It’s no secret that sending spices to cooking bloggers, makeup samples to YouTube-tutorial stars or the hottest headphones to pop idols is a great way to increase sales.

But how do you know which recipe maven is going to maximise your product and yield new customers? The most fabulous recipe is no help if no one sees it. A persuader needs that perfect combination of talent — whether it’s in the kitchen, the recording studio or anywhere else — and audience.

Sending free samples to celebrities has always been a popular marketing ploy, and the previously limited number of media outlets made it easy to target potential customers. But today’s consumer audiences are ever more fragmented by social media. With that in mind, we set out to develop a novel new method to predict the most effective persuaders – those whose adoption of a product or service will result in increased exposure for brands.

Previous methods focused primarily on social influence, defined as a change in behaviour that one person causes in another, whether intentional or unintentional. We sensed the need to go deeper, looking at several different forces central to social persuasion, not only social influence but also factors such as entity similarity and structural equivalence. Entity similarity denotes the degree to which two social entities are similar in their individual characteristics. One entity in a social network can use other similar entities as a frame of reference and adjust his or her attitude, belief or behaviour accordingly.

Furthermore, similarity also exists in structural position within a social network. Structural equivalence matters because the similarity in structural position within a social network can affect individual entities’ beliefs, attitudes and behaviours. To predict top persuaders, it is essential to consider these different forces underneath social persuasion. We integrated all of these factors to create a new method of predicting top persuaders that substantially outperforms the prevalent methods used in previous research.

Without getting too far into the weeds and boring you with lots of esoteric math, we’ll note that introducing persuasion probability — the probability that one social entity persuades another to adopt a given behaviour — was a key factor in how our study differed from previous methods.

This of course is all lovely in theory, but how does it work in the real world? We evaluated our method using real-world social network data from a massively popular online roleplaying game. In the game, users select an avatar, form friendships and alliances and complete quests and tasks. One data set contains 1.34 million records of complete message communications among avatars in the online game over 20 weeks, starting with the launch of the game. Each record contains the timestamp of the message, as well as the respective identities of the two avatars who are participating in the communication.

We used this data set to construct a social network of avatars, and had each social entity represented by an avatar. A relationship between two entities exists if two avatars have message communications with one another.

We went a step further and measured the strength of the interactions between the avatars (or entities) based on the number of messages in their communication string. A larger number of messages equaled a stronger relationship between the two.

Once we’d established relationships between avatars, it was time to test our method. In this online world, players can spend real money to increase strength and purchase items like weapons or armour for their inventory. For our research, we chose to collect data on whether avatars adopted a particular virtual item. This item was not a necessity for avatars, so it was completely optional as to whether they adopted it. For each of the 5,162 avatars we monitored, we collected data on whether they adopted the item, and if so, in which week.

So what are the broader implications for business and society? We feel that our method can offer huge value to social network-centric applications. For example, a firm could use our method to predict top persuaders among its potential customers. It could then entice them to use their product or service — think the cooking ingredients or makeup samples we mentioned earlier — and then hopefully see an increase in sales when the persuaders’ followers purchase the spice blend, concealer, enchanted magical artefact, etc. We feel that the $14.8 billion annual virtual goods market — which is expected to continue to grow — offers a prime opportunity for significant and positive impacts for firms that apply our method.

We believe our method also has major implications beyond monetary gain. Consider the potential impacts it could have on public awareness. For example, top persuaders might be able to promote healthy diet and lifestyle habits in a population. Or think of a remote, underdeveloped area where aid workers are trying to foster acceptance and trust of modern medicine. In such a situation, we could create a social network for the population, and then target and convince top persuaders to adopt the promoted lifestyle practice. Similarly, our method could have positive implications for situations requiring infectious-disease containment. Targeting persuaders could mean the disease is contained and preventative measures are taken before the outbreak could reach epidemic proportions.



Xiao Fang is Associate Professor of Management Information Systems and JPMorgan Chase Fellow at the University of Delaware’s Lerner College of Business and Economics and Institute for Financial Services Analytics. His research is in the area of Data Science and Business Analytics. He serves on the editorial boards of MIS Quarterly and Decision Sciences.


Paul J. Hu is David Eccles Chair Professor at the David Eccles School of Business, the University of Utah.  His current research interests include information technology in health care, electronic commerce, data-driven business intelligence, social network analysis, technology-assisted learning, and knowledge management.