Wednesday 10 September 2008

Viral Models are Incomplete

For several years now “viral marketing” has been a part of most marketing campaigns, online and off. Now that the frenzy has effectively died out it’s clear that it doesn’t work the way many of its proponents would have us believe.

1. It tends to be of far lower impact than it’s assumed promise; success tends to be judged in terms of minimally positive ROI. In fact, the vast majority of viral marketing campaigns return negative ROI and only a tiny (really tiny) percentage ever return anything close to the epidemics popularized by Gladwell’s Tipping Point.

2. It’s virtually impossible to predict which items (movies, songs, fashions, etc.) will succeed in a cultural market (check out recent work by Duncan Watts et al. and Luis Bettencourt). Ironically, this is presumably because of the very forces – informational cascades (localized conformity) and network effects in general – that viral marketing supposedly harnesses.

3. Conversion of hub individuals, so-called “influentials”, isn’t nearly as simple or effective as it seems it should be. More research by Watts suggests that normally connected people have a nearly identical chance of starting epidemics as do the highly connected. The important mechanism seems to be the existence of a significant, connected sub-network of easily infected individuals. Also, it’s not at all clear that the basic influentials hypothesis isn’t conflating causation (do influentials cause cascades) and correlation (or are they highly visible people caught in them like everybody else, just slightly earlier). In social markets, which by definition are rife with externalities, correlation and causation are not clearly distinct. But viral marketing flatly identifies influentials as primary causes.

4. The epidemiological analogy isn’t complete. The idea of susceptibility to a cultural artifact is an unexamined assumption, rather than an explanatory mechanism.

My take on these points is that 2) and 3) explain 1). Controversially, I also hold that 4) at least partially explains 2) and 3).

The contagion model viral marketing claims as justification ideally explains (mechanistically) the spread of a contagion through a population. But a key assumption of the model is left largely unexamined. Susceptibility to a cultural contagion (fashion, song, movie, silly online video, etc.) is a parameter of these models, but the mechanism which actually determines this parameter’s value is effectively a black box. Emphasis is put on the spread dynamics and which structures effect the greatest spread, but very little consideration is given to the nature of the mechanisms that determine each agent’s susceptibility. This is a matter of focus and for contagion dynamics – models of how something spreads through a network – this is an appropriate omission. However, for marketing theory, this omission has significant consequences: it obscures the nature of susceptibility to cultural contagions, focusing us solely on the medium of transmission while assuming that the medium is agnostic as to the content transmitted. Network structure is key to explaining how a contagion spreads, if it’s spreadable. But understanding susceptibility mechanisms will help explain what can be spread if the structure is right. As Duncan Watts has said, the ideas that take off have to be right for the society and this comes down to susceptibility. An analysis of the mechanisms underlying susceptibility will help us develop a much better model for viral marketing specifically and cultural contagion generally.

Understanding these mechanisms will in no way guarantee the success of your viral campaigns, however. You’re still open to myriad chance-introducing externalities (the fragility of some types of cascades and the subsequent dicey nature of path dependent process) inherent in social/cultural markets. At best, it puts you in a position to better set success levels, determine rational outlay and generally make your offering more competitive in the cultural market.

In future posts I'll take a stab at some possible mechanisms, mostly drawn from sociology.

No comments: