Harikesh Nair, Professor of Marketing at Stanford and Chief Business Strategy Scientist at JD.com: where growth will come from in 2020

Dec 10, 2017

AURELIE: My name is Aurelie Guerrieri and I’m the founder of Akila One, a Growth Consultancy helping mobile startups scale. I’m the author of The Mobile Native’s Guide to Marketing, and the creator of the podcast series, “Growth Hacking is Dead, Long Live Growth Marketing.” Today, as part of this series, I speak with Harikesh Nair, Professor of Marketing at the Stanford Graduate School of Business and upcoming Chief Business Strategy Scientist at JD.com. Harikesh, in a few words, can you tell us how you got into studying marketing, specifically marketing analytics?

HARIKESH: Firstly, thank you so much for having me, it’s a real pleasure to be here. I got into marketing because I was doing engineering in Texas, Austin, at the University of Texas in Austin and I was working on transportation problems, trying to understand how people drove on a project for the EPA. And I became really fascinated by the use of data and the use of probability and other mathematical models to understand human beings and how they behaved in a variety of situations, and I had never been exposed to that as part of my engineering training before.

And as I read that literature more and more I found the questions in marketing extremely interesting. For instance, how do people buy products? How do people react to the recommendations of their peers? And how do they decide to react to prices, to advertising, and to external stimuli? And I became really interested in the way in which theories about human behavior got combined with data and with the mathematical spin to create real actionable outputs for firms. For instance, by understanding consumer behavior firms could improve their pricing, their promotion, their advertising, their targeting, their product entry, their market size assessment strategies. And for the first time I thought, “Wow! All of the science is actually very useful.” And I became very drawn to marketing as a field. I dabbled a little bit in thinking about doing a PhD in economics or a PhD in marketing. I found the problem in marketing way more interesting, and here I am.

AURELIE: You just shared with me and that you’re taking this a step further in terms of helping companies understand human beings and behaviors by joining jd.com, one of the largest Chinese e-commerce companies. Can you tell us a little bit more about that future role?

HARIKESH: Yes, absolutely, I am very excited about that. I am going to start working with JD to find ways to take their incredible data assets, which include their entire e-commerce stack within their incredible platform that JD has built in China, in collaboration with some insight into consumer behavior and data from WeChat, which folks on this podcast will know as one of the largest social network based messaging platforms in China.

Using the e-commerce data and the WeChat data, we are trying to understand consumer behavior much better and provide a platform for brands that want to settle into China, especially to tackle growth in rural China where the growth is quite incredible. And the CEO, Richard Liu, has essentially recognized that the way to win in China is through technology and through data. Marketing is essentially now a field in which you can succeed only by bringing the analytical smarts, the creativity and the data together. How do you put all of that together in collaboration with really smart people and in conjunction with the incredible data assets to find a way to sell into one of the fastest growing markets in the world? That’s what I will be doing. I am excited about the ride that I’m in for.

AURELIE: It’s fascinating that you’re going at social from the WeChat angle now because in 2013 you looked at Facebook quite deeply: you conducted an analysis of 106,316 Facebook messages across 782 companies, and you used a combination of Amazon Mechanical Turk and natural language processing algorithms to extract specific findings on the kind of posts that elicited the most user engagement and the best brand visibility, the best brand impact, the best ROI. Clearly social media has changed drastically since then both in terms of audience (and globally that’s even truer), and in terms of product features. So, if you were to take a wild guess today, how do you think the best practices that you found in 2013 are evolving in 2017 and beyond?

HARIKESH: That’s a very good question. I think the way in which the entire targeting and creation of content that appeals to users will evolve is through the application of AI and better personalization. One way to think about the problem is to go to theories of how human beings behave and to applied psychology of engagement, and to human-computer interaction models, and try to reverse engineer from first principles, what works for this type of segment, and what does not work for another kind of segment. Both the academic community and the smart practitioner community have thought about this very deeply for a long time, but a different paradigm has emerged with the collection of data at scale through digital models.

Now we actually don’t really need to rely fully on our gut feel, or our intuition, or on our theories. We can simply look at data at the micro-individual level and seeing what exactly this individual has liked, engaged with, disliked, said no to, has not shared, shared more, etc., in the past. And all of those touch points and all of those behaviors are collected at scale and not aggregated but collected at scale at the micro level.

