By Tanya Rosenblat
In 2009, the military’s Defense Advanced Research Projects Agency (DARPA) conducted a competition to find 10 moored red balloons at 10 secret locations within the continental United States. A team from MIT won the $40,000 prize by using an ingenious social marketing campaign: participants could sign up on the team’s website and refer friends. These friends, in return, could refer others and so on. Whoever found a balloon and submitted the location to the team’s website would receive $2,000 if the team won. If that finder had been referred by someone, the referrer would get $1,000. The referrer’s referrer would get $500 and so on. It turns out that the payout for any balloon was always less than $4,000 and therefore the prize money was sufficient to pay everyone (hint: $2,000+$1,000+$500+… is a geometric series).
The winning team used a technique which is closely related to social or “multi-level” marketing incentives. For example, Dropbox offers users some extra storage space for referring their friends. Vonage and other phone providers offer a free month of service for every successful referral. Google’s Gmail service was by invitation only from 2004 until 2007 when it was opened to the general public. Every referred Gmail user received a small number of invitations to invite friends. Google used the same strategy for many other new services such as Google Voice in 2009 and the Google Glass explorer program in 2013.
Social marketing has become significantly easier with the rise of social media platforms. For example, you can now easily refer a Facebook friend or Twitter follower with the click of a button. This raises an interesting question: how powerful are these social media campaigns compared to traditional advertising such as TV and radio ads or search advertising? Under what circumstances would a company use social marketing instead of broadcast advertising?
On some level, it is not clear why social marketing would ever be superior to broadcast advertising. For example, assume that a French cheese manufacturer wants to send a message to 1,000 potential new customers in the hope of attracting 100 of them as regulars. Also assume that it costs one cent to get someone’s attention for 10 seconds (enough to deliver the smelly cheese ad). The company can pay a TV station $10 to deliver the ad to 1,000 different customers, or it can pay 100 intermediaries 10 cents each to refer the cheese to 10 of their friends. The messaging cost in both cases is exactly the same.
However, social marketing can be superior to broadcasting when referrers have private information about the preferences of their friends. For example, assume that only one out of the 10 friends of an intermediary is a gourmet who appreciates smelly French cheese. In this case, social marketing can reduce the total messaging cost dramatically. For example, each of the 100 intermediaries might only want to tell their gourmet friends about the French cheese. Instead of sending 1000 messages, the messaging volume has decreased to 100 messages.
This example demonstrates that social marketing should be most successful for niche products that appeal to a subset of the population. In such a case, referrers can target their recommendations to those friends who are likely to enjoy the product, which reduces the total messaging costs. This can make companies better off (because they have to spend less on advertising) while also reducing annoying ads (because only gourmets hear the message).
However, companies have to resist the temptation to make their social marketing “too viral.” Imagine, for example, that Google had given every early Gmail user 500 referrals instead of five, or (even worse) would have paid users large sums for referring their friends. While these incentives would have made the referral program more viral, it would also have interfered with targeting: early users might have been tempted to refer any friend in order to get a promotional payment even if the referred friend was only likely to become a marginal Gmail user.
My coauthors and I are currently testing this theory using a field experiment. Experiments have become important tools for economists to check whether their theories can explain real-world behavior.
In our case, we looked at food trucks in California which rely heavily on word-of-mouth advertising to build a customer base. The most successful food trucks tend to offer highly distinctive cuisine such as Korean fusion food, or vegan cookies. Therefore, they fit our model well: clearly, not everyone is a vegan but we usually know who among our friends is a vegan and might be interested in trying out the truck.
In our experiment, we distribute discount coupons electronically to a Truck’s Twitter followers. They can redeem the coupons themselves or send them to their friends. We vary both the maximum number of coupons per Twitter follower as well as the extra financial incentive for referring a customer. Our hypothesis is that a food truck can do better by using a “Gmail-invite” strategy: by limiting the number of discounts per person and the financial rewards for referring friends, the referrer has to think harder about whom to tell about the truck, which generates higher-quality referrals.
This experiment relies heavily on modern web technology: we scrape Twitter in order to pick up referrals, and we collect data from trucks through tablets and our own Android app (“TwitterTrucks”). Fortunately, many of my colleagues and students at the School of Information have a computer science background, which makes it much easier (and more fun) to design these types of studies.