Nielsen OCR has certainly got agencies and publishers talking, as reported by AdNews last week. We are data evangelists and so a move that seeks to encourage the growth of digital isn’t to be dismissed lightly. But it is our view that the behaviour of audiences online is of more value to brands than the simple parameters of age and gender. So with a little less conversation and a little more action, we put it to the test to find out how accurately age and gender correspond to interest.
To do so, we measured the demographic make-up of 15 interest based audiences on the assumption that they would display a strong age and/or gender signal. The categories are set out in the graphs below but included Justin Bieber, Rihanna, Hair loss, Pregnancy Tests and Student Loans. A pseudo-campaign was then created where 1 x 1 pixels were served, appended with the Nielsen OCR tag, to any user who visited a site on the Exponential network who had one of the 15 behaviours in their cookie profile.
The chart below shows the 15 Interest Based Audiences plotted in terms of their predicted age and gender make-up.
And this one shows the results
We found that Interest Based Audiences have a gender bias along a sliding scale rather than fitting into discrete categories. So whilst four out of the five audiences predicted to be predominantly female did have a majority of female users, only the Rihanna audience having very slight male bias (50.4% male), some male interest was highly underestimated – Justin Bieber was surprisingly only 52.3% female. When you look behind the percentage average it’s clear that there is potential for a significant number of wasted ads and missed opportunities to reach the desired audience.
Meanwhile, of the eight audiences predicted to be male, seven were accurately predicted as such and had at least 55% male users, the exception to this was the Hair Loss audience which had a female bias (52.6% female). This fact may suggest women are viewing Hair Loss content on behalf of men.
When we looked under the hood at age assumptions, we discovered a similar pattern. While the average age of the majority of the audiences (10 out of 15) fell between the predicted age ranges, the detail of the age data threw up some interesting insights that would have been missed if we were just looking at averages. In Hair Loss, for example, we found that the highest indexing age groups comes before the 30+ range that was predicted indicating that perhaps men tend to search for solutions at the first sign of hair loss in their late teens to late twenties. Another audience that had a younger than anticipated demographic was Luxury (Cars), a straightforward explanation for which is that while buyers of luxury cars may well be more likely to be male 30+, users looking at luxury car content are more likely to be male 18-34. This could suggest that the Interest Based Audience would not necessarily be the best audience for an ad campaign for a luxury brand, unless it was overlaid with further data driven insight.
Our study, which was conducted in the US but which we will shortly be replicating in Australia, was seeking to confirm or challenge assumptions and a timely reminder that whilst we can tailor behavioural targeting to reach the audience desired by a brand – the right audience may not always have the age and gender we might expect.
Because above all else data is only human in so many regards. It takes people to design and build the technology that collects and organises it; it takes people to understand its limitations; and it takes people to ask it the questions that lead to meaningful insight. Then, when all that is done properly, it takes people to recognise just how much it can tell us about people.
Smart brands will continue to be guided by age and gender but should be diving deeper into what their target audience are actually doing, watching and buying. Because whilst its trues that Nielsen OCR is progress but certainly not the Messiah, the devil – and real insight driven results – is always is in the detail.
Ben Maudsley
Country Manager
Exponential