Susan Etlinger:
Welcome to Leading the Shift, a podcast from Microsoft Azure where we hear from the people at the forefront of data, AI, and cloud technologies. I'm your host, Susan Etlinger. Today our guest is Shirli Zelcer, Chief Data and Technology Officer at dentsu, an integrated growth and transformation partner to the world's leading organizations. dentsu was founded in Tokyo, Japan in 1901. Yes, you heard that correctly, and is present in over 110 countries and regions.
And Shirli is a bit of a unicorn, or I don't know, maybe more like a zebra, in that she oversees both data and technology strategy, which is a role that feels like it could become a real force multiplier in the coming years. We're going to talk about how Shirli is quite literally leading the shift to bring data and AI together to help dentsu and its customers. And in the process, we're going to touch on a whole bunch of topics from how our understanding of the customer journey is evolving to the untapped potential of synthetic data to help us better understand audiences in a privacy-compliant way. So come for the strategy, stay for the anecdotes. Shirli, welcome to Leading the Shift.
Shirli Zelcer:
Thank you so much for having me.
Susan Etlinger:
No, it's great to have you. I'm so excited about this conversation. So early on when we had our first little chat, I was really taken by the fact that we had a similar, a kind of a parallel path except that you started life as a statistician and I come from a liberal arts background and the technology landscape has evolved so much. And I would love you to talk a little bit about how your background has influenced you and what you've seen in the last several years and where you think things are going.
Shirli Zelcer:
Sure. So as you just mentioned, I started out my career as a statistician, and back then 20-plus years ago, it was a very back office type of practice, and we were still getting all of the data and the insights and looking at things in a predictive way, but it was not really at the forefront of decision making. And I think over the years as things have evolved, we turned from statisticians to business analysts and then more recently into data scientists. And this whole concept of using it as a science and having it be at the forefront of strategy and decision-making and marketing. And to me, that's been really exciting because I've always believed in the power of data. And so watching it evolve over time to where it's really taking a front seat for most organizations is very exciting.
Susan Etlinger:
Yeah, absolutely. And more than that, it's become a topic of conversation. I mean I remember early in my career, people would ask me, "What do you do for a living?" And I'd say, "Oh, data, I'm interested in..." And they just sort of slowly back away from me, and now people want to have that conversation.
Shirli Zelcer:
Absolutely. It's definitely something that used to be back of mind, and now it's really in the front of how do we get more data and what do we do with our data? And it's definitely become a hot topic.
Susan Etlinger:
So how did you come to where you are? Tell us a little bit about your journey.
Shirli Zelcer:
Well, I have been with dentsu through the Merkle acquisition actually for, it will be 20 years this summer.
Susan Etlinger:
Oh wow.
Shirli Zelcer:
I've been data essentially my entire career. And so then it just became a natural progression of, I've used the data my whole career, I've thought about how to apply it, I've worked within platforms, and then how do we as dentsu was thinking about this new capability of data and technology and bringing it together. It seemed like a natural fit and something I was really excited about to be able to join and marry together the data and the analytics and everything that I'd been doing along with the tech and the platforms and those capabilities. So here we are today.
Susan Etlinger:
Oh wow. So did you start as... You started in data and took on CTO or is that how it worked or it all came together at the same time?
Shirli Zelcer:
It all sort of came together. So my career was spent largely as in my more junior career it was analytics and data, and then it grew into data platforms. And then we actually just newly formed earlier this year, or I should say last year, the data and technology organization at dentsu bringing everything together and so this is where I ended up.
Susan Etlinger:
So let's talk for a minute about dentsu. dentsu is a company that has been first of all global and has been around for a very long time and I would love to also hear a little bit about your role, which is fairly unique in that you are both the chief technology officer and the chief data officer.
Shirli Zelcer:
Yeah. Selfishly, to me that makes a ton of sense. I think that at dentsu we really wanted to create this organization of data and technology that takes advantage of the power of data and identity and AI and brings that all together to create a true competitive edge for marketers. And so bringing together the data competency and really understanding the depth of data and the different usages of data, and then combining that with our tech and what we're creating and how it's both connected into our client environments and to partner environments, to us, that was the right way to think about it.
Susan Etlinger:
Yeah, yeah, it's interesting because I've met many chief technology officers, many chief data officers and they almost feel like really different breeds. You've got a lot of chief data officers come from the data governance space and really intimately know data, and of course, chief technology officers are looking at technology strategy all up, and marrying those two capabilities I think is so important, especially now when everything is sort of converging. And one of the things that you had said earlier in our conversation really stuck with me because of course in the last two years and now it's just a bit over two years, you said my biggest excitement about generative AI is actually in the data space. And I think that's a really... It's just it's a really intriguing thing to say and I'm curious to hear you expand on that.
Shirli Zelcer:
Yeah, so I think we've seen a lot of tech come across in using generative AI around images and around copy and around all of these really amazing things that I don't want to take away from. And generative AI has done a ton for us there, but where I'm really also interested is how it can help us not only bring together new sources of data that weren't able to be brought together before, whether it's structured data or unstructured data coming together, whether it's combining data without the risk of privacy and security because we're using generative AI, whether it's the rapid speed to insights that you can get with generative AI. I think all of the usage and the use cases around data to me are incredibly intriguing. And then when you combine that with all of the other things that you see out there for generative AI, I think it's incredibly powerful.
Susan Etlinger:
Yeah. So maybe give us a little bit of a deeper dive on that. So what are some examples?
Shirli Zelcer:
So think about, for example, data that you have that we call unstructured data. So ratings and reviews or call center data that typically is not structured in a way where you would be able to combine it with other data. Now we actually have the power to combine that with first-party data, with third-party data and really get so much more insight and so much more predictive capability around how that is influencing behaviors, how that is basically taking advantage of all of the signals that consumers are giving us in order to really make better decisions.
Susan Etlinger:
For sure. And first-party data being, of course, your own data, and third-party data being data from other places, right?
Shirli Zelcer:
That's right.
