All things artificial intelligence with the VP of AI at Samsung
Vice President of AI – Samsung SDS
About this episode
In today’s episode Patrick Bangert, the Vice President of AI at Samsung SDS, talks about the importance of artificial intelligence.
We discuss working from home during the pandemic, pros and cons during Covid, current usage and benefits of AI, and the future of AI.
You might also want to listen to…
- Episode #33Simple TVVickmar Martinez & Yenny Velazquez
Podcast #33 – Artificial Intelligence is the Future of Customer Service
- Episode #32CrowdlinkerAram Melkoumov
Podcast #32 – Data access will open up the possibility for AI to grow
- Episode #31NeoenergiaRenato Suplicy
Podcast #31 – Keep the Human at the Center of your Digital Strategy
- Episode #30PATJohn Ball
Podcast #30 – Reasoning Engines and the Truth About AI
Today we have Patrick Bangert, Vice President of AI at Samsung. Thank you so much for being here, Patrick. And before we went on, we both talked about being from Germany. So I should really say, Patrick Bangert? Hi, tell us, what do you do at Samsung?
Andrea, thank you very much for having me on the podcast today. I work for Samsung SDS, which is the IT company in the Samsung family of brands. So we run the data centers, we write the software for the group. And within that scheme of things, we have an AI department and the AI department do two major things. One is to write the software that the group uses to do all of its AI. And the other group actually does the AI. That’s the scientists and data scientists who use the toolsets we create to solve the problems that our customers have, both customers within Samsung as well as customers outside of Samsung. We do artificial intelligence in every which way. So we do computer vision to analyze images, natural language processing to do things like the Bixby system on the phone, as well as time series and other numerical things. So anything in AI you can think of, our department can handle.
Do you at all work with the customer service departments at Samsung? Or are you removed since you guys are such a huge conglomerate?
Well, the customer service departments within the various Samsung companies do use AI as a component in their services. And my own division, the division does have customers and so we have to take care of them. And we do have some customer service components here as well.
Okay, this podcast is mostly about customer service, because that’s who we serve. But because we’re also an artificial intelligence company, we definitely want to talk about AI. So I’m going to probably switch back and forth with you a little bit on customer service versus AI. Alright, so let’s get started. So we are going to start with customer service, what have you done to promote great customer service in your organization, especially right now, with COVID? And what’s happening with customer service departments and being a little bit thin nowadays?
Right. So before I answer that, I should say our customers are other corporations, typically slightly larger corporations, so our customers aren’t individual people. And in that sense, our customer service is usually the fact that we have one face to the customer, the customer has a specific individual within our organization, who is their contact point. And so there is a relationship that establishes over time, these people get to know each other, the customer representative on our side is a real person, you can get to know that person, you can establish a relationship over time. And so things can get friendly, as time passes by. And then the customer service isn’t necessarily a formal event, or action that is taken, but rather, you know, something between two people who know each other already helping each other out to solve a problem. And that can make things a lot easier, a lot more stress-free, and can also get the problem solved a little bit more pragmatically. So we try to solve the problem instead of trying to get paperwork and forms filled and complaints done and whatnot. And that typically works quite smoothly. And how long is an average customer? 10 years? months? How long did they work with you? Many years, many years. So it’s a real relationship. You really know each other’s products, what you’re working on your businesses.
Okay, that’s really good, because then you can get in there so much more. So have you guys noticed a huge change with COVID-19? Or is it not really affecting you? I just interviewed a couple of people that do business to customers and their smaller companies, and they’re extremely affected. Some of them had to let go of their entire customer service departments. So because you are B2B is it very different with you?
I can’t imagine anyone not being affected, that the differences are gigantic. I mean, first of all, everybody’s working from home. And which means that every aspect of every relationship, even with colleagues is completely virtualized, right, we cannot meet our own office colleagues at all. We haven’t met for over half a year. And, of course, that changes things. And that’s a negative what I just mentioned, on the positive side, we no longer commute. And we no longer need to go from building A to building B to have a meeting. And we don’t have to visit the customer, which is again, a longer commute, the customer doesn’t come to visit us, which is a commute for them. And we simply get on the call. Now, for anybody in America, and there are different time zones involved in this country, as well as for Samsung, who has customers in virtually every country around the globe, our customer service is a little bit tricky, because of all this time zoning, which, in pre-COVID ages, men taking the airplane quite a lot of the time, nowadays from the comfort of your own couch. And you can have a call very early in the morning, late at night. And it doesn’t really disturb you all that much. So the upshot of that is that the number of calls has gone up quite a bit. And every person who has Customer Contact has more Customer Contact now than they ever did before them, which actually serves to improve these relationships rather than let them zero rate despite the fact that we never meet them.
