JEREMY: To start, I want to get a feel for what marketing looks like today using data. In particular, it still feels that a lot of people feel like marketing is kind of a bifurcation—that on one hand, you have "traditional" marketing where you're going to be talking about customer discovery, broad campaigns, a lot of relationship management. And then on the other, you have the world of marketing analytics, really broadly defined, where you start having really customized campaigns that you send out—really hyper-focused marketing to individuals based on what you think you know about them. So, one of the things I wanted to ask you is, is this way of thinking valid in terms of this split? Or, has marketing evolved?
DAEWE: Yeah, from my experience at P&G, I think commercial marketing has evolved, definitely, and to be honest, mixes the two ways you mentioned. We have a philosophy that every action and decision should be data-driven.
For instance, if you know Gillette men's deodorant—if you search on Amazon, you will find the color of the product is actually changing from dark blue to light blue to attract more customers and more consumers. Or Tide Pods, the front side [has fewer] words now with straightforward functionality like "Coldwater Clean." The first time you have those ideas may be from UX or some traditional user testing, but what's next?
You cannot just spend a large amount of money and change all your brand imaging without testifying, right? So now marketing analytics comes in. With more data points, you can tell how much lift improvement this change has brought to the business, or even more detailed, what kind of consumers love this change. And [that can tell you] if you want to target specifically those kinds of consumers or if you want to broaden your idea. So after that, you can expand your activities to the national level, even regional level.
In other words, marketing today is actually bridging data and business to enhance traditional marketing practices rather than replacing them altogether.
J: Are you finding that people going into P&G's marketing group, or just more broadly in the industry, they have to have—it might not be a full data science skillset, but they have to have some understanding of data-driven decisions or at least some experience with some kind of analytical toolkit?
D: From my experience at P&G, I think we have both. Like, there are groups of people—they know data analytics, they know a little bit of SQL or Python, but they are super, super talented in the marketing area. Then I am actually in a position with more data science skills, and every decision is actually made by two groups together. So, we work together closely and then make every decision together—not like some people on the marketing side just make a sole decision without data science, or data scientists just provide recommendations without marketing knowledge. So, it's mixed.
J: When you see mainstream reporting on machine learning and AI and data science, a lot of times it's really the cutting edge—with ChatGPT, it's going to change everything—but I think a lot of what is actually happening in terms of how marketing analytics is done is a little bit more of the boring side of things. And what I mean by that is it's not this transformational, flashy thing. It's instead behind the scenes, possibly has been there for a while, but those things that are actually making people that are customer-facing better, faster, smarter at what they do without necessarily having this zero-to-one transformation happening.
Can you give the environment of what that part of the equation looks like in terms of getting more into the detail on how you're using analytics?
D: Yeah, yeah, definitely. I can give you a specific example. It's boring by the way, but it's essential. That's distribution.
So, with the scanning data, the retailer analytics team and distribution manager—they can easily and quickly check which products are unsellable this week and predict which products will potentially be out of stock next week. It's boring. Now, the magic happens when you have per store, per SKU data.
Imagine we don't have that in an automated dashboard with frequently updated data, which was actually the situation that occurred eight years ago when I did my internship with a luxury goods company. At that moment, I had to receive emails from each physical store and load the stock number in my Excel file and do all the manual stuff. And that was only 200 stores. Imagine for P&G it is all the retailers [in the entire U.S.] So, if you are doing manual stuff, it is crazy.
But now today, P&G is definitely a leader in digital transformation and data science in the consumer goods industry. And it has to be a leader because of all of that data, especially with tons of product IDs and retailer stores all over the world. And I will say it's boring, but it's essential that we have automated dashboards for each different use case to support all the different marketing teams and also the customer team. So yeah, the distribution is definitely one of the super good examples here.
J: Okay, so moving past the boring ones—and realistically, not really boring and in fact absolutely critical to what you do. Well, let's actually turn to the other side of that. The ones that have the bigger bang, the ones that we tend to see either have or have the potential of being really transformational in terms of how we do things. Have you seen anything or what are some of the approaches that you've found that really seem to have changed the game or are changing the game in terms of where we're going?
D: Yeah, so there's a post from our P&G CBO, Chief Brand Officer. He mentioned that the biggest opportunity for market growth is the multicultural market. In the past 10 years, 100% of U.S. population growth came from an increase in Black, Hispanic, Asian, Pacific Islander, multicultural, and multi-race segments. Their buying power was more than $5 trillion. So when markets grow, everyone benefits and we grow the picture and create business versus taking from others. I know people don't like the pie chart, but growing market is actually a more important kind of growth, and we are actually not increasing a percentage of the pie chart, but we're increasing the entire size of the pie.
A super cool approach I observed is micro-targeting based on demographic attributes. It comes from the data science method which with a much more granular mindset, you will be able to understand or find who your product resonates with and see more of them. For instance, if I'm talking about a university neighborhood, maybe people are busy with midterms or finals and they want to use Tide pods to throw just one or two small bars into the laundry machine and finish their laundry work because they still have to do the exam, right? They need to do the review, they don't have time.