Susan Etlinger:
I mean, it reminds me of several years ago I was doing a project with a big hotel chain and they had this property in Hawaii and the property always got three-star reviews and they were like, why are we getting three-star reviews? And they were looking at all these different sources. They were trying to look at market research, they were looking at NPS, they were looking at their social signals, and what they discovered was that it wasn't really a three-star review, it was a bunch of one-star reviews and a bunch of five-star reviews. And the five-star reviews were like, oh, this is the most beautiful place ever. The gardens are so lovely, their restaurant's so quiet. It's just so elegant. It's so great for our anniversary. And the one-star reviews were like, there's nowhere to go after 9:00. It's really boring. There are all these old people there.
Shirli Zelcer:
So that's actually an amazing example because imagine if we could take all of those ratings and reviews and now we combine them with who the consumer is and what their demographics are and what their purchase behavior is and everything else. And then all of a sudden you can create audiences and tailor experiences that are specific and okay, this specific property in Hawaii is meant for this type of demographic or consumers who are looking for this specific experience and that type of in a different property would be the better fit for people who are looking for the big party life. And it allows us to connect data in a way that we really haven't before.
Because even in the example that you gave, we could do that anecdotally. We can go and we could sift through the ratings and reviews and there's lots of great tools to be able to do that and connect the sentiment. And we've been doing sentiment analysis for a long time, but now we can actually take that raw data and combine it with the actual behavioral data and get a lot more powerful in how we're giving customers essentially what they want.
Susan Etlinger:
Exactly. Exactly. And then you can tell who thinks the towels are scratchy and who's still mad they have to pay for Wi-Fi.
Shirli Zelcer:
Everybody. The answer is everybody is still mad if they have to pay for Wi-Fi.
Susan Etlinger:
I know. It's funny. And it's been that way for 10 years. I think one of the things that really has been, I don't know, just inspiring for me is seeing how all of this comes to life through generative AI and through the ability to query just all of this incredible data. So can you talk a little bit about your predictive analytics Copilot, sort of how you're using that specifically actually for this kind of use case, maybe not the towel use case, but you know what I mean.
Shirli Zelcer:
Not the towel use case. Essentially, for us, it was important to try this use case of how is generative AI working for data. And so what we did is we really took an example of taking organizations' full suite of data. So like you mentioned, first-party data, which is their own data, whether it's transactional data, whether it's data that they've acquired through their sites, etc. And then third party data or demographic data, or any data that they've been able to collect from partners, and then all of this unstructured data, whether it was ratings and reviews or call centers or anything that they're able to get.
And then bringing it all together and really transforming the way that they're accessing that data and accelerating their speed to insights. So being able to use just regular easy conversational language to ask questions that typically would live in dashboards or typically you'd have to write code to be able to access and getting answers to questions and then being able to say, "Okay, well, let's forecast for the future certain scenarios, and how would they look based on all of this data?" And really giving our clients the ability to use these tools in a very powerful way.
Susan Etlinger:
So can you give an example of what kinds of questions can they ask?
Shirli Zelcer:
It's really anything that we have data for. So it's everything from are there meaningful client segments that are responding to this certain message in a specific way or how is my creative performing or if I have an extra budget of $1 million at the end of the quarter, where would be the best way to do it or on which channel, etc. So it really gives you the power to not only gain insights but also really think forward-looking into what you want to do.
Susan Etlinger:
Yeah, I had the opportunity to interview a gentleman who is actually the chief digital officer at ABB, which is one of Microsoft's, one of our customers. And we were talking about exactly the same thing except that that was from the industrial manufacturing perspective. So the people on the shop floor can ask, can query using a chatbot, and find out how the machinery is working and whether anything is about to fail. And that gives them just this incredible... It empowers them actually to be able to take these actions that otherwise they would've had to wait until the thing failed and maybe their emissions were now out of control or there was something else happening.
And so it's given them the opportunity not just to anticipate things at the shop floor, but then also once you have that capability, it opens up this portal of now what can we do? And it feels like this is a really consistent theme with dentsu, which is of course a global agency to industrial manufacturing to lots of other different types of industries. So it makes me wonder how that's coming to life culturally because you must have people at multiple levels of the organization who are now using data in a really different way.
Shirli Zelcer:
Yes. To me, it's super exciting because it gives the ability for every different level to get the insights and the answers to questions that they have that are relevant to making their day go better and making the decisions that they need to make better. So whether it's an executive who wants to see performance overall, but there's some kind of anomaly and they're trying to dig in and they get those questions answered quickly or whether it's a campaign manager or whether it's a product owner trying to understand how their product performance is going.
So it's really nice that now these insights are at people's fingertips very quickly and it takes all of the back and forth that we used to have where... I mean, a lot of people have very rich dashboards and capabilities, but if there is something different or if there's an anomaly, the question would usually have to go to somebody who would write some code, who would then pull the data and get that information and then send it back. And that cycle is becoming a lot smoother and a lot faster so that you can get insights much more rapidly.
Susan Etlinger:
Yeah. For sure. And so you can get those insights more quickly. You can generate audiences more quickly, you can do all sorts of different things, which I think probably opens up that creativity as well. I've been really intrigued and coming to data as somebody who grew up in the liberal arts world as opposed to the world of mathematics and statistics, I always found unstructured data so interesting because it's human data, it's all the things that people express in the different ways that they operate throughout their day. So maybe they're making a comment on a social channel or maybe they're responding to a survey or maybe they're having a conversation.
I can always hear my spouse in the other room having conversations with folks in call centers and feeling very, very bad for them. But you can now connect, interpret those signals in a way that is so much more precise than what was possible in the past, and connect them. I wonder also what you think about in terms of the opportunities of unstructured data going forward because it feels like there's more opportunity to enable that connected experience. Maybe it's connecting the person, but also just connecting an idea.