And so I would say productivity has improved. And general effectiveness has improved, even though people feel lonely through these work from home technologies. And of course, to assist all of this, what has to shut up, what had to change is use of technology. And so you went out you’re using Zoom at the moment, that’s I think the primary technology people are relying on. But for example, to beyond the relationship aspect of customer service, you need to record it, you know, what exactly has gone wrong. Where has it on what is the error message or the fault that’s occurred? And then you have various other technologies, like Microsoft Teams, or JIRA, and tools like that to, you know, document what’s gone wrong, what action has been taken to correct that error.
And so a variety of tools, software based cloud based, are used to assist this. And, and between all these technologies, we’re getting the job done quite well without any interaction. Yeah, so yeah, we’re affected quite a lot. Yeah.
I like that. You said that, though, that people are working different hours. Now. I mean, things are so different. It’s going to be so interesting what it’s going to be like, once we go back to work, how many people are going to want to keep what we’ve been doing. And some people are going to be like, No, no, no, I want to go back to the office and the nine to five. I’m a work from home type of person. So I love this environment. I’m going to ask you something. That’s why I’m switching the question a little bit. Usually, we ask the same five questions, because usually we have mostly customer service professionals. But I’m so excited to have an AI vice president here. So I want to ask you more about what you’re doing with your department. So how do you measure the success of your AI team in your organization?
Well, we have different objectives. And one is that we’re a revenue generating department. So one of the primary successes for us is quite simply revenue. And revenue comes in a variety of forms. We sell our software, and we sell our services. And of course, we sell support to that as well. So there we see how much is incoming, how much is renewing. And what are we getting done. Definitely, because it’s AI and AI is a science slash data science. And we do have accuracy measures as well. You know, how good are our models? How accurate are they? You know, if it’s a classification, what’s that? What’s the confusion matrix look like? So are we managing to deliver to our customers better and better models as time goes by, how can we differentiate ourselves from other AI companies that may or may not want to develop better models than we do? And then what is the amount of time that is required both in terms of days as well as in terms of person-hours to make that model. Of course, AI has the ambition as we go on and on over the years to use ever fewer person-hours to generate a really good model. But so it’s, if you will, the ratio between person-hours and accuracy. And that’s quite interesting, how much effort do I have to invest? And then the other side of the coin is, of course, how much computer hours do I have to invest in order to get the same accuracy out. And we would hope that that decreases as well over time, respectively, by using more computer infrastructure, I can get ever better models again. And that’s actually the core offering of my department is we have a software package called the red X AI accelerator, which takes a training task and splits it across a network of computers to make it much faster. And by that speed improvement, you can do all sorts of auto machine-learning tricks to make the model better. In that way, you’ve accomplished all of these things, you’ve made the model better, by using fewer people hours, because you’ve transformed the task from a person’s task to a computer task over that distributed network. So you’ve actually improved all of these different metrics, then you can deliver better models to your customers, and hopefully, then they pay you for it. And then the whole thing becomes one, that rounded business model
Okay, so I’m gonna have questions about this. So can you explain because a lot of people that are listening, they’re not understanding the depth that obviously, that you have of artificial intelligence, their customer service people that might have never actually used at all, and they’re now about to get into it, and get their first chatbots on their website and get their first messengers and get their first knowledge centers. So since your Samsung is huge, what is at the very basic, like, what is the product that you would recommend for every customer service department to have? So think about the smaller folks that are just now starting to get into this world?