Or, how can a business utilize a back-to-school season in August and September when people are buying things crazily? Do we give more of a discount or do we make the price a little bit higher? Even more granular, what are the products that African-American consumers love more at a historically black college and university. So multicultural, multi-demographic targeting is a super cool and advanced approach today a huge business opportunity.
J: The micro-targeting you're talking about—there's got to be a lot of data underneath that, and so one of the things that I know some folks would love to hear about is the kind of data that you're able to access. And I don't necessarily mean the in-house P&G piece, although feel free to talk about that if you'd like, but really what you can buy.
I think a lot of people, especially people just getting into data science, think the data begins and ends with what a company has and doesn't realize how much data is buyable, that you can purchase a lot of this data, and the data that's out there. When you start thinking about data privacy, these types of things come up. Could you talk about what some of the data sets are that let you and let a large organization really start targeting at this level?
D: Yeah definitely, so there are two kinds of data. One is more super sensitive, like sales data, scanning data—and to be honest, we [do not own] the data. Actually, the consumer who paid for the product owns the data and the [retailers] manage the data based on the consumer's agreement. So, that's the more highly sensitive part.
But I am super surprised that we have some [less sensitive, high-quality] data. When I was at Duke, when I did a data project, we always went to Kaggle or UCI library and download data and we think, “Oh my god, the quality is not that good and how can I build an algorithm on that?” But if you're just searching on Google, maybe Carlo, Spatial AI, you will have a data set based on the U.S. census group.
For instance, they have in per census group, how many people are from age 10 to age 20? How many people are from age 21 to age 30? How many people love to hear about beauty supplies? How many people have a guitar at home? How many people have a laundry machine? It's more survey-based and also based on the U.S. census survey. That data, to be honest, are not expensive. If you're doing research in a university setting, you can actually buy that.
You can also [buy] some e-commerce data. I'm not talking about Amazon sales data, it's not doable, right? Because Amazon owns that. They have high sensitivity, but you can go back and see, [for example], on Twitter, how many people are focused on what specific account? How many people love to use Instagram over Facebook? You can purchase [that data] as well in several products. So, the cost is not like $100 million or whatever. It's not that huge and it is frequently updated—every month or every half year—and the data quality is good.
J: Two follow-up questions for that. One is for anyone watching that didn't know that kind of data is out there and are thinking that someone is watching—can you give an idea of generally how you think the people that sell that data, how they're getting it—is that, for instance, survey data, is that people agreeing to when you sign a customer agreement? Do you have an idea of where that data comes from originally?
D: Yeah, definitely, I can tell you that. It's not an advertisement again, but there's one super good data set vendor called Spatial AI. It's more like a mid-size, entrepreneurial right now in this system and in this dataset world, and what they do is [gather data through Google Chrome cookies.] Every time your device is searching for something on Google and then links to a specific page. Let's say you want to order Thai food—you will have a cookie that's linking from your device location and this specific Thai food location or IP address—that's one record. And if you click “accept cookies,” this is actually saved in your Chrome, and if you [allow that information to be shared] with the data provider, sign the agreement—people always mark, “I agree,” right?
So that's one thing that they do, yeah. They actually collect your data from your device, and another one is some big company—for instance, Google. Every time you use Google Maps from your home to somewhere else, they actually track your route and how many times you take it, how busy it is, and we can draw a circle from the destination to your home and say, “Oh, it's 10 minutes walkable or drivable.” And then let's say you have a library and you want people to read more books and go to your library, you can actually use that data to draw a circle or draw a walkable distance or drivable distance, and then you can find your target customers. Like, what’s [their age, gender, or race?] So yeah, that's another kind of data, but Google, those big companies, they own the data and sometimes you have to be a giant company to collaborate with them, and that's the data with more cost.
J: I tend to find that fascinating and I generally know what happens when I accept a cookie. I think other people might not realize the level that can come out. So, it's probably worth talking about the ethics around this world because one of the things that you're talking about any time technology is moving very quickly, there are these ethical considerations on where do we want to be, what are we thinking about? We could spend weeks talking about the nuances of that. But from a high level, can you give an idea of what the ethical landscape looks like in terms of decision-making bodies? Who's at the table? How are those thoughts? How are they brought together?
D: Yeah, let me tell you one example for more, let's say more highly sensitive, the sales data. I am on the developer side and I need to do the coding, right? I still don't have all the full data. I need to request that data from each retailer team and from all the marketing teams to get access to that specific part of data. So it is, to be honest—I'm always thinking 80% of my time is doing data cleaning when you do a project. No, actually 80% of the time is getting the data access from them. You have to tell them—hey, what is my purpose, how this algorithm can be more valuable to this company and to this industry. You cannot just use that data to beat the other company. You have to make that happen to make your entire industry more successful based on your algorithm usage. And so that's more like a high-sensitive [data.]
For the low sensitive [data], the U.S. census group, the demographic future, the demographic index—they are actually getting access per buyer. So, you don't need to have a specific access or license to get that data. All you need to do is to buy the data and then tell the provider [what you will be doing with it.]