 Live from New York, extracting the signal from the noise. It's theCUBE, covering RapidMiner Wisdom 2016, brought to you by RapidMiner. Now, your hosts, Dave Vellante and Jeff Brick. Welcome back to RapidMiner Wisdom 2016. We're here at the Aventi Hotel in New York City. Boris Scharinger is here. He's the director of IT audit at Siemens. Boris, thanks very much for coming into theCUBE. Welcome to America. Thank you. Good to have you here. So, very interesting background. We were talking off camera. You've done model-driven architectures, SOA, process management. I tell, we're here at a data science conference. What's an audit person doing here? That's a good question. No, I mean, audit really benefits from an analytical approach from a data scientist-driven approach. Audit is really something where it's hard to get opinion out of all sorts of audit discussions that you have if you do audits in many areas. They are interview-driven, or they used to be interview-driven, so data science adds so much value to our daily audit work by sort of replacing opinion by facts, working evident based, and that's how data mining and process mining comes actually into our daily work. So, Siemens, huge company, complex organization. When you think about auditing, you think about the bread and butter every day, financial auditing and reporting and things of that nature. So, talk about the scope of your activities that's obviously beyond that. Yeah, it clearly is. So, we are globally around 300 auditors worldwide, working really on a global base, and of course, this type of hygiene, this hygiene type of audit, financial systems and processes, reporting guidelines. That's an important part of our audit work, but not the only one. So, maybe it makes up 70 to 80% of our work, but then there is 20 to 30% of our work, which is really going out to business units, helping them to become more efficient, helping them to detect root cause issues in their process quality, in their throughput, in their lead times, in factories, and all of that stuff is really based on us being able to consult, to be responsive to their needs, to understand their processes, their business domain, and then make an analytical difference by a highly skilled, highly analytical stuff and the weapons, the analytical weapons of data science. So, are you a service organization for the lines of business, or are you a CEO, big brother? Or... We are a bit of both actually, and sometimes that's an interesting balance, but actually, by being service oriented, even if you're in a governance function, you can be very service orientated and you feel with the responses coming from business units that we can really be helpful and then we are regarded as being helpful. And your data sources are, I presume a combination of internal and external, but a lot of internal data, is that fair to say? Yeah, absolutely. I mean, just as a side note, Siemens has around 280 SAP systems out there, right? Which is an incredible number. Lucky us, they are pretty standardized, but that's an incredible source of transactional data, whether it's on sales, whether it's on procurement, whether it's on payments, whether it's on sort of factory scheduling and all of that stuff. And that's the main source that we use when it comes to efficiency and process consulting. And then obviously, if we, in our audit work, are going for fraud detection and stuff like that, we are using other data sources as well. Right, okay, so you've got this massive amount of data from all over the organization and you're applying data science to what end? What's the business objective? Is it to help optimize the lines of business? To identify, like you said, fraud? I mean, a lot of different use cases. Yeah, we have indeed a lot of different use cases. That's why our data analytical work in our audit department is big in size, actually, out of these 300 people, 70 people are in the IT practice of our internal audit. 70 sort of data scientist people. And we do all sorts of stuff, like fraud detection and others, plus this efficiency-related process. How much of your effort is on those types of use cases? Business outcome-driven, detecting fraud or other churn or quality, et cetera, versus applying data science to improve your predictive capabilities, if that question makes sense. Yeah, maybe that's a 50-50 thing. Yeah, okay. So a fair amount to each. And are you providing data science services to the line of business to help them improve their predictive capabilities? Or is it more... So far, they would be approaching us sort of problem-driven. We seem to have an issue here in that area, in that factory, with the quality of those products. So audit, can you help us? And then we would actually step in and help them. And typically, a big portion of our help is working out a strategy to approach it data-driven, evidence-driven, which then brings all the data science stuff. So very interesting. The concept is you've got visibility over the entire organization, you've got the data, so let's inject some data science skills and then give you that charter. Yeah, because you kind of answer, I was going to say, where did it come from? When did you move from doing kind of the financial audit and the majority of your business to helping in these other areas? Because I would think at first blush, they would look at you as kind of part of accounting and part of audit, and are we doing our thing versus being experts in the field of data science and being able to apply that to their problems compared to, as we saw in the keynote, all the old-school guys in that particular division that are working on intuition in the way we've always done it and their traditional tools. So did you see the opportunity and say, hey, we can help you in these? Or they say, we need help, we've tried everything, or is there some marginal economics that you guys can squeeze out that you see through your auditing process? Well, first of all, I think it has been a huge journey. Siemens had this corruption scandal in 2007 that when really a global audit organization was sort of founded and the focus after 2007 was really to get that sorted. And then maybe three to four years later, we are constantly, we started to move into this transformational type of activities. How can we really help business units to improve their efficiency and operational things as well? Right, right. And since then we are sort of going there and by now we have a reputation where people are actually calling us up and asking for help. And does some of it come from seeing change over time within a particular department, whether their improvement's going up, their efficiency's going up down where you start to see patterns and say, hey, we're seeing this is going the wrong way, can we help you? It's going the right way. Can we accelerate it? We saw in your sister organization, they're doing this, maybe you guys should do this. I mean, what are some of the dynamics? That's actually the requests for help are less driven by the coming out of the traditional audits. It's more that people hear what we did in other places of the company. They sort of understand the success stories. And then that's the moment when your telephone starts to ring. So you have a backlog of, do you charge for your services internally? Partially yes, partially no. Partially we are this audit department that is centrally funded. But in any case, we have a constantly a list of candidate projects, people