London’s Social Media Week got off to a good start this morning with an examination of the lies social media statistics are telling and how companies can generate better insights.

Associate director for social at OgilvyOne Worldwide, Karin Robinson, outlined three problematic areas within social analytics. “We haven’t been thoughtful enough or considerate enough about what social insights should look like,” she says. Insight doesn’t equal data, it requires an understanding of what that data means.

She points to demographics as a key area that social data fails to represent accurately. Twitter is a problem in this area because its data is thin and often skewed in algorithms. Facebook, on the other hand, has brilliant demographic information, but most of it is hidden behind privacy settings. Only .45% of Twitter users identify their age in their profiles. Location settings are notoriously untrustworthy unless input manually, and even then, can be controlled by the user.

Additionally, she points to sentiment analysis as an area where insights can yield a better interpretation of the situation than raw data. Sentiment is hard to code. Robinson says, “Sarcasm is really hard for computers,” as just one example of language causing problems for social media measuring tools. To accurately understand sentiment, brands have to either examine every comment or at least check a random sample of data to ensure their tools are not failing to code things accurately.

Finally, Robinson outlines the problem of ‘dark social.’ Like dark matter, no one seems to know where exactly it is or what it is comprised of, but it makes up most of the social universe. About 71% of all referrals made online come from dark social – private messages, emails, Whatsapp conversations and the like. “It’s incredibly powerful,” she says, only it’s the least quantifiable aspect of social media analysis.

These problems amount to a mountain for brands to climb when seeking to make sense of social media statistics. But Robinson says companies can still generate effective insights. She advises they remain sceptical and always test data in context. Dark social may be pervasive and statistics may easily seduce those responsible for social media, a bit of understanding goes a long way to creating real insights.

Follow @communicatemag for more reports live from London's Social Media Week.