Wharton's Kartik Hosanagar discusses his research on brand content and consumer engagement on social media.

In the world of social media advertising, the biggest win for firms is when consumers are delighted by the content they see, want to engage with it and eventually buy something. Kartik Hosanagar, Wharton professor of operations, information and decisions, has co-authored research that takes a closer look at brand posts on Facebook to determine the type and mix of content advertisers should aim for to get results. The paper, “Advertising Content and Consumer Engagement on Social Media: Evidence from Facebook,” which was co-authored with Dokyun Lee of Carnegie Mellon University and Stanford University’s Harikesh Nair, is forthcoming in the journal Management Science. Hosanagar recently joined Knowledge at Wharton to discuss his findings.

An edited transcript of the conversation follows.

Knowledge at Wharton: This paper looks at the increasing importance of something called content engineering. Can you explain what that is and how businesses have started to use it?

Kartik Hosanagar: Content engineering or content marketing is essentially about designing the right kinds of content that might engage consumers. It’s becoming increasingly important on the web today because firms are producing so much content, be it on Twitter, Facebook, Pinterest and so many other social media platforms. Firms have a presence in all of these platforms and are constantly generating content. Some of that content sticks and some does not. The question is, can we figure out what sticks and generate content that customers want to engage with?

Knowledge at Wharton: You point out in the paper that reach used to be the focus, and now companies have moved more towards engagement.

Hosanagar: Right. During the early days of social media marketing, a lot of firms were focused on acquiring followers or fans on these networks. For example, acquiring a lot more Facebook fans and spending a lot on advertising to get these Facebook fans. But over time, they realized that a lot of these so-called followers or fans weren’t actually engaging much with the content.

Now, the value of Facebook for most of these brands is ultimately that they can use these followers to instantaneously spread information about new products, about new promotions. When their brand-loyal customers engage with content by liking, sharing or commenting on it, the friends of these brand-loyal followers also learn about the brand. So, there’s the opportunity for viral spread of the message. But that wasn’t happening as much because while a company might have had millions of followers, only 1% of these Facebook fans were actually engaging with the brand. More recently, the attention of firms has moved from merely acquiring followers on social media platforms to actually getting these fans to engage with them because, ultimately, that’s where the value lies.

Knowledge at Wharton: This paper tries to get at that question of what content works best. I was surprised to learn that this is an under-studied area. Why would you say that is, and what do you try to do differently in the study to get at that problem?

Hosanagar: Given that advertising has been around for a while and there is so much emphasis on advertising communication, you would think that there’s going to be a lot of work done that’s getting to the heart of how should we design content. We were surprised to find that there hasn’t been a lot of systematic study of the subject, and our conclusion is that’s largely because it’s a tough problem.

If you look at traditional advertising, a brand comes up with ads, a marketing person figures out what words to use in those ads, what imagery to use in those ads and that person makes a creative decision that we’re going to talk about the following product features but not these other aspects of our brand. Those creative decisions may well be intuitive. But at the end of the day, it’s very hard to test systematically and really assess whether this makes sense because we are talking about doing this at scale and with unstructured data. That has changed with the web because you’ve got so many firms creating lots of social media messages, and now you have the opportunity at analyzing this at scale by looking at hundreds of thousands and perhaps even millions of messages posted by various brands and looking at which ones people engage with.

We’re able to do that only recently because you have that kind of data at scale, but it’s a hard problem because you still have to process all that content and figure out what is in a message. Is this message about the brand’s product? Is it about the price? Is it about deals? Is it comparing its product to its competitors? Is it emotional? Is it humorous? Is it talking up a brand’s philanthropic initiative? All of this requires a lot of natural language processes and at scale, and we’re only now in the last few years starting to have both the data and the technology to do that.

Knowledge at Wharton: You found a really interesting data set to work with. Tell us about that.

Hosanagar: We were fortunate to partner with a large social media analytics company. This company powers the analytics for a lot of large brands on Facebook. They shared with us a data set that spans a period of about six months and that is every Facebook post posted by 1,000-plus companies on Facebook. These Facebook pages belong to some of the largest corporations, so we’re talking about millions of followers associated with each of these brands. For each of these posts, we knew daily engagement information, meaning how many likes, how many comments, how many shares and how many clicks on these happen on a daily basis over that six-month period.