knocking on our door. And at any point in time, for example, on this process mining area, at any point in time, we are doing two to three projects of such kind in parallel for some Siemens units out there. Do you... How do you prioritize those projects? Is it first in, first out? Or do you look at the business case and the ROI and the technical feasibility? How do you adjudicate? That's a non-trivial question actually because sometimes prioritization can also be driven by the opportunity to educate business units of what data science can do for them. So that's not a direct return on investment. It's not a direct quantitative business case in the first run. Saving potential is certainly a factor. But then also, can you plant some seeds that maybe will show their full potential in two or three years from now because you have actually educated a Siemens business unit to become data scientists themselves? So it's not a sequential picking jobs. It's not a pure sequential thing. Yeah. So following up on that question, what are some of the things that you've done with the business units who weren't necessarily data science? Savvy's probably not the right word, but centric to get them to make that turn. What are some of the things you could share that gets people to turn the light on and say, hmm, maybe this is a different way, a better way to do things? Yeah, maybe a use case that I will be talking about later on as well here on the RapidMiner Wisdom Conference. We did actually a project with our regional company in Brazil about imports into that country, Brazil, because if you are importing products, goods to Brazil, it always seems to be a lengthy, messy process with customs involved and all of that stuff. So we found out actually that only a very small percentage of import transactions would follow a direct line, the shortcut into the country and the majority of import cases would take up to 60 days, the whole process of importing goods into Brazil, where in the best case, you can do it in two weeks or 20 days. And so we were trying to understand the full process, including customs and paperwork and SAP transactions to identify root causes, why the majority of the cases would take so long. And simply by doing that, we could develop a very, very concrete set of actions and measures, some related to training, some related to SAP configuration topics, whereby now the majority of those cases is following the ideal path, the fast track into the country. You talked before about educating the lines of business on the potential of data science. So I infer from that that your strategy is not simply to do projects that support the business units, but to actually teach them how to fish, if you will, so they can develop applications and predictive solutions to improve their business. So assuming that's the case, when you look at 280 SAP systems, systems of record, essentially, are you, do you see yourselves building analytics, predictive analytics systems that are extensions of those systems of record, and those bringing analytics to those transaction systems, or do you see them as more separate green field applications? I think there's a strong trend in actually integrating analytics directly with transactional systems. Part of our process mining engine is also now coded into SAP HANA and in memory and all of that stuff on one side. On the other side, it's really, in many cases, about asking the right question, and it's asking the right question that needs so much work in advance that, you know, including analytical prototypes, if you like, that it is a project-driven approach, and sometimes you develop those sort of autonomous project solutions for your analytical work because the question is just so specific. So there's an overall trend, and then there's reality of how it's done today. Yeah, right. And that is basically sort of having a separate analytical nucleus project-driven. That's a great point, because so much is said often in this vision that you can just throw the data into Hadoop and the magic will come out right. But really, as you said, you really need to know the question or at least start with a question. It's a start with a hypothesis and then maybe work your way to a journey that wasn't specifically where you thought you were going to end up. And the question is domain knowledge specific, right? You need to be sitting in the middle of a factory to fully understand what this is about, right? Whether it's engineered to order or whether it's mass production of a product, right? And even just these two use cases, engineered to order and mass production, is so different. It's so different in how SAP systems are configured. It's so different in how the factory works and is organized. And data analysts need to understand that. And our auditors, by the way, are in a perfect position to understand that because they are seeing so many different business units over their audit time that this is actually knowledge they can absorb, right? So as this evolves, do you envision a thousand data scientists blooming within the lines of business? Is that sort of the model that Siemens will take or will it be a more centralized model? Or as Professor Weissman is saying, a hybrid approach? I think it's sort of a hybrid approach where I was clearly thinking that this discussion versus central versus de-central, you can have that discussion in a company of the size of 40,000 people, in a company of the size of 350,000 people. It's going to be de-central, yeah? Because each of the business units is so big that it's a company in its own. You have a CEO and a central. And for us in the internal audit, we are a growth and development organization. So what we basically do is we take in auditors that actually do not have auditing as a career vision. We educate them and train them for four years to be analytical consultants, to become trusted advisors. And then we develop them as a core part of our mission. We develop them into leading positions in the business units. So we are taking people, training them, and then sending them back into the business units to make a transformational difference out there. And that's so much nature of the way we work and what we do daily that this is having an impact over mid to long term in how Siemens develops analytical skills overall. So we're out of time, but last question is, from a standpoint of the technology industry, is it obviously very crowded, right? A lot of people trying to do predictive analytics and what do you want to see from the technology industry that would make your life better? I can clearly say that analytics is not process driven enough, right? I'm a big promoter of exactly that discipline of merging data mining and business process management, which translates actually into process mining, because that has a very, very concrete hook into operational business. And in many areas, data mining isn't having this hook. It's not concrete enough, it's not process driven enough, it's not actionable enough, yeah? Then it becomes a matter of sales forecast, okay? Yeah, we've been doing this for years, yeah? Yeah, it's removed from the fundamental business process. You want it embedded. Bar, it's a great vision and excellent perspective. Thanks so much for coming to the CUBE. It's a pleasure to meet you. Thanks for having me here. Thank you so much. Right there, buddy, we'll be back right after this word rapid miner 2016 wisdom. Right back, this is the CUBE.