Shirli Zelcer:
I think it's connecting the entire customer journey, because to me, if a consumer is taking the time to write a review or call in, those are very meaningful signals. It's actually one of the most powerful things that you have. And so if you can take that and not only understand what they're saying and the sentiment but rather where they were in their journey, where they left that. Were they in the decision-making part of their journey? Were they post-purchase? Were they ready to make a secondary purchase? Where did they give you those signals and how did that influence the next step that they took? And using all of that information and really curating experiences for consumers that are much more meaningful, that really take into account what they tell the brand. I think that's really powerful.
Susan Etlinger:
Yeah, yeah. I'm super intrigued by the context around those signals because I remember in the early days I thought, well, social is so different from everything that we have and in that people are just expressing themselves in the moment. And also it reaches people who would not necessarily be reached by a survey because in order to survey someone, you have to know who and where they are and how to reach them and have at least... You have to have the permission to be able to reach them. And then by the time somebody is at an emotional state where they pick up the phone or they open up a chat box to speak with a call center, they've clearly had something happen where they have a need and maybe they have a strong feeling.
It feels like that just gives us so much more insight. But also we think about a purchase funnel, which it really feels like an antiquated term in the sense that everybody's always in all of these states with all the different brands. So I just feel like this could be something that really helps us understand fundamentally what people want and helps brands actually do much more precise work to be able to meet people where they are.
Shirli Zelcer:
I completely agree. I think, like I said before, these types of signals are to me a lot more powerful than what segment you're in or where you live, or the information that we would've typically used historically. Not taking away from that. I think that's also very powerful data, but I think the data that speaks to what the consumer is actually telling the brand is so powerful, and we've really only been able to use that as sentiment or directional. And now with generative AI, we're really able to bring it much closer to the behavioral aspects and really change the way that we're curating experiences.
Susan Etlinger:
Yeah, for sure. I mean, as an example, many years ago when my son was born, a certain social platform, which I will not name, started sending me targeted advertisements for losing stubborn belly fat. And needless to say, that was not really the message that I wanted to hear at that particular point in time, but there was a certain feeling that I had when I got that message. And I do think that we have this potential now to actually open up more empathy in our relationships with customers and consumers simply because we have a little bit more insight into what they really want. And I think that ability to create that personalized interaction at scale is what I think is really important and interesting.
Shirli Zelcer:
It's going to also be really interesting to see how customers actually then react to brands, because if consumers are actually starting to see that this type of data, is giving them a better experience are they going to be more vocal and interact more? Right? In that example that you gave, did you actually go in and say, "Hey, this is not relevant to me and why are you showing this to me?" But now that we actually have the ability to use that data and to curate better experiences, it's interesting to see how consumers are going to react to that and if they're going to take the time to actually give that type of feedback or connect more with the brands to get that better experience.
Susan Etlinger:
Well, and it really gets into the question of responsible data use and responsible AI and what the culture and what individual's specific desires are at a point in time. I mean, I remember having a conversation with an identity management company a long time ago, and they said, "Well if you want the ability to have your privacy controlled, we have to know who you are." And I thought like, oh gosh, that hadn't occurred me in the past. But of course, you do because you have to know what I want versus what Shirli wants versus what somebody else wants.
And we may be in different places in our lives, or we may be in different places with regard to what we want technology to do for us. And I think that is a really important piece is not so much the... I mean, I'm not underestimating the importance of privacy. It's critical, but the control over the privacy, I think because we may have different needs and different desires in terms of what we believe privacy really means. So let's talk about that though. Let's talk a little bit about data governance, about responsible use of data. What are some of the issues that you're seeing that you're thinking through and certainly that you're focused on in your role?
Shirli Zelcer:
There's a wide variety of those things.
Susan Etlinger:
I bet.
Shirli Zelcer:
So there's obviously the data compliance that we think about and the data privacy and both keeping data safe and making sure that we treat the data with the right level of security and try as much as possible to not move data around and keep it in secure environments. So that's number one. Number two, there's also the compliance from legal standpoints, right? So in different markets, you have different levels of fidelity of data that you're allowed to use. Obviously, there's in EMEA, there's GDPR, and there's a lot of regulation around what you're able to do with data and with AI specifically in data. And there's a lot of new laws and regulations, especially in the EU around AI. So there's that component.
And then there's also the component that to me is a little bit softer, but I think equally as important, which is making sure that you're using AI in a responsible and ethical and not creating bias that is unintended. And so I think all of those things are top of mind day in and day out as we're using so much rich data and making sure that we're not abusing the data and that we're both keeping it safe and compliant as well as using it in a way that does not make unintended decisions or unintended bias in our day-to-day.
Susan Etlinger:
Yeah, absolutely. And I think one of the things that we've talked about is how important it is to have certainly the governance piece from a compliance and legal perspective but also to have the processes supporting that also to have the education and skilling so that people have that awareness. And then of course, to have the tooling that enables you to detect bias in an algorithm or understand how a decision was made. And I think that's something that's really important. On the flip side though, one of the things that's been super intriguing to me lately is some ways that we're using technology to do things in a more privacy-compliant or responsible way. So for example, synthetic data, one of the ways that we can protect privacy is by using data that we actually create for particular purpose. So can you talk a little bit about that?
Shirli Zelcer:
Yeah, this is an area to me that I think is very untapped, and I think over the next year or two, we're going to see a ton more of this, both from a persona standpoint and creating synthetic personas and seeing how they react and seeing how they behave and using that. But then I think also more broadly for creating audiences that we can map to or we can activate against, because especially in privacy-compliant regions like the EU, we're not really able to do addressability at scale because of the challenges around PII. But using synthetic data and getting at audiences and behaviors and then being able to activate through certain signals to me, gives us a lot of opportunity to get a much more personalized experience while maintaining that compliance.
Susan Etlinger:
So yeah. So let's break it down. So PII is personally identifiable information. And so synthetic data, if I understand properly, gives you the ability to create a data set that mirrors an actual situation in the world, but without actually using their data. So you could say, okay, so I want to make a data set that is women in France from ages 23 to 42 or something like that who have done X, Y, and Z. You can actually create a data set to do that without using their specific data.
