This post was originally published on the GreenBook Blog on Wednesday, September 5, 2012.
Ever since the popularity of Malcolm Gladwell’s book ’The Tipping Point’, there seems to have been a fascination in some marketing circles for identifying influencers and thereby targeting spend more effectively. This has been called ’influencer marketing’ and almost every social media huckster has a story of how they can identify the influencers and make you rich.
The problem is that most of the efforts to chase the mavens, connectors, and salesmen are based on myths and confusion. In most cases there are not large collections of people (e.g. 10% who influence the rest) who can be predicted to be influential in the next brand success.
There is considerable doubt about Gladwell’s suggestions about how influence manifests itself and about how market researchers and marketers might set about identifying influencers. The two key issues for market researchers are:
- 1. Most of the scales that seek to identify influence use self-evaluation, typically based on work citing Feick and Price.
2. And, most of the studies about inference are post-evaluations of phenomena, with poor predictive outcomes.
The most cited research in terms of influencers is work done by Feick and Price, for example, The Market Maven: A Diffuser of Marketplace Information.
Following on from Fieck and Price’s research a number of scales have been used to identify influencers. However, typically these scales ask people to evaluate themselves and how others see them. The scales tend to include things like ’People ask my views’,’People follow my lead’, ’I tend to buy new products before my friends’ etc. Generally men score higher than women, i.e. they are more likely to say people listen to their views – but there is little validation of these scales, and little reason to believe that men are more influential than women.
The problem with much of the research into influence is that it is based on the human tendency to see patterns, even if patterns do not exist. If a ball rolls down the mountain its path seems entirely predictable in hindsight – but in most cases if we roll another ball it takes a new route – which also, in hindsight, seems predictable.
I tend to illustrate the problem of influence by substituting an outbreak of flu for product adoption. If we look at last winter we could examine a town and work out who caught flu first, last year, and call them early adopters. We could look at who tended to catch it from whom, and we would identify some nodal points, people who interfaced with lots of people and who seemed to be implicated in the spread. We could even ask people who they thought they caught the flu from. However, if we try to use this model of flu ’influence’ next year to predict the path of the next flu outbreak we would probably be wide of the mark. Worse still, if we try to use our flu model to predict the pattern of the next outbreak of glandular fever (AKA mono) we would be even wider of the mark – but this is what typical product influencer models try to do.
Indeed, if you re-read Gladwell you will note that one of his stories is about the spread of Hush Puppies from a small group of people in Manhattan. This group was responsible for a global trend. However, if these people were ’influential’ there would have been other stories about products they had been the early adopters of.
The patterns we see, when we look at data about product adoption or messages, are usually the result of the way information moves through scale-free networks. Items do not tend to pass smoothly through the network, with every node being equally likely to carry the message. When we look at the network, in retrospect , there will always be nodes (i.e. people) who carried more traffic. However, in most cases this does not mean they can be predicted to carry heavy, relevant, traffic in the future, and even when a node carries heavy traffic it does not mean that they are ’influential’ – post hoc does not imply propter hoc – being connected to a lot of people can mean following a lot of people, rather than leading a lot of people.
Note, where nodal models work, in diseases, is when they are hard to catch, such as sexually transmitted diseases (i.e. if you are not intimate with somebody you don’t catch the disease). In these cases treating the vectors of the disease (the people doing the spreading) makes a lot of sense. The analogy of these situations in marketing is more likely to occur in B2B or pharma, rather than consumer markets.
Another issue in influencer model is the difficulty of separating cause from effect in human behaviour. Studies that show liking something leads to purchase could often be used to show that intending to purchase something leads to liking. Reading recommendations might lead to purchase, but equally planning to purchase leads to reading recommendations.
In many ways the current interest in mavens is a lot like the interest in early adopters (from the Rogers diffusion model). Most successful new products are adopted by early adopters, but the vast majority of the products liked by early adopters do not become successful.
It is true that there are influential people, but they tend to be far fewer than people like Gladwell indicated and they tend to be the usual suspects. Television chiefs are often influential in new food choices; sports journalists can often influence sentiment about a manager or player.
Beyond key influencers, e.g. the journalists, the pundits, the fashion stars, the level of influence is pretty marginal. The Economist had an interesting article in 2010 on mining networks to find predictive elements. The article reported that influence does exist, but it also shows that the effects they are reporting are actually very small.
The book ’I’ll have what she’s having’ by Mark Earl provides a really good insight into how ideas spread and why prediction is so hard. The basic mechanic is social copying. We all do it, and we are all part of the system. Things are popular, largely, because they are popular – which is not very helpful in trying to predict what will be popular, or in predicting who will use it.