Maximising influence on social media

Celebrity influencers may not be the most effective way to maximise the spread of a marketing strategy, says MIT Sloan Professor and author of The Hype Machine, Sinan Aral  

How do marketers choose a set of influencers to maximise the spread of an idea or behaviour in society? Popularised in The Tipping Point in 2000, research into influence maximisation has focused on demonstrating real results. The problem was first formalised by computer scientists Pedro Domingos, author of The Master Algorithm, and his student Matt Richardson, now at Microsoft, in 2001. Since then computer scientists and marketers alike have proposed solutions to influence maximisation with increasing sophistication, and these solutions have been implemented in industry.

Selecting influencers

One obvious approach is to choose the most popular people, with the largest following – celebrities. That’s where influencer marketing started, with celebrities like Kim Kardashian who have vast reach. Picking the most popular people is logical, sensible, and more effective than choosing influencers randomly. But there are drawbacks to this strategy, as marketers discovered over time. First, popular people tend to be connected to one another, so their follower networks overlap, creating redundancy in their influence. Second, the cost of their influence is high. Kim Kardashian charges up to $500,000 USD for an Instagram post, making her influence cost-prohibitive. Finally, it’s difficult to predict the success of any given influencer message – their effectiveness is what marketers call ‘high variance’. So a strategy that deploys a portfolio of highly paid influencers is inefficient, meaning the influence per dollar is low.

An alternative strategy later emerged based on the friendship paradox. The friendship paradox describes the understanding that: ‘Our friends tend to have more friends than we do.’ Discovered by sociologist, Scott Feld, this pattern arises because people with more friends are more likely to be connected to others, so the friends of randomly selected people tend to be highly connected. A marketer can use this pattern to select influencers in a way that reduces redundancy or the doubling up of influencers, and enables their influencer marketing campaign to work when the social network is not known or is costly to obtain, such as in public health interventions in rural villages.

To find effective influencers with this strategy, a marketer or public health official employs a two-step process. First, they take a random sample of the population in which they wish to spread an idea or behaviour. Then they randomly sample the friends of those who were sampled (randomly) in the first step. This two-step process identifies people with a lot of connections who are themselves spread out over the network. It’s an approach that identifies highly connected people who are less likely to be connected to each other.

Case study: adoption of multivitamins in Honduras

Nicholas Christakis [a Yale University Professor] and his team used this strategy to spread adoption of multivitamins in 32 rural villages in the Lempira district of Honduras in 2012. They divided the villages into three groups of nine randomly selected villages, then assigned each group to three different influencer marketing campaign strategies to spread the use of multivitamins.

In the first group, the researchers seeded the most popular individuals in the village – the top 5% – with a bottle of 60 multivitamins, health information about the vitamins’ benefits, and tickets to redeem multivitamins that they could share with people they knew in the village. In the second group, the researchers provided influencers with the same products and information but seeded the network according to the two-step process – they first randomly sampled 5% of the village, then randomly sampled a single contact of each of the people in the first sample. In the third group, as a control, they randomly gave 5% of the population the vitamins, health information, and tickets to share.

The results confirmed that the two-step process, which chose the friends of the randomly selected individuals as influencers, significantly outperformed the other two groups in terms of the number of tickets that were redeemed for vitamins at a local store. In the two-step influencer villages, 74% of available vitamin tickets were redeemed, while only 66% and 61% of the vitamin tickets were redeemed by those in villages targeted by the other two strategies.

Limitations and the effectiveness of the ‘ordinary’

While this strategy is certainly compelling, it has weaknesses as well, as became apparent. First, the more influential someone is, the less susceptible to influence they are. Whether they are the most popular celebrities or are chosen by a two-step nominating process, they tend to be less susceptible to the original idea or product and to cost more to activate.

Second, there is a trade-off between popularity and engagement. The more followers an influencer has, the less influential they are over any given follower. Remember Dunbar’s number? Human beings have a hard time meaningfully engaging with the massive audiences some people garner on social media. So as their networks grow, their engagement with their followers weakens. Industry research has confirmed, for example, that Instagram influencers with more followers get fewer likes per follower on their posts, as their audience feels less intimately connected with them than the audiences of influencers with fewer followers. The average number of likes per follower is 8.8% for influencers with 1,000 to 5,000 followers, 6.3% for influencers with 5,000 to 10,000 followers, and 3.5% for influencers with more than 1 million followers. As someone becomes more popular, they lose their grip on the attention of their followers.

A model proposed by Duncan Watts, Jake Hofman, Winter Mason, and Eytan Bakshy suggested a different strategy of seeding those they called ‘ordinary influencers’, those who have fewer followers but more engagement per post, at a lower cost. My research at MIT with Paramveer Dhillon [now at the University of Michigan] confirmed the effectiveness of this approach. Our models, tested on real social media data, showed that ‘optimal seeds . . . are relatively less well-connected and less central nodes, and they have more cohesive, embedded ties with their contacts.’ Our research, in other words, was pointing to the importance of ‘microinfluencers’ and ‘nanoinfluencers’. And that is exactly how the industry has evolved in recent years.

One key takeaway from the research on influencer marketing and microinfluencers is the importance of attention. The reason a portfolio of diverse microinfluencers can outperform celebrity influencers is because they spread attention over segments of the network without redundancy. Although their reach is smaller, they have a firmer grip on their followers’ attention and can therefore generate more engagement per follower at a lower cost.

This is an edited excerpt from The Hype Machine: How Social Media Disrupts Our Elections, Our Economy, and Out Health – and How We Must Adapt by Sinan Aral (Currency, 2020).

Sinan Aral is a Professor at MIT Sloan School of Management.

You may also like...