Shirli Zelcer:
Well, you would want to have a seed of data, so you still need to have a seed that is compliant, that is allowed for usage and approval. So you would use that seed data and then you would essentially blow it up into a full data set that has a lot of characteristics to it that you can then apply models on or activate against or do really cool things with that you wouldn't be able to do otherwise.
Susan Etlinger:
Yeah. And I've heard examples of this, for example, in the medical field where you can't necessarily do modeling on a population, let's say for brain MRIs or something like that, but you could take a subset of brain MRIs and actually create a population and actually model that for different kinds of outcomes so-
Shirli Zelcer:
Yeah. Because the nice thing is in that scenario, you could kind of scrub any of the personal information to create that synthetic data, so you're not actually using anybody's personal information, but then you're creating a rich data set that will allow you to get a ton of learning out of.
Susan Etlinger:
And you could do forecasting against that and different things like that. So I love that idea. Can you actually talk a little bit about zero-party data? What is zero-party data?
Shirli Zelcer:
Zero-party data is when a consumer actually gives their data consentedly. Is that even a word? Consentedly?
Susan Etlinger:
It is now.
Shirli Zelcer:
Let's go with it. To a brand. And a lot of times it's an exchange of value of some sort. So if there's a loyalty program and I willingly, give my information because I'm joining a loyalty program or because a brand asked me to do something. So when a brand gets all of that information and stores it and it's consented and privacy compliant, that is zero-party data. So different from transactional data, which we call first-party data and behavioral data, and then different from the third-party data, which is data that you get that is from different companies that you purchase.
Susan Etlinger:
So it's really a mosaic that we're thinking about from all the different ways that we can either encourage people to share data. We can use the data that in our transactions with them, we can use data from the outside. I mean, ultimately this feels like a mosaic and it feels like a way of using this technology to bring together as much information, as much knowledge and insight as we can in order to be able to achieve business outcomes. And in many cases, I think we haven't... We're still so early on in this conversation, in this whole platform shift that we're in that we haven't really even seen I think a lot of the impact that we're going to see in the future. But I wanted to just bring it home a little bit with what you've learned. In the last couple of years, you've been through a lot, we've all been through a lot and we've all learned a tremendous amount. What pops out at you?
Shirli Zelcer:
It's interesting. I think that at the beginning of this boom of generative AI, everybody was super excited about it and there was so much hype. And I think where companies have been successful is when they've said, "Hey, we still have the same goals as a company and we still have the same KPIs we want to achieve. Let's just see if AI can help us get there faster or more efficiently." As opposed to companies that have said, "Hey, we're going to bring in all of this cool AI and figure out what to do with it." And so I think that the biggest area of learning from your... I guess the advice that I would give is to think about the use cases that you have as a business or as an organization and what are the specific use cases that you're trying to achieve, and can AI help you with those specific use cases as opposed to trying to fit AI as a tool for everything. So that's sort of the big thing for me.
Susan Etlinger:
Yeah, start with the business problem.
Shirli Zelcer:
That's right.
Susan Etlinger:
I mean, I think it's interesting though. You start with the business problem. You start with a real problem, but at the same time, then there are things that you just discover.
Shirli Zelcer:
Along the way.
Susan Etlinger:
Exactly spontaneously. And I think that combination of business rigor and spontaneous ideation is also really powerful, and it's something that I hope we see more of. When you think at... So we were talking a little bit, I know we're going to geek out in the final moments. So there's a saying that I know you know by the very famous statistician, George E. P. Box, and potentially, he said, although we're not entirely sure if he said it, "That all models are wrong, but some models are useful." And I thought that was super interesting when you apply it to unstructured data because humans are so complex and the way that we express ourselves and our language is changing all the time. In fact, you just coined a word in the middle of this podcast, language changes all the time. And so do you think all models are wrong? Do you think some models are useful? How do you think about some of these first principles-
Shirli Zelcer:
It's funny-
Susan Etlinger:
... as it relates to where we are?
Shirli Zelcer:
... no model is a hundred percent accurate, no tool, no model. So in a sense that it's actually right, all models are wrong, but I firmly believe that they're also incredibly useful and they're better than no models. So I think I would actually change the saying a little bit to say all models are wrong, but properly, both models are better than no models.
Susan Etlinger:
I think that sings, and I think that we should just propagate that as our motto for the future. Amazing. Well, thank you so, so much, Shirli. It's been such a pleasure chatting with you and-
Shirli Zelcer:
It's been great. Thank you for having me.
Susan Etlinger:
Of course. Thank you. Wow, that was such a fantastic conversation. And there's a lot going through in my head right now, but there are a few things that really stuck with me. The first thing is everybody talks about how data is the fuel for AI, but Shirli raises a really, really intriguing point, which is that it's actually kind of a two-way street. Data fuels AI, but generative AI is also transforming the data space by enabling organizations to bring all these new sources of data together to do all sorts of things that couldn't be done before. And then right alongside her appreciation of the opportunities of rich data and advanced models is Shirli's respect for using data and AI responsibly. She has this incredible background in statistics, which gives her tremendous insight into how important data privacy and security, and responsible AI really are. And finally, Shirli's journey is a microcosm of the platform shift we're all living through.
She started her career as a statistician, which was frankly seen as a back-office practice back then. But given all the technology innovations of the past two decades, and particularly in the age of AI, strategic use of data is so critical to an organization's competitive advantage that technology and data strategy are now C-suite priorities. And Shirli's career is a perfect illustration of that. Thanks for listening. I learned so much from this conversation and I hope you did too.
We hope you've enjoyed this episode, and as I'm sure new podcasts live and die on engagement. So please like, comment, share, tell your friends, and let us know what you'd like to hear about. We're listening. If you'd like to learn more about how people and organizations are innovating with Microsoft Azure, visit azure.microsoft.com.
--------
Credits:
Leading the Shift is a production of Microsoft Azure. Susan Etlinger is our host. Our executive producers are Hailie Meehan, Aaron Russell, and Erik Williams. Production support for this episode provided by Indigo Slate.
