An HKUST researcher has developed a methodology to optimize promotions in online networks. The methodology does not only consider the likelihood of whether an individual is interested in the promotion (i.e., whom to target), but also how it impacts other individuals in the network (i.e., social influence of the target) and the time it is most effective (when to target). Along with Netherlands and U.S.-based collaborators, Ralf van der Lans modeled ripple effects in decisions of users in a massive online game. They showed that their methodology increased the effectiveness of in-game promotions by 60 percent in terms of game usage, compared to traditional targeting approaches.

Social influence permeates the study of consumer choice: we don’t make purchase decisions in isolation but are influenced by others. For marketers, the tantalizing prospect of a way to identify the biggest influencers in a social network could unlock huge profits via better timed and targeted campaigns. As the researchers note, “choices by influential consumers might be considered as more profitable for marketing communications and promotional activities as they would trigger desired actions by other consumers.”

With today’s ubiquity of social media, peer effects are more important than ever—and they leave a digital trace for researchers to analyze. However, uncovering social influence from digital traces is one of the main challenges in network research, because the observation that two individuals make similar choices is not necessarily attributable to social influence. For instance, two friends that simultaneously play an online game may do so for many reasons other than social influence, such as similar interests (birds of a feather, flock together) or exposure to the same information and similar timetables (other exogenous factors). To disentangle these confounding factors, Van der Lans and collaborators divided social networks into “communities” in which people naturally sort themselves. They argued that “Consumers from the same community often have similar interests, speak the same languages, [or] live in the same geographic regions,”. Their real-world data confirmed their predictions and shows that these clusters can be used to accurately predict social influence.

Van der Lans and collaborators applied their method to a network of 25,000 diverse players of a popular online roleplaying game—for which real data are available—over 30 days. Using their approach, they explored how in-game campaigns could encourage gamers to step up their activity, and how this could ripple to other users through social influence. To keep users active, the game runs promotional campaigns aimed at subsets of players, and several of these took place during the study period. As described in the paper, “The regular promotion rewarded users with in-game items (e.g., magic potions and costumes) if they logged in. Impressively, the model reproduced the diverse responses of real-life players to the targeted campaigns. The researchers also compared competing approaches to targeting the biggest influencers in a marketing promotion. While previous studies have assumed that “hubs”—users connected to many others—are the most influential, the HKUST study shows that it is important to target those who are both well-connected and whose friends are responsive to their actions.

Given the similarities between game-playing platforms and other online social networks, marketers can use the insights and methodology of this study to identify latent communities within consumer populations and target users who are most likely to influence others to respond to a promotional campaign.