So, by looking at what you’ve done I can essentially bring in the old economic idea that revealed preferences are superior to stated preference, you know, what you do is more powerful than what you say you’ll do or what you’ll express you’ll do. So by collecting that kind of data history I can actually do very intense personalization, and by aggregating all of that data through machine learning algorithms and large scale high dimensional statistical algorithms I can actually borrow some information across consumers to understand what cuts across contexts and cuts across consumers and does well for the average consumer. But I can use the micro data for an individual consumer to do well for each individual consumer as well, right?

I think what’s going to happen is increased personalization, at the level of a consumer and based what is acceptable, [delivering] a great creative content ad based on what worked for you individually in the past. You know, what’s red for you is black for me and what’s black for me is red for you. So, it’s quite useless arguing what’s better in terms of tastes and things like that. It’s really simply better to use user behavior to provide a recommendation or a piece of content that is tailored to each user.

We are slowly getting there as a community, finding a way to do that, and those powers are going to become incredibly awesome with the collection of data and the use of algorithms. We will actually get content that is not really meant to be ads, we’ll actually get content that is incredibly useful and tailored to you and completely non-annoying and really valuable, and AI is going to help us do that.

AURELIE: That’s fascinating, because you’re just talking about data collected today but as the number of collection points will increase (I saw a forecast that connected devices will reach 26 billion by 2020), the progression of AI that you mentioned but also continued progression of computational powers (Moore’s law is still helping us), cloud storage capabilities, those analytical capabilities are going to increase further and further and that micro individual personalization and the creation of valuable content and valuable advertising capabilities are really going to increase.

Do you think we’re moving towards a world of predictive advertising? For example predicting products and services that someone will need based on their recent behaviors and not just saying, “This person likes casual games so I’m going to show them another casual game, because I’ve seen they use casual games and that puts them in a casual gaming category.” But be a little bit more specific and say, “This person likes casual games and I’m noticing that they’re getting tired of playing the current casual game that they are playing, now is the right time to advertise a new one to them,” so that kind of predictive advertising, do you think that’s pie in the sky or do you think that’s going to be part of the future of advertising?

HARIKESH: I absolutely think that that is the future of advertising and content in general as opposed to being pie in the sky. I think it’s absolutely imperative that industry and science find a way to get us there because the current models are simply not sustainable. A lot of the current model is giving really high quality, expensive to produce content away for free, in order to force consumers to pay attention to some kind of advertising that they really would not want to consume otherwise. And that model is just not sustainable because free content is simply too expensive to produce and give away for the kind of money you can make out of advertising except of course if you are Google or Facebook.

The consumer attention is a scarce commodity on the internet and there are just too many options on which to consume really good quality content, so showing people advertising or content that they really don’t want or cheating them to click on something through click-bait is not going to work.

It’s absolutely important that we better predict what works for you, that we predict what kind of content you would like to consume and which is useful. For instance, if I’m in the market for a car it’ll be useful for you as the advertiser or the platform to show me content that’s related to a car in the specific category that I’m looking for, like luxury cars, or SUVs, or whatnot, without overwhelming me with a plethora of information that I’m not able to process and my brain just shuts down from choice conflict. Instead, tailor information that’s actually useful.

And not show me ads that just covers the entire page and just annoy me and force me to look away. So, it’s absolutely important that industry and science develop ways by which you can predict what content is important; what ads are usually important. And then deliver them in a seamless way, in an interesting way, in a useful way, perhaps without having the user ask for it through search.

Along that line are two other points, one is: what is content, what is advertising? That distinction is fast going away with native and things like that. It’s very important of course that what is paid for is appropriately labeled in a salient way as the FTC wants. They’re sponsored, but I’m quite okay consuming sponsored content as long as it’s useful for me. So, content is advertising and advertising is content and I don’t think we need to worry too much about the distinction of what’s one, what’s the other as long as it’s disclosed appropriately and there is no deception.