Welcome to Leading the Shift, a podcast from Microsoft Azure where we hear from the people at the forefront of data, AI, and cloud technologies. I'm your host, Susan Etlinger. Today our guest is Shirli Zelcer, Chief Data and Technology Officer at dentsu, an integrated growth and transformation partner to the world's leading organizations. dentsu was founded in Tokyo, Japan in 1901. Yes, you heard that correctly, and is present in over 110 countries and regions.
And Shirli is a bit of a unicorn, or I don't know, maybe more like a zebra, in that she oversees both data and technology strategy, which is a role that feels like it could become a real force multiplier in the coming years. We're going to talk about how Shirli is quite literally leading the shift to bring data and AI together to help dentsu and its customers. And in the process, we're going to touch on a whole bunch of topics from how our understanding of the customer journey is evolving to the untapped potential of synthetic data to help us better understand audiences in a privacy-compliant way. So come for the strategy, stay for the anecdotes. Shirli, welcome to Leading the Shift.
Shirli Zelcer:
Thank you so much for having me.
Susan Etlinger:
No, it's great to have you. I'm so excited about this conversation. So early on when we had our first little chat, I was really taken by the fact that we had a similar, a kind of a parallel path except that you started life as a statistician and I come from a liberal arts background and the technology landscape has evolved so much. And I would love you to talk a little bit about how your background has influenced you and what you've seen in the last several years and where you think things are going.
Shirli Zelcer:
Sure. So as you just mentioned, I started out my career as a statistician, and back then 20-plus years ago, it was a very back office type of practice, and we were still getting all of the data and the insights and looking at things in a predictive way, but it was not really at the forefront of decision making. And I think over the years as things have evolved, we turned from statisticians to business analysts and then more recently into data scientists. And this whole concept of using it as a science and having it be at the forefront of strategy and decision-making and marketing. And to me, that's been really exciting because I've always believed in the power of data. And so watching it evolve over time to where it's really taking a front seat for most organizations is very exciting.
Susan Etlinger:
Yeah, absolutely. And more than that, it's become a topic of conversation. I mean I remember early in my career, people would ask me, "What do you do for a living?" And I'd say, "Oh, data, I'm interested in..." And they just sort of slowly back away from me, and now people want to have that conversation.
Shirli Zelcer:
Absolutely. It's definitely something that used to be back of mind, and now it's really in the front of how do we get more data and what do we do with our data? And it's definitely become a hot topic.
Susan Etlinger:
So how did you come to where you are? Tell us a little bit about your journey.
Shirli Zelcer:
Well, I have been with dentsu through the Merkle acquisition actually for, it will be 20 years this summer.
Susan Etlinger:
Oh wow.
Shirli Zelcer:
I've been data essentially my entire career. And so then it just became a natural progression of, I've used the data my whole career, I've thought about how to apply it, I've worked within platforms, and then how do we as dentsu was thinking about this new capability of data and technology and bringing it together. It seemed like a natural fit and something I was really excited about to be able to join and marry together the data and the analytics and everything that I'd been doing along with the tech and the platforms and those capabilities. So here we are today.
Susan Etlinger:
Oh wow. So did you start as... You started in data and took on CTO or is that how it worked or it all came together at the same time?
Shirli Zelcer:
It all sort of came together. So my career was spent largely as in my more junior career it was analytics and data, and then it grew into data platforms. And then we actually just newly formed earlier this year, or I should say last year, the data and technology organization at dentsu bringing everything together and so this is where I ended up.
Susan Etlinger:
So let's talk for a minute about dentsu. dentsu is a company that has been first of all global and has been around for a very long time and I would love to also hear a little bit about your role, which is fairly unique in that you are both the chief technology officer and the chief data officer.
Shirli Zelcer:
Yeah. Selfishly, to me that makes a ton of sense. I think that at dentsu we really wanted to create this organization of data and technology that takes advantage of the power of data and identity and AI and brings that all together to create a true competitive edge for marketers. And so bringing together the data competency and really understanding the depth of data and the different usages of data, and then combining that with our tech and what we're creating and how it's both connected into our client environments and to partner environments, to us, that was the right way to think about it.
Susan Etlinger:
Yeah, yeah, it's interesting because I've met many chief technology officers, many chief data officers and they almost feel like really different breeds. You've got a lot of chief data officers come from the data governance space and really intimately know data, and of course, chief technology officers are looking at technology strategy all up, and marrying those two capabilities I think is so important, especially now when everything is sort of converging. And one of the things that you had said earlier in our conversation really stuck with me because of course in the last two years and now it's just a bit over two years, you said my biggest excitement about generative AI is actually in the data space. And I think that's a really... It's just it's a really intriguing thing to say and I'm curious to hear you expand on that.
Shirli Zelcer:
Yeah, so I think we've seen a lot of tech come across in using generative AI around images and around copy and around all of these really amazing things that I don't want to take away from. And generative AI has done a ton for us there, but where I'm really also interested is how it can help us not only bring together new sources of data that weren't able to be brought together before, whether it's structured data or unstructured data coming together, whether it's combining data without the risk of privacy and security because we're using generative AI, whether it's the rapid speed to insights that you can get with generative AI. I think all of the usage and the use cases around data to me are incredibly intriguing. And then when you combine that with all of the other things that you see out there for generative AI, I think it's incredibly powerful.
Susan Etlinger:
Yeah. So maybe give us a little bit of a deeper dive on that. So what are some examples?
Shirli Zelcer:
So think about, for example, data that you have that we call unstructured data. So ratings and reviews or call center data that typically is not structured in a way where you would be able to combine it with other data. Now we actually have the power to combine that with first-party data, with third-party data and really get so much more insight and so much more predictive capability around how that is influencing behaviors, how that is basically taking advantage of all of the signals that consumers are giving us in order to really make better decisions.
Susan Etlinger:
For sure. And first-party data being, of course, your own data, and third-party data being data from other places, right?
Shirli Zelcer:
That's right.
