Prof. Dr. Heiko Beier - Know your Customer: Deep Level Understanding of Customer Interaction with Smart Data Analytics

Yes. Hello. Good morning. I'm talking to you as a professor for international media communication. So I think firsthand, this sound may sound pretty exotic to you, media communication. What does this have to do with information security, identity management?

Well, I'm a professor at language of applied languages. SDI. The S is for language German.

The, you can see it for, I like interaction. And if you put the D for digitalization, you have more or less what I'm talking about, and which is the, the subtitle of my talk. And maybe also interesting for you to know I'm not only professor in, in 2001 already. I founded my own software company called Moss and we started in 2001 working on semantic and cognitive computing in days when I think, well, particular, the latter one wasn't really invented. So the good thing is I will be very precise, very concrete with you in, in real life use cases.

And yeah, maybe to get the whole story and to understand what inspired me to all, to do all that and why I'm here it's would be interesting to take a little step back in history because I started out being a particle physicist. So my, I did my PhD in particle physics, and I had the honor of doing big data analysis already in the mid nineties. So that was also the time when Tim be Lee invented the worldwide web, the semantic web at CERN.

And yeah, what we did there is pretty similar to what we do today. And there. What we did is we did pattern mining. So there were a lot of data data, well like 10 million events per second.

We, we processed petabytes of data with the idea to find what's behind a signal what's really relevant behind a signal disclosing the truth behind something. There are some of course main differences are the main difference. And that's also the reason why I left this business was, was so, so very, very cost intensive. Yeah. To find this little signal, we spend about 3 billion euros on, in, at, at Daisy and Hamburg.

So, and this, you have to be aware, the 3 million euros were mainly for producing the data. So, and now we are living in a different world. We have more than enough data. So the focus is on making value on these data. So I want to introduce you to, to the topic with a couple of use cases, very concrete use cases, which at firsthand, maybe again, to you are not so clear. What does this have to do with, with information security, but at the very end, you will see it's, it's all about new approaches for smart customer identity. These are use cases from corporate banking.

It's about display advertising and digital assistance. So here, so your little use case from corporate banking doing sales and corporate banking.

Well, everybody of you who has been in contact with banks, tried to get a loan financing, might know that these guys are not very aware of what are you actually doing? What's your market? What are your chances in the market? Yeah. What are you planning to do simply because they don't have any chance to do that. They don't have the time to do that. They evaluate and assess you based on, on the past and on numbers and credit scores that are derived from that. So given the situation of, of a bank officer who works in sales, it's a great thing.

If he would be able to get signals, automatic signals, what he could sell to a client, what is the company planning to do? Where is it about to invest and what you see here in those little pictures, our analysis we have done on corporate bank on, on, on annual accounts from companies to automatically extract investment signals, to, to give the sales officers hints, whether they're expiring finances with, with other banking partners.

So this cognitive expert advisor helps you to really understand your customer, to know what he is about to do, what is planning to do and what market is use case we have discussed yesterday also, which of course is very interesting for, for supporting the onboarding process and knowing your customer better also for risk reasons. Okay, let's switch. Say it. I come through concrete use cases, something very, very different. Some of you might heard of Google losing big parts of the media budget spent on YouTube advertising.

Here are some quotes from recent magazines from newspapers wise that, well, unfortunately, some of the ads that, and not some, but really regularly Google showed and delivered ads next to, well, I would say compromising content, close to terrorism news and things like that. And of course, foreign advertiser, this is, well, this is something you definitely need to avoid. So they stopped many of their budgets and their campaigns. So you might be surprised who comes to that might mean Google the king of context. Yeah. Knows the world knowledge.

Well, actually it's not that easy. And even for Google, it's not that easy. Obviously. I want introduce you to that with a little example. Imagine beautiful Bavaria. Yeah. Some of you might wanna travel here around going to the Alps. This is my, what you, what you might see and yeah. You might go by train. Yeah. This is a little advertising movie from Deut Shaban German railway and yeah. Looking from the perspective in the front of the train. So they want you to, to sit back, relax. Yeah. So you might want to imagine that, but there is something you don't want to imagine.

Something quite horrible because I think it's about two years ago, a horrible accident happened right at this spot. I showed you before about I blink and two commuted trains in the very early morning hours, they crashed front by front head to head. I think there were like 30 death people, many, many injured people. So the problem is exactly this ad was shown in news reports about exactly this accident. Huh? So disaster for Deutsche ban. Why is this so difficult?

Well, in principle, we've seen, it's nice to, to show train or, or commercial ads for Deutche barn around what, what we've seen, but from a cognitive perspective, when people read this text, they have something in mind and they know what's going on and they have heard about this accident. So you need to be actually aware of all things that happen around and with cognitive analysis, you can do that.

We, we we've built a system that we are now introducing. And the nice thing about it is that it not only predicts to advertises where your ads will be shown, it's also fully compliant by GDPR because there's no tracking involved, no cookie tracking, no behavioral data collected and yeah. Similar for digital assistance.

So, and I'm not talking about ordering a pizza or asking about the weather. So very schematic, very straightforward questions. But I'm ask, I'm talking about moderating and answering complex questions where it is about knowledge, service requests, things like that. So also here machines can actually learn what, what is a concern to people?

They can understand what a customer is about to ask what we are actively doing this, for instance, in telecommunication, with Deutsche telecom, what the system can measure exactly what what's the intent, what, what concerns people, what kind of product related requests are giving what people want to do with what? And if you do these kind of analysis, if you have this knowledge in your machine, you can actually moderate service dialogues.