The other point is that the unit of analysis typically was a user, and the underlying assumption behind that is that a user is defined by a set of tastes. I’m a user, I’m Harikesh, I like black, or I am Harikesh and I’m a price sensitive consumer so I would like to see value ads in every different context. But actually that notion is very flawed, the real unit of analysis is a user-occasion combination. Me, Harikesh, in one occasion I can like black, in some other occasion or another context I can like red. And in one situation I might be quite price sensitive but when it comes to let’s say my daughter’s health, I’m completely price insensitive, I’ll pay anything to help that out.

So, the power of digital is that you can move away the analysis from a user to a user-occasion context, and you can see, “Well, Harikesh in this context, how has he acted before?” And then micro-target advertising content and other stuff to that level of unit. It’s absolutely imperative, it’s already there I would say, in best-in-class forms or it’s coming. And that’s my view.

AURELIE: Yes, I agree with you on the native aspect, which is for me not just a format question but it’s a context question as well. It’s not only user location! I would even add timing: there are times to which you’re open to certain advertising and times where your preferences change.

HARIKESH: Absolutely.

AURELIE: I want to talk a little bit about mobile because a lot of what you’re talking about in terms of content being expensive, attention span being shorter and shorter, real estate being scarcer and scarcer is aggravated on the mobile device. You’ve studied it and you’ve actually demonstrated that mobile advertising helps validate brand perception. That was a really interesting finding that I read. Can you tell us a bit about this experiment and what you think it reveals about the consumer psyche in the mobile context?

HARIKESH: Correct, as you said, on mobile it’s very important to provide ads that are relevant and useful, and the real estate is limited, and the bulk of advertising revenues are going towards mobile. One of the big debates about native advertising, which is becoming very important on mobile, has been whether or not to disclose or how to disclose the fact that a particular piece of content or ad unit is actually sponsored and paid for by a firm as opposed to organic content produced by the platform.

There is a big debate about this, both in the regulator community and in the advertising community. The regulator, which in the United States is the FTC, wants clear disclosure. Some proponents of mobile native advertising say, “Well, if I show you the fact that this content is sponsored and paid for, our users will trust it less,” so it’s a question of trust. Some others say that if I hide it or don’t reveal it appropriately then user trust on the platform will decrease.

Generally, the consensus seemed to be native works because users confuse what is native ads for what is organic unbiased content. So we wanted to get in and test these theories. At one level, one abstract theory is that people just don’t trust any advertised claims. Why would they, because they know this is not an unbiased claim by a third party but this is actually paid for by the advertiser.

At another extreme is an interesting theory from economics which has been around for about 20-25 years, which is called signaling, and says: let’s think about who would advertise. If I don’t know who’s a good firm and who’s a bad firm I would click on the ad and then go visit the firm and then I’ll realize, “Hey, this is a really good restaurant and how come we never knew about this.” Or alternatively I’ll go there and say, “Hey man, this pasta sucks, this is a really bad restaurant.”. So, if it’s really true that advertising causes people to come there, only the really good ones would want to advertise because the really bad ones, after spending some money on advertising, will bring consumers in and those consumers when they arrive there they will realize that this restaurant is not very good and there would be no repeat purchase, right?

AURELIE: So, if you have a bad product you’re not going to spend any money advertising it, right?

HARIKESH: You really don’t want to, yeah. I mean, it’s true that you may want to do a little bit of subterfuge and try to spend money but you can’t spend lots of money on this as a sustainable strategy. So if the fact that only the good restaurants will want to advertise is true, you can imagine what consumers may think in that kind of world. They might say, “Okay, I see this ad, this restaurant that’s advertising, and when they’ll go there to that restaurant, ‘hey this is actually a really good restaurant,'” because It must be the good restaurants that are advertising. There are expectations that firms that are advertising are not rejected by what’s happening in the marketplace and slowly those beliefs get reinforced

There are two implications of that theory: if a firm is advertising then A/ it’s more likely to be the better firm that’s advertising than the bad firm, B/ consumers react positively to the fact that the firm is advertising, because they realize or at least they have some notion that it must be the better one. Now, to give you some structure in that, imagine you wanted to visit a restaurant and then you walked into Stanford University and you went to this little coffee shop and on the bulletin board you saw a small flier from a restaurant, “Oh this is a new restaurant in Palo Alto, go for it.”