Susan Etlinger:
I mean, it reminds me of several years ago I was doing a project with a big hotel chain and they had this property in Hawaii and the property always got three-star reviews and they were like, why are we getting three-star reviews? And they were looking at all these different sources. They were trying to look at market research, they were looking at NPS, they were looking at their social signals, and what they discovered was that it wasn't really a three-star review, it was a bunch of one-star reviews and a bunch of five-star reviews. And the five-star reviews were like, oh, this is the most beautiful place ever. The gardens are so lovely, their restaurant's so quiet. It's just so elegant. It's so great for our anniversary. And the one-star reviews were like, there's nowhere to go after 9:00. It's really boring. There are all these old people there.
Shirli Zelcer:
So that's actually an amazing example because imagine if we could take all of those ratings and reviews and now we combine them with who the consumer is and what their demographics are and what their purchase behavior is and everything else. And then all of a sudden you can create audiences and tailor experiences that are specific and okay, this specific property in Hawaii is meant for this type of demographic or consumers who are looking for this specific experience and that type of in a different property would be the better fit for people who are looking for the big party life. And it allows us to connect data in a way that we really haven't before.
Because even in the example that you gave, we could do that anecdotally. We can go and we could sift through the ratings and reviews and there's lots of great tools to be able to do that and connect the sentiment. And we've been doing sentiment analysis for a long time, but now we can actually take that raw data and combine it with the actual behavioral data and get a lot more powerful in how we're giving customers essentially what they want.
Susan Etlinger:
Exactly. Exactly. And then you can tell who thinks the towels are scratchy and who's still mad they have to pay for Wi-Fi.
Shirli Zelcer:
Everybody. The answer is everybody is still mad if they have to pay for Wi-Fi.
Susan Etlinger:
I know. It's funny. And it's been that way for 10 years. I think one of the things that really has been, I don't know, just inspiring for me is seeing how all of this comes to life through generative AI and through the ability to query just all of this incredible data. So can you talk a little bit about your predictive analytics Copilot, sort of how you're using that specifically actually for this kind of use case, maybe not the towel use case, but you know what I mean.
Shirli Zelcer:
Not the towel use case. Essentially, for us, it was important to try this use case of how is generative AI working for data. And so what we did is we really took an example of taking organizations' full suite of data. So like you mentioned, first-party data, which is their own data, whether it's transactional data, whether it's data that they've acquired through their sites, etc. And then third party data or demographic data, or any data that they've been able to collect from partners, and then all of this unstructured data, whether it was ratings and reviews or call centers or anything that they're able to get.
And then bringing it all together and really transforming the way that they're accessing that data and accelerating their speed to insights. So being able to use just regular easy conversational language to ask questions that typically would live in dashboards or typically you'd have to write code to be able to access and getting answers to questions and then being able to say, "Okay, well, let's forecast for the future certain scenarios, and how would they look based on all of this data?" And really giving our clients the ability to use these tools in a very powerful way.
Susan Etlinger:
So can you give an example of what kinds of questions can they ask?
Shirli Zelcer:
It's really anything that we have data for. So it's everything from are there meaningful client segments that are responding to this certain message in a specific way or how is my creative performing or if I have an extra budget of $1 million at the end of the quarter, where would be the best way to do it or on which channel, etc. So it really gives you the power to not only gain insights but also really think forward-looking into what you want to do.
Susan Etlinger:
Yeah, I had the opportunity to interview a gentleman who is actually the chief digital officer at ABB, which is one of Microsoft's, one of our customers. And we were talking about exactly the same thing except that that was from the industrial manufacturing perspective. So the people on the shop floor can ask, can query using a chatbot, and find out how the machinery is working and whether anything is about to fail. And that gives them just this incredible... It empowers them actually to be able to take these actions that otherwise they would've had to wait until the thing failed and maybe their emissions were now out of control or there was something else happening.
And so it's given them the opportunity not just to anticipate things at the shop floor, but then also once you have that capability, it opens up this portal of now what can we do? And it feels like this is a really consistent theme with dentsu, which is of course a global agency to industrial manufacturing to lots of other different types of industries. So it makes me wonder how that's coming to life culturally because you must have people at multiple levels of the organization who are now using data in a really different way.
Shirli Zelcer:
Yes. To me, it's super exciting because it gives the ability for every different level to get the insights and the answers to questions that they have that are relevant to making their day go better and making the decisions that they need to make better. So whether it's an executive who wants to see performance overall, but there's some kind of anomaly and they're trying to dig in and they get those questions answered quickly or whether it's a campaign manager or whether it's a product owner trying to understand how their product performance is going.
So it's really nice that now these insights are at people's fingertips very quickly and it takes all of the back and forth that we used to have where... I mean, a lot of people have very rich dashboards and capabilities, but if there is something different or if there's an anomaly, the question would usually have to go to somebody who would write some code, who would then pull the data and get that information and then send it back. And that cycle is becoming a lot smoother and a lot faster so that you can get insights much more rapidly.
Susan Etlinger:
Yeah. For sure. And so you can get those insights more quickly. You can generate audiences more quickly, you can do all sorts of different things, which I think probably opens up that creativity as well. I've been really intrigued and coming to data as somebody who grew up in the liberal arts world as opposed to the world of mathematics and statistics, I always found unstructured data so interesting because it's human data, it's all the things that people express in the different ways that they operate throughout their day. So maybe they're making a comment on a social channel or maybe they're responding to a survey or maybe they're having a conversation.
I can always hear my spouse in the other room having conversations with folks in call centers and feeling very, very bad for them. But you can now connect, interpret those signals in a way that is so much more precise than what was possible in the past, and connect them. I wonder also what you think about in terms of the opportunities of unstructured data going forward because it feels like there's more opportunity to enable that connected experience. Maybe it's connecting the person, but also just connecting an idea.
