What Facebook thinks about market research

Neither, as it turned out – the presentation was affable, helpful, and revealing about Facebook’s approach to research. Facebook has two goals in research, according to Bruich: it wants to understand the world around it and how its users lives are changing, and it wants to understand how consumers interact with one another and (crucially) with advertising on the site. To do this it looks at three kinds of data – basic usage data (who looks at what when), “revealed preference” data (people’s interests and social networks) and finally solicited data – quick polls and surveys.

Bruich’s presentation answered many of the basic questions about Facebook data. Is it representative? Well, it’s as good or better than polling data. Is it predictive? Facebook users’ declared intention to see a movie strongly correlated with opening weekend take. Is it engaging? Participation rates in Facebook polls are sky-high, because the experience is embedded into Facebook and therefore ultra-convenient – nobody is being asked to leave their trusted environment and go into a bespoke survey space.

Bruich offered strong points of view on social media and influence, rooted in Facebook’s own philosophies and the culture of its social network. He feels the great advantage of Facebook over other social media is authenticity, which he identifies with the offline space, people’s real names and real-life connections. He also described – by way of an Oprah clip – how the popular Gladwellian model of powerful influencers doesn’t really work. Facebook suggests that influence isn’t top-down, it diffuses across networks via several thousand small platforms, not one or two big ones. And it turns out there’s no consistent predictor of which content or “likes” will become viral. On the other hand, Bruich claimed that small-scale social proof – seeing that your friends have seen ads – increases recall and purchase intention.

So what should research be doing? Facebook dislikes long, externally located surveys, preferring one- or two-question mini-polls. This obviously makes most sense if you’re sitting on as much data as Facebook is. But it’s a direction many panel providers and research firms are also looking to go in. Bruich pointed out that river sampling – plucking users out of Facebook into a longer survey – is less efficient because it tends to miss light Facebook users, who actually have higher ad recall than the heavier ones.

Bruich joined the chorus of conference sceptics about text analytics, making the now-familiar points about its shortcomings at understanding context. A focus on the chatter itself misses out on its “viral reach” – the potential these mentions have to reach others. He showed a copy of a General Motors survey from the 1930s, pointing out that the questions were largely the same today (he was too polite to mention that the 1930s survey was rather more attractively designed). GM made the results of that survey social by turning them into a “buyer’s guide”, directly connecting the research and the customer. It’s clear Facebook feels that for research to prosper it needs to become social once again. This was the central theme of Bruich’s talk – by focusing on individuals, be they respondents or influencers, research fails to take networks into account.