Alternatively, you open your browser and you go onto “The New York Times” and on the home page you saw the restaurant advertised. When do you think you are more likely to go to that restaurant? Most people would think, when you see it in “The New York Times,” and the reason is, they have enough money to advertise on “The New York Times,” so it must be good, right? Why would they spend so much money advertising on “The New York Times” if they were a really crappy restaurant?

Under that theory, revelation of the fact that content is advertised is a good thing. This is the exact opposite prediction of the other theory that we laid out that says people just mistrust advertising (when they see content is advertised they won’t believe it as much as non-advertised content).

So, we ran an experiment, to understand this. It was a fairly large-scale experiment with a mobile restaurant search platform like Yelp, it’s called Zomato, they own Urbanspoon in the United States. We did this experiment in 11 cities in Asia and we randomized large numbers of consumers into conditions in which ad units were revealed to be sponsored, or not sponsored. Firstly, we found that better quality restaurants are advertising on the platform consistently with that signaling story.

And to our surprise, the demand for a restaurant was roughly about 77% higher with the same content in the same position, when shown to a random set of consumers so the consumer characteristics were held the same, everything was the same except in one condition it was revealed to consumers that this ad is paid for. Click-through rates and demand (purchase intent) was 77% higher, which was consistent with this so-called notion that advertising signals quality. That has implications for the firm and for the regulator, the regulator likes it because firms now have an incentive to disclose on their own without active regulation.

The platform likes it because the platform wants to sell ads and useful ads. You want ads to be useful for consumers and it’s useful because consumers by their clicking and calling behavior are revealing that they like it and find it useful. And for the advertisers it’s good because it helps sort between the bad ones and the good ones, and good firms are willing to pay to get their advertising on the top of the page and it produces increased demand. So, that was the point of the paper.

AURELIE: So, advertising bolsters credibility and in this case, that’s fascinating.

HARIKESH: Yes, but it wouldn’t work in all contexts. For example, when I’m reading “The New York Times” or let’s say I’m going to “The Wall Street Journal” and I want to read about what happened today. Some of the advertorials seep into my consciousness or my plan of action through interruption.

There is an annoyance factor in ads, but ads on which we were running our experiments were essentially search ads. In response to a search on a mobile platform, content will be shown, and most surveys show that the search ads are the least annoying because they are tailored to the search keywords and they actually turn out to be useful. If I want to take my wife out for dinner I would really like a nice recommendation for a nice place which has the characteristics of the thing that I’m searching for and I find that useful. So what if it’s sponsored, if it’s good I’ll go there, so search ads are useful.

I think when advertising seeps into our actions through interruption, any positive aspect of the signaling could be contaminated by a negative effect of annoyance, but I think in search platforms the negative effects of annoyance are diminished and therefore the positive aspects of signaling has a higher chance to bubble up to the surface.

AURELIE: It’s the whole “user-initiated versus intrusion” paradox. Besides that, you mentioned this movement towards increased personalization, micro-targeting. Let’s project a little and fast forward to 2020. Can you describe for me a specific advertising scenario that we are going to be able to deliver a few years from now with all this increased data and computer processing power and artificial intelligence? Let’s think about the user viewpoint and the advertiser viewpoint, so if you could describe what’s a good advertising scenario for the user at that point in time with an example and what specifically about that scenario would deliver a good ROI for the advertiser and maybe they’re different scenarios, or maybe ideally it’s the same scenario that’s valuable to both the user and the advertiser at the same time, you tell me?

HARIKESH: Of course, I can describe two different scenarios where one is great for the user but not necessarily for the advertiser, unless the advertiser discovers a better business model, and one which is great for the advertiser but possibly for the user as well. The one that would be great for the user but not necessarily for the advertiser would be voice-enabled search. The ability of Alexa or Google Home or Okay Google or any of these other voice-enabled chat bots to now understand our questions and do natural language processing is just mind-blowingly awesome.

It’s quite friction-full, if that’s a word, to go to Google and input my keywords and search. Why can’t I just wake up in the morning and sit in my room and say, “Hey Alexa,” or, “Okay Google, find me a particular kind of jeans and add it to my app and I’ll buy it there.” In other words, why can’t I speak my search into the atmosphere and that search result would be delivered and actions would be taken? I think that’s going to happen with the way voice recognition is going and the increased willingness of consumers to put these devices in their homes, which Amazon Echo has shown us is possible.