Shirli Zelcer:
I think it's connecting the entire customer journey, because to me, if a consumer is taking the time to write a review or call in, those are very meaningful signals. It's actually one of the most powerful things that you have. And so if you can take that and not only understand what they're saying and the sentiment but rather where they were in their journey, where they left that. Were they in the decision-making part of their journey? Were they post-purchase? Were they ready to make a secondary purchase? Where did they give you those signals and how did that influence the next step that they took? And using all of that information and really curating experiences for consumers that are much more meaningful, that really take into account what they tell the brand. I think that's really powerful.
Susan Etlinger:
Yeah, yeah. I'm super intrigued by the context around those signals because I remember in the early days I thought, well, social is so different from everything that we have and in that people are just expressing themselves in the moment. And also it reaches people who would not necessarily be reached by a survey because in order to survey someone, you have to know who and where they are and how to reach them and have at least... You have to have the permission to be able to reach them. And then by the time somebody is at an emotional state where they pick up the phone or they open up a chat box to speak with a call center, they've clearly had something happen where they have a need and maybe they have a strong feeling.
It feels like that just gives us so much more insight. But also we think about a purchase funnel, which it really feels like an antiquated term in the sense that everybody's always in all of these states with all the different brands. So I just feel like this could be something that really helps us understand fundamentally what people want and helps brands actually do much more precise work to be able to meet people where they are.
Shirli Zelcer:
I completely agree. I think, like I said before, these types of signals are to me a lot more powerful than what segment you're in or where you live, or the information that we would've typically used historically. Not taking away from that. I think that's also very powerful data, but I think the data that speaks to what the consumer is actually telling the brand is so powerful, and we've really only been able to use that as sentiment or directional. And now with generative AI, we're really able to bring it much closer to the behavioral aspects and really change the way that we're curating experiences.
Susan Etlinger:
Yeah, for sure. I mean, as an example, many years ago when my son was born, a certain social platform, which I will not name, started sending me targeted advertisements for losing stubborn belly fat. And needless to say, that was not really the message that I wanted to hear at that particular point in time, but there was a certain feeling that I had when I got that message. And I do think that we have this potential now to actually open up more empathy in our relationships with customers and consumers simply because we have a little bit more insight into what they really want. And I think that ability to create that personalized interaction at scale is what I think is really important and interesting.
Shirli Zelcer:
It's going to also be really interesting to see how customers actually then react to brands, because if consumers are actually starting to see that this type of data, is giving them a better experience are they going to be more vocal and interact more? Right? In that example that you gave, did you actually go in and say, "Hey, this is not relevant to me and why are you showing this to me?" But now that we actually have the ability to use that data and to curate better experiences, it's interesting to see how consumers are going to react to that and if they're going to take the time to actually give that type of feedback or connect more with the brands to get that better experience.
Susan Etlinger:
Well, and it really gets into the question of responsible data use and responsible AI and what the culture and what individual's specific desires are at a point in time. I mean, I remember having a conversation with an identity management company a long time ago, and they said, "Well if you want the ability to have your privacy controlled, we have to know who you are." And I thought like, oh gosh, that hadn't occurred me in the past. But of course, you do because you have to know what I want versus what Shirli wants versus what somebody else wants.
And we may be in different places in our lives, or we may be in different places with regard to what we want technology to do for us. And I think that is a really important piece is not so much the... I mean, I'm not underestimating the importance of privacy. It's critical, but the control over the privacy, I think because we may have different needs and different desires in terms of what we believe privacy really means. So let's talk about that though. Let's talk a little bit about data governance, about responsible use of data. What are some of the issues that you're seeing that you're thinking through and certainly that you're focused on in your role?
Shirli Zelcer:
There's a wide variety of those things.
Susan Etlinger:
I bet.
Shirli Zelcer:
So there's obviously the data compliance that we think about and the data privacy and both keeping data safe and making sure that we treat the data with the right level of security and try as much as possible to not move data around and keep it in secure environments. So that's number one. Number two, there's also the compliance from legal standpoints, right? So in different markets, you have different levels of fidelity of data that you're allowed to use. Obviously, there's in EMEA, there's GDPR, and there's a lot of regulation around what you're able to do with data and with AI specifically in data. And there's a lot of new laws and regulations, especially in the EU around AI. So there's that component.
And then there's also the component that to me is a little bit softer, but I think equally as important, which is making sure that you're using AI in a responsible and ethical and not creating bias that is unintended. And so I think all of those things are top of mind day in and day out as we're using so much rich data and making sure that we're not abusing the data and that we're both keeping it safe and compliant as well as using it in a way that does not make unintended decisions or unintended bias in our day-to-day.
Susan Etlinger:
Yeah, absolutely. And I think one of the things that we've talked about is how important it is to have certainly the governance piece from a compliance and legal perspective but also to have the processes supporting that also to have the education and skilling so that people have that awareness. And then of course, to have the tooling that enables you to detect bias in an algorithm or understand how a decision was made. And I think that's something that's really important. On the flip side though, one of the things that's been super intriguing to me lately is some ways that we're using technology to do things in a more privacy-compliant or responsible way. So for example, synthetic data, one of the ways that we can protect privacy is by using data that we actually create for particular purpose. So can you talk a little bit about that?
Shirli Zelcer:
Yeah, this is an area to me that I think is very untapped, and I think over the next year or two, we're going to see a ton more of this, both from a persona standpoint and creating synthetic personas and seeing how they react and seeing how they behave and using that. But then I think also more broadly for creating audiences that we can map to or we can activate against, because especially in privacy-compliant regions like the EU, we're not really able to do addressability at scale because of the challenges around PII. But using synthetic data and getting at audiences and behaviors and then being able to activate through certain signals to me, gives us a lot of opportunity to get a much more personalized experience while maintaining that compliance.
Susan Etlinger:
So yeah. So let's break it down. So PII is personally identifiable information. And so synthetic data, if I understand properly, gives you the ability to create a data set that mirrors an actual situation in the world, but without actually using their data. So you could say, okay, so I want to make a data set that is women in France from ages 23 to 42 or something like that who have done X, Y, and Z. You can actually create a data set to do that without using their specific data.
