Can Algorithms Predict the Shareability of Content? Should They?

Unruly Media have just released Sharerank, a "proprietary algorithm which allows advertisers to accurately predict the ‘shareability’ of a video, before it is even launched" and so maximise the impact of their content by calculating in advance the amount of earned media a piece of content is going to secure.

The system uses data from their Viral Video Chart overlaid with viewer response data (over 10,000 data points). Multiple regression analysis showed the relationship between viewer responses and actual share data and enables the identification and contribution of factors that impact on shareability (psychological, social, content triggers and valence). It's claimed that this can correctly predict shareability 80% of the time, a figure that is predicted to increase as more data is added and the algorithm gets smarter.

It's interesting work, but my first thought about it (a question that Econsultancy also asked) was around the risk of clients wanting to create content and ads for the algorithm. Unruly have stressed that the formula is there to help (and not replace) the work that creative agencies do by providing context and insight into the best sharing triggers and content characteristics that brands can use. But it's a thin line between using it in that way and stepping over to using it earlier in the idea generation process and the temptation for clients to apply some degree of predictive capability to something so inherently uncertain will no doubt be very strong.

Serendipitously, the next article I came across after reading about the algorithm was Sarah Carter's excellent piece on renowned BMP creative John Webster (whose creative work is memorable to me and many others more than 30 years after it first appeared). "It's no coincidence", says Sarah, "that Webster's greatest ads were often the ideas that were added into research as the wildcard, 'nothing-to-lose' option, but emerged as the most successful. Most groundbreaking ads, almost by definition now, seem to me to emerge in spite of, rather than because of, regimented creative development systems and processes".

It's a certainty that algorithms will be playing a much larger part in the future of our industry and will touch many areas. But it's important that we remember, I think, their limitations and the value of combining machine-driven learning and optimisation with more human-powered creativity and inspiration.

Comments

[Clay Parker Jones]

Hey look! Commenting like it's 2006. :)

I'm on both sides of this; to wit, I have a post sitting in my drafts called "Data and the End of Creativity". Part of me believes that we can't, and we shouldn't, use machines to help engineer content that spreads. That part believes in the core creativity of humans, and in the magic that happens when an agency (or whoever) magically gets it right.

But the other side looks at organisms like Buzzfeed, and sees bona-fide engineering of success in action. Perhaps they're not *actually* mechanizing spreadability, but their editorial team has certainly become machine-like in terms of repeatability and pattern recognition.

So I'm mostly certain that this task *can* be done. Whether we want it done is another thing.

[neilperkin]

@clay heh. Nice point about BuzzFeed etc. I can see the value in being closer to understanding what spreads and what doesn't. It's a natural evolution for editorial teams to get much closer to the metrics. I guess my concern is that it then becomes so easy to start chasing the numbers which inevitably impacts creativity I think. There are no doubt many triggers to spreadability, but originality is also one

[Rachael]

Morning,

This certainly provides a lot of food for thought - I suppose I shouldn't be surprised that this has come about, what with the vast amount of data now available to begin working these things out and the amount of time marketers spend monitoring share-ability anyway.

But the thought of creating content becoming so much more formulaic rather than creative feels rather sad.

[Mark Earls]

Great stuff Neil. There's another POV as well.

If by "spread" we mean social diffusion of the classic sort, then this approach is by definition looking in the wrong place - at the THING rather than the PEOPLE.

It's genuinely impossible to predict what'll spread because this is primarily to do with people interacting with others who are interacting with others, not with qualities of the thing itself.

That's why "lighting fires" approach etc is so important - helping yourself be lucky.

It's not a thing-thing, it's a people-thing http://bit.ly/VpMQXh

[neilperkin]

That's a great point Mark. And I suspect that the circumstance/situation/context is just as important as well