So, search is going to shift to voice and that’s fantastic, it’s going to be great for the user and actions could be taken against that. The reason it may not be that good for advertising is that right now search advertising is so good that right after the search I can show you an ad on the browser or on the mobile phone but right after a voice-based search it’s hard to show an ad. Where will I show an ad, will I show it on the mirror in my bedroom? It’s going to be tough and that’s going to be a very intrusive experience.

Probably that kind of search advertising is going to die out; then the advertiser essentially has to find a different model to deliver that content. And the platform also, in this case Google or another platform, may have to find a different model. Probably Google as a platform can extend that voice-based query into a transaction by facilitating the transaction of actually buying this elsewhere, but it’s harder to see how the advertiser will develop a business model.

Maybe they’ll have to pay Google more money because it was really Google that captured and facilitated the transaction. Right now the transaction doesn’t necessarily occur on Google, only a lead is sent by Google so you pay less for that but the transaction occurs on your website or somewhere else. The power of the platform that owns the voice and has the ability to facilitate the transaction may become bigger, which may be bad for the advertisers because of the rapid dual-polarization of the ad markets, with Google and Facebook.

AURELIE: Yeah, It’s owning a completely new channel effectively, and that’s why there’s so much competition in this space.

HARIKESH: Correct. The things that’ll be great for advertising and marketing as a community is VR, and within VR I would say augmented reality. The ability to superimpose images, superimpose suggestions and things like that into what the user actually sees, is just incredible and, as human beings we process most of the information visually. To be able to see myself in the mirror wearing that suit makes me more likely to buy that suit and to be able to see the lamp that I’m thinking about as I look at the table in my bedroom by augmenting the view that I’m seeing in my bedroom with an image of the lamp, etc. Keep in mind that all of this is through my own device, and my past data including where I looked at, will be tracked. So now what lamp I’m going to show can be completely micro-targeted and personalized through algorithms, which checked through incredible amounts of data and delivered that image in real time.

So, it’s going to be micro-targeted augmented reality and it’s very obvious that that’s coming and that’s going to be great for advertisers and increase the amount of ad serving opportunities by a large amount. Not sure how consumers will react to that, they might find it too intrusive but if industry can find a way to add it in a useful way this might be a really big market.

AURELIE: You made me think of the new version of the old sales technique of putting the product in a consumer’s hand and, once you put the product in the consumer’s hand they’re much less likely to give it back to you so they will end up buying it. It’s almost the same that you’re describing here: visualizing yourself with the product in your hands or on you or in your house so that you are committed to buying it.

HARIKESH: That is true, absolutely yes. It would be a big deal.

AURELIE: That will be fascinating to see too. So, if you had a magic wand, and could change one thing in growth marketing, what would that be?

HARIKESH: I would say it is the way in which advertising is assessed and return is assessed, and I think the best-in-class companies and the best-in-class advertisers and the best-in-class brands do a superb job of thinking through assessing advertising and measuring return. But at the middle of the stack and at the bottom of the stack there is just too much confusion about data and complete defiance of logic and too much confusion between correlation and causality: just because sales occurred or did not occur after a user saw advertising does not mean that the advertising caused or did not cause that sales.

Event A occurring after Event B does not mean that Event B caused Event A. This is the first thing that we learn in logic and there is too much of that. I think we can change that to thinking a little bit more formally about trying to measure. Usually a measurement revolution precedes a decision or managerial revolution, so we’re probably moving towards better assessment and methods for decision-making. But there’s a little bit of the academic in me, it makes me a little frustrated. So, that’s what I would change.

AURELIE: For sure. Not just academics get frustrated about faulty measurement, because it leads to faulty decisions.

HARIKESH: That’s right.

AURELIE: Well, thank you so much for your time Harikesh; it was fascinating to envision the future of marketing with you.

HARIKESH: Thank you so much.

Love this interview?  Please share! And, don’t forget to follow Aurelie @ItsAurelieG and Harikesh @jd_corporate and http://fb.com/jd.cominc