Effectively, we had in our data set nearly 800 different U.S. brands and about 106,000 messages or posts that these firms shared on Facebook. For each of those 106,000 messages or posts, we had daily information on likes, comments, shares and clicks by roughly 450 million Facebook fans of these different brands.

Knowledge at Wharton: Tell us about some of the key findings for the research.

“The attention of firms has moved from merely acquiring followers on social media platforms to actually getting these fans to engage with them because, ultimately, that’s where the value lies.”

Hosanagar: I run an interesting experiment in class when I’m talking about the study where I ask the students, “What do you think will happen?” I tell them that I have two kinds of variables: content variables that are product informative, like price, deals and so on; and content variables that are about brand personality, like emotion, humor and something called friend-likely, which we use as a variable that represents whether this is casual information sharing along the lines of what a friend might post.

I usually ask my students, “Which one do you think matters?” And usually, my students will tend to pick humor first. Probably about 30%, 40% of them will pick humor. The second most popular tends to be deals. A third one is emotional. A lot of them also pick friend-likely. I see votes split across all of these variables. Now, what we found across 800 different companies from many different industries was brand personality content was associated with much higher levels of engagement — specifically, likes, comments and shares — than product informative content.

Generally, product informative content had lower levels of likes, comments and shares, unless they were combined with brand personality information. In other words, if a brand is sharing information that is about deals or about price, unless you combine with emotion or with humor, you are not going to see very many likes, comments and shares. Within these two high-level buckets, if we drill down into the individual variables, it turned out that emotion was the most influential variable. For example, we found that emotion drove over a 20% increase in the number of likes an average message gets, assuming it does not have emotion already. Similarly, for an average or a typical message, if we add emotion into it, the number of comments went up by nearly 20% again. That’s a pretty significant increase in engagement by adding one new variable into your content mix.

Knowledge at Wharton: You also reported some interesting findings around deals.

Hosanagar: As far as likes, comments and shares are concerned, we saw that it is generally associated with a lower level of likes, comments and shares relative to an average message that does not already have deal in it. Just adding deal information did not really add to these engagement measures. In fact, it took away some of the engagement. But when we looked at the number of clicks that messages were getting, clicks were similar to likes, comments and shares in terms of brand personality content getting more clicks and product informative content getting fewer clicks.

But the one sense in which it was different was deals. Deals did not get likes, comments and shares, but got lots of clicks and lots of purchases. Deals was interesting in the sense that it drove path to purchase, even though it did not drive social engagement through likes, comments and shares. We found that puzzling. I think the way we have made sense out of it is that likes, comments and shares are socially visible engagement. Clicks, on the other hand, are not socially visible. When we see a brand post humorous or emotional or philanthropic content, we like, comment and share it because that’s socially visible and that says something about our own personality. But when we like, comment or share on a deal, we don’t want to appear cheap. But we are happy to click on those and then go pursue the deals and even buy.

“When we see a brand post humorous or emotional or philanthropic content, we like, comment and share it because that’s socially visible and that says something about our own personality.”

Knowledge at Wharton: You also make the point in the paper that companies should not do all one or all another. Even though some of this brand information posted doesn’t get as much engagement doesn’t mean that companies don’t need them or that they can’t be beneficial.

Hosanagar: Absolutely. I think what we advocate is a portfolio approach. If you found emotional content works and you start doing one emotional message after another, at some point, people are going to wonder what’s up with this brand. I think you have to have a mix. All we are advocating is in the mix, while you’re trying to promote product information, price information and just as much of the brand as possible, recognize that consumers are often looking at brands on social media to understand brand personality. Don’t forget the brand personality side. But overdoing just brand personality without the product information can hurt you.

At the end of the day, as a brand you’re trying to use this to ultimately push sales or push loyalty and things of that nature. As I already mentioned, things like deals — even though it’s not getting likes, comments and shares — it’s driving path to purchase. People are coming to your website and buying. Ultimately, you still need a nice mix of content. What we are saying is that if you are going to push product information and deals and pricing and sales, figure out if there are clever ways in which you can pair it with brand personality.

Knowledge at Wharton: That presents an interesting challenge for brands because it’s very easy to compose a dry post on, “This is my product and how much it costs.” But if you’re going to do some of this brand personality content, you have to figure out your personality as a brand. Then you have to be able to communicate that to the people who are actually writing your posts.