You can, the machine can anticipate what a user is about to ask. You can give hints, you can ask questions. You can have a call this in a chatbot, but also can do it in a, in a search like interface with a very smart and intelligent to complete giving, giving the relevant hints. In this use case here, the machine learns from community. It learns what people are asking. It learns what people are answering. And actually in this use case here we came across about something that might be very familiar for you. A security issue. We had to anonymize all the incoming requests.

The, the application was so successful. People also started typing in their credit card, numbers, numbers of the builds, asking questions.

Of course, all these data starts in a, in a database for analysis. So we had to anonymize it. This was for security reasons. Good news is that even if you anonymize it, if you get rid of the identity and all knowledge you have about the identity, you can still serve your customer perfectly. So now I'm getting closer to my message and why this customer interaction, smart customer interaction, cognitive computing is so close to what you guys do here. Look at this Bold's eye. What are we doing? Principle in the focus spot. You have the identity in the next step. We are enriching this identity.

We're enriching it. What you probably all do with metadata. You have behavioral data.

When, at what time, where, where was the excess done in the next step? We enrich it with smart data. And then we go to the next level, which is the intent. So enriching it with smart data means understanding of all interaction, of all communication of requests, of information, of communication. What's in there. What's really relevant. What's behind it.

Topics, business concepts, persons, organizations. And once you really understand, what's what's, what's in there. You can go to the next step and you can build contextual models. We've successfully deployed successful Mo contextual models for sentiment emotion analysis, where use it for understanding user intent. As I've shown you from the automation and, and, and telecommunication service, we are understanding market segments, target groups, and something very exciting stuff we're actually doing right now is classifying for confidentiality.

So, and here's the clue of what I'm telling you. Once you have this contextual model, you actually can get rid of the identity.

In many, many cases. If I understand the intent, I can remove the identity.

This, this intent serves as a perfect proxy for the identity. At least if you wanna address the question, how do I serve the people? How do I understand for what purpose someone is approaching me? So this contextual models can achieve a lot. You've seen finding sales signals, delivering ads at the right time and giving the right answers to people which have concrete questions, but it's even more regarding information security coming to GDPR times of wild west are over.

That's what I usually say to people which are not familiar with this topic, collecting data, bulk data, and later on, looking for ideas. What can I do with that? I think that's, that's over data has to be collected by a special purpose where you have consent by the user. But otherwise, if you have collected all this data, you need to make a re-engineering for what purpose has it been collected?

So to get into a little summary, to the very end, here's a framework, which includes much of what I've told you, which also includes how to cope with problems that are given, and that are arising through GDPR for data addressed world. That's how the world looks. On one hand, you have the open world, you have social media, you have news articles, corporate webpages. On the other hand, you have your enterprise, you have data addressed. You have all your corporate knowledge, you have your marketing communication. And in between that you have the communication channels.

So what we start doing, where we start from is we do contextual learning on billions of webpages, starting to understand what people are talking about. How do they express themselves with these contextual models?

Actually, you're able to very effectively process the inbound communication. As I've shown you, you can analyze it. You can match it to your internal, know how, and you can yeah. And implement very smart service bot, which are giving the right answers to people. And that's the first thing what this framework is doing, but it's more and anonymization, we've talked about that. You've seen it in the, in the real use case of Deutsche telecom. So what you do here, it's secure, You can apply this to, to your data addressed.

So what we actually do is data discovery, given all your data that you might have collected in different applications for different purposes, what's behind it. What's the signal behind it. Looking back at what I did as a, as a particle physicist with this reminds me, it's exactly the same thing. I'm trying to understand. I'm trying to discover I'm I'm discovering. Yeah.

What, what data is there and in the next step, of course, once I know that I can make sense of it, I can classify it. I can classify it for different purposes. I can classify regarding confidentiality and then I can neatly separate the data and yeah. Treat it properly. And the last thing, which is something I, we actually haven't thought of before, but which came across is this is also tool for data loss protection.

Because with all these contextual models know, knowing what's going on in the outside world, you can match this to your internal knowledge, neatly matching the pattern of knowledge that you have insight. And in the outside world, one final remark on this, this framework is also meant as an architecture. So what you see here is a very hybrid approach. And I definitely recommend to do all that.

Well, we cannot do that with just one system with one central installation. What we do is we are learning in the open world. We are learning in, in a cloud based area where we also offer this as services, but we apply this on private data in private clouds, on, at rest, on premise. And this also of course adds to the security architecture of this framework. Okay. Thank you. There are two questions first removing the identity is a good idea. But what about accountability in services processes?

Like, so how do you distinguish between cases where actually the anonymity is well accepted and even due to G GDPR necessary and other cases where you actually need accountability? Sure. So of course, regarding identity access, at some point you need the identity, that's clear. Yeah.

What I, what I've shown you is, is really interesting for actually in the onboarding process. It's very interesting. If you are, are in process where, where the user still anonymous yeah. Where you wanna identify him and where you wanna get him and, and serve him right. And properly. Okay. And where customers find predicting intent cool or creepy. I think it's my impression. They will find it. Cool. But they find it creepy if they come across that they are being sped on. Yeah.

I mean, Difficulty distinguish, do you think? No, actually not. Because for instance, what, what I've shown you with this display advertising solution, there's no spying involved. We actually have no idea of the identity. We don't know who's looking at the text, but we are looking at the text as the machine is looking at the text as a human being is doing. And the machine knows all the environmental issues like the human beings are doing. So there is nothing involved with spying on data.

We actually actually are implementing this with advertising at server companies who actually want to get rid of all their data. They have collected. Interesting. Okay. Thank you very much again.