Future of Media Research: Data fusion – the holy grail?
Bringing data sets together will become the holy grail, according to ITV’s head of commercial research & insight Sarah Messer (pictured).
Speaking at MediaTel Group’s Future of Media Research event at Specific Media last week, Messer said the shift towards data – where data solutions will measure outcomes based on a variety of different numbers (what people are viewing on their TV, to what they are buying, for example) – “brings home how important data fusion is”.
Ipsos MediaCT’s senior director John Carroll agreed that data fusion should be used more often that it is. In his view, fusion has come a long way and is more transparent, though he pointed out that there is only a “niche group of experts who can do this work… and there is no off the shelf solution”.
From the floor, Steve Wilcox, managing director at RSMB, commented: “In the past, data fusion did get bad press. Data fusion techniques have been tested and validated. It all starts with the information you’ve got available to you – so that is what makes or breaks it. However, data fusion is not always transparent to everybody and it is not always the best solution.”
James Smythe, owner of Culture of Insight, was positive about data fusion – saying there is real value in software which can turn masses of data into answers. He hopes UKOM will be able to link with other media currencies, to give a view of reach across all platforms. “Measuring media in its completeness is the way it’s going… we want to be able to tie in with cinema, outdoor etc,” he said.
However, Chris Worrell, European research director at Specific Media, wondered how quickly fusion can react in a world where real-time data is available. “It is going to be a difficult challenge for data fusions to capture the fast-moving digital space,” he said.
Jim Kite, strategic development director at Starcom MediaVest Group, admitted to being a “fan of fusion” and said he is surprised by how slow the take-up has been, especially with companies such as RSMB, which in his view is the “best and most robust way of doing fusion in the UK”.
Although he wonders whether “fusion will be just a pathway to a greater day?”. According to Kite, the industry wants a complete picture and clients are changing their mindset because of digital – “they want to know how many people did something, not just saw the ad,” he said, hinting that this kind of complete view will soon be a reality for TV and press.
First of all I wanted to address the fact that respondent level fusion is not always the right way to integrate two datasets. Fusion is one of variety of modelling solutions to the integration of two datasets. In many ways, fusion could be regarded as a way of configuring or presenting the results of an integration. Its advantage is that it creates a database with a familiar structure, which is transparent to data users. But sometimes the fusion approach doesn’t make logical sense, either because of the nature of the input data or the analysis application. Indeed we use a variety of different models for the integration of the different media research currencies into TouchPoints.
There is also a concern over the ‘black box’ nature of data fusion. ‘We’ve done a fusion so it must be right’ is not the right attitude and ultimately gives fusion a bad name – despite our best efforts there are still sceptics out there! It is the responsibility of both buyers and sellers of fusion to question the performance of fusion and any other modelling technique. Of course all the fusions we have done for JICs (and I include the IPA) have been vigorously evaluated. JICs are not in the habit of undermining their trading currencies with sub-standard modelling techniques. I recognise that such rigor costs time and money and JICs have an advantage here. But as far as we can it is important not to use the complexity of modelling techniques as an excuse for lack of explanation and validation.