Hosanagar: I think there is a sort of philosophical take on how marketing has changed. It used to be the case that marketing was completely a creative discipline. Many people have argued that over the last 10, 15 years, it’s become a quantitative discipline, how everything is measurable and you can use data and optimize marketing. This is an environment where we are interestingly looking at data and analyzing and crunching the numbers, and ultimately we are finding that it’s actually a mix of the two.

Yes, you want to focus on the numbers and that matters. But at the same time, we are saying that brand personality matters. It’s very easy to craft a Facebook message that has just very basic information of product attributes, about product price. It’s harder to craft a message that has humor in it, that has emotional appeal in it. It means that there is still some art left here.

Knowledge at Wharton: Can you think of a brand that is doing a really good job of this, that’s getting it right?

Hosanagar: I have always liked what one of our own Wharton companies, Warby Parker, does on Facebook. They have created a very loyal base of followers on Facebook. They were engaged with those followers during the early days. They would actually post photos on behalf of their customers to help those customers select the right frame. Customers would take photos of themselves in different Warby Parker frames and post it on their own Facebook wall. But Warby Parker would also post on Warby Parker’s wall and get feedback from them. Over time, that has changed as the company has grown. But still, they use Facebook very effectively even today.

Knowledge at Wharton: When brands are posting on Facebook, they have to contend with Facebook’s mysterious algorithm about what kind of posts people are seeing. How did you account for that in this research?

Hosanagar: That was really interesting because if we observed that a certain post got a lot of likes, we did not know for sure if it was because consumers liked that post or because Facebook liked that post and showed it to lots of people. Part of our study was trying to separate, is it Facebook liking a post and showing it to lots of people? Or is it people themselves liking it? We had to reverse engineer Facebook’s newsfeed algorithm a little bit so we knew which customers of a brand were eligible to be shown a post and which ones were actually shown a post. By looking at the difference between the pool of people who could be shown a post versus those who were shown a post, we could kind of say, “Here’s how Facebook decides who to show a post to.” And we could separate that effect of the consumer’s choice.

“We find that if you design the content better, you can increase the reach and engagement of your post, even without having to pay for it.”

One of the interesting findings was that firms can also understand what Facebook does in terms of selecting a post, and this is somewhat analogous to the world of search engine optimization or SEO, where firms are trying to figure out what is exactly going on behind Google’s algorithm. Why is my page showing up for some search queries but others? How do I change my pages so that it will show up at the top of Google search pages? There are some nice parallels there. We have found that it is possible to similarly reverse engineer, not fully but to a good extent, how Facebook chooses posts. Then you can craft your post not just to ensure consumers engage with it, but to also ensure that Facebook’s algorithms will select your post and show it to lots of people.

Knowledge at Wharton: Where do paid posts — sponsored posts or ads — come into play here? All of the social media platforms, including Facebook, are doing this now and seem to be pushing brands in that direction.

Hosanagar: Our study was focused on the organic part of social, nonpaid part of social. When a Facebook page of a brand has a bunch of followers and the brand pushes a post, what drives the engagement of those posts among its followers, assuming no paid promotion on the part of the brand? But I think there’s a nice intersection with the paid portion of social media marketing as well. Firstly, a lot of brands are finding that they’re not getting enough engagement unless they pay Facebook to show those posts. We find that if you design the content better, you can increase the reach and engagement of your post, even without having to pay for it.

Secondly, if you are going to pay for it, the question becomes which of your Facebook posts are you going to promote as a sponsored post? Because all these brands typically promote some of their own posts. Again, our algorithms can tell you which posts are likely to get higher levels of engagement, and those might be the ones you might want to push out more. It comes to emotional content, brand personality content. If you’re going to do informative, pair it with brand personality content and then you can pay for those and promote those additionally.

Knowledge at Wharton: What’s next for this research?

Hosanagar: I think one of the innovative things that we did in this study was to take text content at scale. We’re talking about 100,000-plus messages and applying machine learning techniques to code this content and say, “What is in the post itself?” Which is hard to do for human beings at scale. We can maybe tag 200, 300 posts, but not 106,000. Next, we want to take those techniques and apply them to large-scale analysis of content broadly. We could go beyond text. We could start looking at images on Pinterest, on Instagram. Even with text, we could go beyond the 100,000 we looked at on Facebook and we can look at Twitter, blog posts and so on. More broadly, we can look at differences between these social media platforms. I think that would be a very fruitful direction. We hope to go there. It’s a matter of getting lucky again with a good data partner.