Well, that would be the chatbot, hands down. So if you visit one of the bigger websites these days, um, and you arrive there at the bottom right hand corner, typically, there’ll be a little pop-up window, saying something along the lines of if you have any questions, you know, you can ask me. And so you ask a question, and you get an answer back. That’s not a person writing back to you that is a computer model slash an artificial intelligence model. And what it does is you provide the question or a statement, and it provides some sort of hopefully intelligent reply to that. These chatbots are usually customized to the application that they’re dealing with. So let’s say you visit a car website, you know, you want to buy yourself a new car, if you start asking pointed questions about the cars that this company is selling, you’re probably going to get a reasonable answer. But if you ask questions about how the AI is feeling today, and what it might do for fun tomorrow, it will probably have issues giving you a good answer to that. And again, if you ask a question, you’ll get a pretty good answer to that. If you ask a sequence of questions, though, and you expect the answers to make sense respective to the previous answer and the previous to the previous answer, you’ll probably be disappointed. And so that tells you that whatever this thing is, that’s generating the answers, it’s not intelligent. Okay, so if you think AI is actually intelligent in the same sense that Andrea here or Patrick is intelligent, it isn’t. And it’s basically an automation technique to give you something sensible back for an inquiry, but that’s something that is automation technology that allows a great many, let’s say 90%, maybe even a little bit more than that of inquiries to be dealt with automatically, without having to have a person. Right. So if you want to ask, you know, how many doors does this car have versus that car, the AI can take care of that. That’s the sort of level of complexity. Yeah. So how many gallons per mile or, you know, this, these sorts, that sort of level question, the AI will deal with just fine. And if you want to ask like, I have five kids, you know, are they going to like it? Can they play in the back, the AI will probably be unable to give you an intelligent answer to that.
So there, you need emotional insight. And that’s the point where you need to pick up the phone and talk to the person. And so that’s that’s the sort of level that you’re at, if you expect that the vast majority of your customer inquiries are pretty simple and factual. a chatbot will enable you to automate a way 90% of that traffic from people, the other 10%, you still need to answer by people. So you can get rid of your human customer service department completely. But it goes a long way. So that would be the first technology that I would advertise. And there are many systems out there that will make that realistically possible, even on a smaller budget.
Yeah. All right. So let’s take that the basics, the chatbot, to know, I want you to take your Samsung hat off and just put on your thought leader in the AI space. Where do you think AI is gonna go in the future? Maybe, broadly, in general, but also, when it comes to customer service? Like what do you think’s going to happen to AI and customer service in the future?
That is an excellent question. Now, where AI as a whole is concerned, there is really a conflict between three elements. One element is fear. By the general public of artificial intelligence, they think of the movie Terminator and think of movies like that Hollywood’s great at generating fear of AI. Right, there have been, I don’t know, 20-30 movies, recently, that basically addressed the topic of once you have achieved artificial general intelligence, basically, humanity is in trouble.
And that and over the robots always take over in these movies. Exactly. We always live. Exactly. Yeah. So the one thing to say to that is, please don’t be afraid of this. And we are extremely far away from that level of capability. We are in fact, so far away from it, that we don’t even really know how to define general artificial intelligence either. So don’t be afraid of the robot, a couple. Apocalypse is not not going to happen tomorrow, or during my lifetime, is my prediction.
So the second component is academic AI. AI done at universities, the vast majority of AI researchers are academics. And, and they are pursuing AI in various ways. Some of them are related to this artificial general intelligence or creativity and things like that things that don’t have an immediate commercial application. And they’re pursuing their research goals. And the third component is commercial AI systems, some of which are built at universities and some of which are built by companies like Samsung SDS. And these are typically extremely specialized, very, very targeted applications, like a chatbot for a car sales website that can answer questions about these cars but can’t answer questions about anything else really. And or a cancer detection system. It can take an MRI scan and tell you whether you have cancer or not, but they can’t do anything else. So that’s state of the art those aspects now what’s the future I think that one of the aspects of the future is that these disparate systems have to be more and more unified. So instead of having cancer-detecting AI and the COVID, detecting AI and a, I don’t know measles detecting AI, I want to have a health AI that could detect all of these things. And instead of having an app that you know, calculates my resting heart rate or my sleep pattern or anything else. I want to have an all-encompassing person assistant that can help me with my health and nutrition and whatnot, and so on and so forth across all categories of life. So that would be one thing. And the other thing is the commerciality of it. I think the business models have to change that are behind AI. Currently, they incentivize the specialization. And they incentivize very heavy models. And the most recent model that was released, that’s going to be of interest to customer service for short is called GPT. Three. It’s the most sophisticated language processing system out there. So it’s going to be extremely relevant for this chatbot application we’ve been talking about.