Shirli Zelcer:
Well, you would want to have a seed of data, so you still need to have a seed that is compliant, that is allowed for usage and approval. So you would use that seed data and then you would essentially blow it up into a full data set that has a lot of characteristics to it that you can then apply models on or activate against or do really cool things with that you wouldn't be able to do otherwise.
Susan Etlinger:
Yeah. And I've heard examples of this, for example, in the medical field where you can't necessarily do modeling on a population, let's say for brain MRIs or something like that, but you could take a subset of brain MRIs and actually create a population and actually model that for different kinds of outcomes so-
Shirli Zelcer:
Yeah. Because the nice thing is in that scenario, you could kind of scrub any of the personal information to create that synthetic data, so you're not actually using anybody's personal information, but then you're creating a rich data set that will allow you to get a ton of learning out of.
Susan Etlinger:
And you could do forecasting against that and different things like that. So I love that idea. Can you actually talk a little bit about zero-party data? What is zero-party data?
Shirli Zelcer:
Zero-party data is when a consumer actually gives their data consentedly. Is that even a word? Consentedly?
Susan Etlinger:
It is now.
Shirli Zelcer:
Let's go with it. To a brand. And a lot of times it's an exchange of value of some sort. So if there's a loyalty program and I willingly, give my information because I'm joining a loyalty program or because a brand asked me to do something. So when a brand gets all of that information and stores it and it's consented and privacy compliant, that is zero-party data. So different from transactional data, which we call first-party data and behavioral data, and then different from the third-party data, which is data that you get that is from different companies that you purchase.
Susan Etlinger:
So it's really a mosaic that we're thinking about from all the different ways that we can either encourage people to share data. We can use the data that in our transactions with them, we can use data from the outside. I mean, ultimately this feels like a mosaic and it feels like a way of using this technology to bring together as much information, as much knowledge and insight as we can in order to be able to achieve business outcomes. And in many cases, I think we haven't... We're still so early on in this conversation, in this whole platform shift that we're in that we haven't really even seen I think a lot of the impact that we're going to see in the future. But I wanted to just bring it home a little bit with what you've learned. In the last couple of years, you've been through a lot, we've all been through a lot and we've all learned a tremendous amount. What pops out at you?
Shirli Zelcer:
It's interesting. I think that at the beginning of this boom of generative AI, everybody was super excited about it and there was so much hype. And I think where companies have been successful is when they've said, "Hey, we still have the same goals as a company and we still have the same KPIs we want to achieve. Let's just see if AI can help us get there faster or more efficiently." As opposed to companies that have said, "Hey, we're going to bring in all of this cool AI and figure out what to do with it." And so I think that the biggest area of learning from your... I guess the advice that I would give is to think about the use cases that you have as a business or as an organization and what are the specific use cases that you're trying to achieve, and can AI help you with those specific use cases as opposed to trying to fit AI as a tool for everything. So that's sort of the big thing for me.
Susan Etlinger:
Yeah, start with the business problem.
Shirli Zelcer:
That's right.
Susan Etlinger:
I mean, I think it's interesting though. You start with the business problem. You start with a real problem, but at the same time, then there are things that you just discover.
Shirli Zelcer:
Along the way.
Susan Etlinger:
Exactly spontaneously. And I think that combination of business rigor and spontaneous ideation is also really powerful, and it's something that I hope we see more of. When you think at... So we were talking a little bit, I know we're going to geek out in the final moments. So there's a saying that I know you know by the very famous statistician, George E. P. Box, and potentially, he said, although we're not entirely sure if he said it, "That all models are wrong, but some models are useful." And I thought that was super interesting when you apply it to unstructured data because humans are so complex and the way that we express ourselves and our language is changing all the time. In fact, you just coined a word in the middle of this podcast, language changes all the time. And so do you think all models are wrong? Do you think some models are useful? How do you think about some of these first principles-
Shirli Zelcer:
It's funny-
Susan Etlinger:
... as it relates to where we are?
Shirli Zelcer:
... no model is a hundred percent accurate, no tool, no model. So in a sense that it's actually right, all models are wrong, but I firmly believe that they're also incredibly useful and they're better than no models. So I think I would actually change the saying a little bit to say all models are wrong, but properly, both models are better than no models.
Susan Etlinger:
I think that sings, and I think that we should just propagate that as our motto for the future. Amazing. Well, thank you so, so much, Shirli. It's been such a pleasure chatting with you and-
Shirli Zelcer:
It's been great. Thank you for having me.
Susan Etlinger:
Of course. Thank you. Wow, that was such a fantastic conversation. And there's a lot going through in my head right now, but there are a few things that really stuck with me. The first thing is everybody talks about how data is the fuel for AI, but Shirli raises a really, really intriguing point, which is that it's actually kind of a two-way street. Data fuels AI, but generative AI is also transforming the data space by enabling organizations to bring all these new sources of data together to do all sorts of things that couldn't be done before. And then right alongside her appreciation of the opportunities of rich data and advanced models is Shirli's respect for using data and AI responsibly. She has this incredible background in statistics, which gives her tremendous insight into how important data privacy and security, and responsible AI really are. And finally, Shirli's journey is a microcosm of the platform shift we're all living through.
She started her career as a statistician, which was frankly seen as a back-office practice back then. But given all the technology innovations of the past two decades, and particularly in the age of AI, strategic use of data is so critical to an organization's competitive advantage that technology and data strategy are now C-suite priorities. And Shirli's career is a perfect illustration of that. Thanks for listening. I learned so much from this conversation and I hope you did too.
We hope you've enjoyed this episode, and as I'm sure new podcasts live and die on engagement. So please like, comment, share, tell your friends, and let us know what you'd like to hear about. We're listening. If you'd like to learn more about how people and organizations are innovating with Microsoft Azure, visit azure.microsoft.com.
--------
Credits:
Leading the Shift is a production of Microsoft Azure. Susan Etlinger is our host. Our executive producers are Hailie Meehan, Aaron Russell, and Erik Williams. Production support for this episode provided by Indigo Slate.
