BARB and the future of data integration
BARB is now running two prototype fusion contracts. Here, MEC's David Fletcher examines what this means for anyone with a stake in the TV measurement body
Nothing says 'notice me' quite like 'a billion'.
In the latest of BARB's TV Player Reports, the first stage of Project Dovetail reported total UK online TV viewing for week ending 12th June 2016 at 1.18bn minutes.
The TV Player Report (TVPR), has been in production since August last year and has steadily increased scope - which players within which platform - since inauguration. There are nine players, from All4 through to UTV Player, across four platforms (web, iOS, Android and games consoles) and two viewing types (live, on-demand).
Each requires a different set of coding to be developed, implemented, tested and audited for quality before being reported. That's a 72 cell matrix of technology challenges. By my reckoning the TVPR is now reporting on 25 of these - although undoubtedly the largest elements have been prioritised so we are well over half way there in terms of total audience.
For anyone with even a cursory understanding of UK television the results make for intuitive reading: EastEnders and Made in Chelsea top the on-demand charts; Sky Sports channels dominates the list of the most live-streamed (indeed accounting for over 10% of total player viewing); there are some tweaks in order by platform but not so much as to make a big deal of it; and so on.
It's true that all these numbers remain dwarfed by the vast majority of TV-set/STB-based viewing that still dominates consumption, weighing in at c.90bn minutes a week. But as the reporting scope of the TVPR technology extends, and with the natural grain of consumers extending viewing via new devices and platforms further, so the importance of these numbers in understanding the whole picture also increases.
And it's the whole picture - whether you're a broadcaster, programme maker or advertiser that's the one you really want to prioritise. Kids apart, mainstream viewing typically bears only limited short-term repeat viewing, so whereas you might find a show online to re-live a particular dénouement, the chances are that you're watching online/on demand because you missed the programme or are unable to view the TV set. This means that what we get from the TVPR is likely to add audience - to your channel, your show, your campaign.
Which brings us onto stage two on Dovetail - how do we join together the event-level data of the TVPR with the 5,100 home panel that defines the larger landscape of viewing.
At this point, stuff gets technical. Increasingly the advertising world is driven by algorithms and engineers. Whereas the principles are, on a good day, expressed simply enough, the details are often hidden either behind "proprietary", secret-sauce approaches, or through an impenetrable forest of acronyms, jargon and PowerPoint 'smart-art' (often neither!). Worse, it's difficult in the real world to invest time and effort in testing to identify the best approach.
So thank goodness that BARB is having a go at making this easy to understand as well as getting it right.
Getting things right has always been a challenge for industry currencies whose mandate needs to span the needs of both buyers and sellers, with competing interests within each side as well as across the central divide. The status quo inevitably favours some more than others, so any change can be problematic, and of course these innovations are never without cost.
BARB - and by extension the wider TV industry - has the benefit of some unflattering history. When - under previous management, it should be stressed - a new contract's panel deficiencies led to overnight falls in reported ratings for younger audiences, the whole industry was in uproar.
The BARB we have today is not only thinking further ahead, and with less step-changes in the supply chain, it is consulting more actively and with customers who understand the importance of getting things right.
Stage two of Dovetail is the awarding of two, competing, data fusion prototypes with different research specialists, to be independently evaluated, and then used to set the specification for a future standard.
This is necessary because data fusion is as much art as it is science. The principles are straightforward enough - I have two data sets and I make them into one combined dataset by allocating each record in the smaller data set to a record (or selection of records) in the larger set, based on some common thread that identifies both sides as a meaningful match.
As the IPA points out, the success of the integration depends on the quality of the two data sets and the quality of the algorithm which matches respondents from the two datasets.
This is where the fun starts. BARB's 5,100 homes are richly coded for a wide range of demographics - but the TVPR data is demographically agnostic. We might reasonably assume the characteristics of an iOS Made in Chelsea viewer, but a primary purpose of BARB is to do precisely this characterisation for us - as soon as we build such assumptions into any algorithm we pre-determine the outcome and the research loses its core integrity.
Other approaches will doubtless include data that BARB already collects from panelists (such as device ownership and likelihood to view on-demand/stream); related device analytics data (location data to yield regionality, geodems); third party data matches etc. Each comes with a mix of strengths and limitations - some of which will get ruled out early, others only through testing.
Contracts awarded to Kantar and Nielsen, to be independently reviewed by RSMB, will, in the words of the BARB press release, "allow BARB to determine which data fusion approach is best suited to the needs of its customers."
All BARB's customers should look keenly to the outcome and interrogate the results closely - the more we all do this, the better the outcome.
David Fletcher is chief data officer at MEC