And it just got exclusively licensed to Microsoft a couple weeks ago. So what’s going to happen with that? And there are rumors, and it’s not quite certain who is right, but it costs somewhere between four and $10 million in the electricity bill, the electricity bill, I emphasize that to train this model once. Okay, so that tells you that ordinary mortals cannot train models like this. Yeah, not because we’re dumb or not capable or under educated or whatnot, we simply don’t have the resources computationally to do something like this. And that I think, is worrisome, because it concentrates the power behind AI into the hands of a very few large corporations or governments. And that needs to change. And I think a future of AI will include a reliance on less resources than this. Yeah, there are some directions and in terms of customer service, I see basically a continuation of automation, more and more automation, especially based on natural language processing. In other words, that chat bot should be able to handle more and more complex questions, and maybe even tasks, right. So currently, these chatbots cannot handle tasks. Like if I asked for a refund, that chat bot cannot give me a refund.
Right. It lacks the authority, it lacks the understanding it lacks the tie into financial systems. It doesn’t have any of that. But why not think that eventually, I could initiate the full, you know, return and refund of a product that I purchased via an AI system. It’s not unreasonable.
Yeah. So, let’s talk a little bit more about customer service. And that is also my last question. So I want you to think about this. What is the number one thing that you recommend for customer service folks, we’re only talking about customer service, folks, when it comes to AI, what should they be doing? I know we said the chatbot. That’s the very basics. But in general, what should they do? What’s the number one advice that you have for customer service professionals?
The chatbot is the technology. And in order to enter AI, I think it would be good to really analyze very deeply, what are the different activities that your customer service department actually engages in. And be as compartmentalized as you possibly can be? Because AI right now state of the art today, as I said, AI is a very disparate collection of very specialized things. So the more divisional you can get in that analysis, the more help you’re going to get from AI. Right. So what do you need to do? Some of the things you need to do is answer really, really easy questions. Okay, great. chatbot will handle that. You might have to answer very complex, maybe emotional, weird situational queries. Okay. Sorry, I can’t help with that. You got to have a person on retainer to deal with this. And do you get a lot of returns for your products, if you do consumer products, maybe clothing or shoes or something and you’re going to get 30% of your wares sent back.
They’re probably a whole bunch of activities in that process that you can automate. And so I know that there are some companies that before return is activated, they send out barcodes to the customer and ask the customer To stick on the barcodes that then automates the entire logistics in the warehouse that takes care of this individual product that’s maybe not exactly artificial intelligence, but it’s an automation technology that allows you to have less manual handling in the warehouse. And how do you do refunding? Again, that’s, it could be at least partially automated. You know, again, maybe asking the user to input again, credit card information, this and that. And then you can automate that process, so analyze all the tasks that you’re doing in a very fine grained way. And then for each one of them, you can ask yourself, Is it simple enough for either just a plain software system, or more sophisticated an AI system to handle this aspect? And then you can ask yourself, How do I make these aspects work together? And is it kind of triviality? Someone has to push a button? Or do I have to write some interface because that computer program is to talk to that computer program somehow.
And in which case, you can typically get the manufacturer of those computer programs to write those interfaces for you because they want to make more business. And you can tie in these systems together until you finally arrive at a largely automated customer service type department. And I’m not advertising reduction of headcount. But that means that you can, your people now can focus on the complicated cases, and help those people out. And in the end, you’ll probably get your customer service queries handled faster, and less with less friction. And customers will be more happy with that, you can advertise that fact, hey, guess, guess what, you know, with us, this type of customer service issue that returns for example, refunds is a lot faster than those guys. You know, instead of having to hang in, in a eternity of a phone call, for, you know, an average of two hours with us, you can get it done in 10 minutes, because we’re AI assisted and all powerful and whatnot, then then, then you have an actual sales argument towards you know, your customers. But if you think you want a system that kind of does everything out of the box for you, then you first of all won’t get it. And second of all your attempts to get it will just simply disappoint and frustrate you. So this ideation process of what exactly am I doing, and dividing it up into very fine slices of individual tasks? I think that will lead you to the goal much faster and will save you a lot of frustration.
Welcome to the Future of Customer Service Podcast. I’m Andrea Palten, from Inbenta and I will be interviewing customer support and service professionals to see what is currently working well, what issues they’re trying to overcome, and the future success of customer service.
Yes, I love that. That’s such great advice. It’s something that a lot of times gets overlooked.
All right. Thank you for having me.
Interested in Customer Service Automation?
Download our ebook!
Three smart solutions that can make an impact on your AI-powered customer service automation
Case studies related to each solution
Tips to choose the best technology to automate your customer service