at

TN: Where do you see machine learning and AI being disruptive in the enterprise of the future?

MN: I think that we are on a journey from lab bench to production to large scale enterprise production with machine learning. And, over the next five years we’re going to see significantly increasing acceptance of the types of solutions that machine learning can provide. They are new and we are in many ways uncomfortable to traditional enterprise IT departments who strongly rely on certainties, who strongly rely of the predictability of given solutions, whereas everything machine learning does, is my nature stochastic. It can insure an x percent matching rate. It can insure a 5% improvement over the business performance over a given baseline. But, you can’t quite say how it’s going to handle every individual case.

So, I think over the next five years as we go from lab bench to large scale production deployment and pervasive use in large enterprises, this mind shift change from certainties to probabilities and fractions is what’s going to be the most noticeable thing for businesses. This is also where large enterprise vendors can help by making the journey a smooth and a painless one. And, by bringing in our substantial user interface and user experience expertise in making the overall systems of humans working in an enterprise organization and process constellation with machines taking on certain tasks. Of making that overall system of systems, work reliably and create confidence in the types of predictions and the types of recommendations that machine learning solutions bring to the human experts so that they can come to trust, depend and rely on the help of the machine.

What is machine learning’s killer enterprise app? (Image: Adobe Stock)

In many ways, I think it’s parallel to the impact that robots have made on manufacturing assembly lines.Today, we have found a very good, strong balance that leading manufacturers between the sorts of the things that humans do exceedingly well, sorts of things that machines do exceedingly well and operations discipline of designing the best possible shop floor environment.

TN: How will machine learning impact your popular financial solutions?

MN: Our financial management solutions today touch the vast majority of business transactions worldwide. Whether it be sending or receiving and, the advantage of these solutions today is that they’re rock solid and really under the world’s commerce. We have the opportunity to make these solutions more intelligent and to take a lot of the tasks that are today being repetitively or manually done, in shared service-like environments and to progressively automate them with machine learning.

One of the examples that we’re working on first and foremost, is the questions of the invoice matching for receivables and help back payments coming in via electronic interface. You have receivables in your SAP system today, but because things never quite match exactly, still there’s a large share of shared services work in manual and tiresome work ongoing, like trying to find out who paid for what with these wire transfers.

TN: Will machine learning revolutionize the back office?

MN: We see that as a progressive journey where really the segmentation in transactional repetitive tasks can be machine learned. Because they have lots of variations, but they’re essentially the same thing over and over in high volume. Whereas the work is really creative, unstructured, solving new problems every day, dealing with the human interactions. Everything that falls into the transactional nature, especially if there’s training data in SAP systems for decades and decades of these tasks being already performed, are a strong candidate for machine learning to try and accelerate and automate some of the lower value transactional tasks here.

TN: SAP already offers machine learning through its CV matching to recruiters. What kinds of benefits have you seen?

MN: The fact of the matter is that shuffling the virtual equivalent of paper still occupies a lot of their time. And, they’re usually in one of many situations. It could be that there are posting for which they get thousands of replies and then it’s a scarcity problem, I can only afford to interview four or five candidates, we should know the right ones to invite.

It could that I’m getting fantastically strong candidates, but they’re not the right ones for this particular posting, but we feel strongly we would like to have them in the company. A large enterprise may not have visibility of the other posts that are there, especially with hundreds of recruiters working the case. Solving the hand over problem is also a critical one. And last, but not least, the capability to go out on social media and find candidates who haven’t applied yet, but who would be perfect matches for this position, can help in situations where they’re not an option for a particular posting.

TN: What other machine learning projects can you tell us about?

MN: That’s a tough question for me because I can talk new and upcoming functionality really largely only to the existing customer base. What I can say is that we have collected over 130 use cases, conversations with our customers without partners, with SAPs product and account management teams. They are all across the digital framework, whether they are in our digital core or our financial & logistics systems. Whether they are in customer space with sales, service, and marketing. Whether they’re in the business network with Ariba & Concord tying together procurement and edge finance. The people space with all the talent, recruiting, succession management and other HR functionalities provided by our Success Factors products are last, but not least, in the internet of the things and asset space where SAP has made a massive investment already last summer, to bundle all of our IoT efforts in a new product unit. We’re getting high numbers of potentially attractive cases from all of these. We’re continuing to evaluate them and we’ll be talking about our next wave of incubations when they are ready.

We believe that machine learning has the potential to drive massive new business value for enterprise customers because it has the potential to automate, semi-automate or support tasks that are currently being done manually. Our approach is to work very closely with some of the applications powering the largest scenarios in the market today. Our approach is to work closely with the application teams and to provide a native direct integration of new machine learning functionality into these applications to make the experience for the end user seamless.

Follow Markus on Twitter