Nearly 100 years on from the invention of the television by John Logie Baird, TV is now a fusion of audio visual experiences, delivered to internet-connected PCs, laptops, smartphones, tablets, games consoles, set-top boxes and DVRs and yes, not forgetting, TV sets. And, it’s delivered linear, time-shifted, place-shifted, catch-up and on-demand.
Our role is to accommodate as much of this proliferation as we can and transform from TV to TV, from TeleVision Audience Measurement to Total Video Audience Measurement. Yes it’s still Television, TV, but it’s TV in all of its forms. Kantar’s measurement system, our measurement blueprint, is delivering this transformation.
In this era of not just big, but colossal data, I make no apologies for stating that now and for the foreseeable future, a high quality, representative panel remains at the heart of our audience measurement system. Undoubtedly, there’s an important role for big, behavioural data, but big data must be anchored to our knowledge of how individuals behave. Under conditions that we are certain that we control and fully understand.
Measurement at its core
We know that the TV set will continue to account for the majority of TV viewing for years to come. So we will continue to deploy the best PeopleMeter technology to monitor what’s being viewed on each TV set in the home. The design of the hardware evolves, but philosophically we believe that for the core panel a dedicated, modular TV PeopleMeter will deliver the optimum balance of precision, consistency and flexibility.
How is content on the TV identified? Audio matching (or fingerprinting) is still the industry-endorsed technique, given its independence and comprehensive channel coverage. We use a fingerprinting algorithm that works, and is proven in many markets and many environments. It’s an algorithm that we are developing to deliver finer measurement granularity and to be more resistant to the audio compression and time lags that come with digital delivery of content.
Alongside fingerprinting, there is a growing role for audio watermarking. Our watermarking engineers are without doubt the leaders in their field. And we expect to see more and more markets where watermarking is deployed. Sometimes this is as the primary detection technique, but often it’s for specific purposes such as isolating simulcast. The latest incarnation of our watermarking technology has been developed specifically to identify pieces of content that are not part of the standard broadcast chain. From on-demand programmes to dynamically or geographically served ads. And there are other possibilities too. Imagine re-defining the exposure to TV ads down to viewing for at least two seconds. That would make TV ad exposure equivalent to typical Online Video metrics, so like-for-like comparison.
Capturing online behaviour
Not only is a representative panel at the heart of core TV set measurement, but we believe that panels play an important role in measuring internet-connected devices. This may seem a little perverse in the age of enormous server-side datasets, but the panel tells us about individuals’ behaviour and therefore demographics and duplications.
We’ve learnt many, many lessons on our journey and one of the most important is the challenge of panellist recruitment and compliance. Homes just have so many connected devices nowadays. This has been very much front of mind with our latest technology, codename FocalMeter. FocalMeter is installed within the home network, a bit like a router, to monitor home network traffic. Once installed, all new devices joining the home network are automatically registered – tablets, smartphones, smart TVs and computers. FocalMeter will be part of the recently-announced TVOV Total Video system we are setting up in Norway, and more countries will be announced soon.
Using census data for more granular insights
Panels are a core asset within our TV measurement system – a source of truth. But of course now we have access to, and can substantially enhance our measurement, with census data – or more accurately big, behavioural data. We’ve been working with Return Path Data, RPD, for over ten years now. That’s over ten years refining the algorithms needed to clean the set top box and server data, suppressing those occasions when the box has been left on and no-one is watching. We also know that we can accurately model individuals and demographics from a reference dataset.
And our experiments, for example in the UK and Pakistan, have proven conclusively that we can integrate larger volumes of RPD into currency panel data. The resulting merged ratings are more stable and more granular, substantially reducing the number of zero viewing cells.
The other key piece of census data in our system is of course online server-side data generated by tags installed on the web players. This is very large scale, device level, behavioural data which helps us in several ways. The census ratings data are interesting in their own right. Controlled implementation of sophisticated video streaming tags on web players allows us to generate consistent and detailed device-level ratings data at the census level. This has been an important step in the UK to show audience volumes, what BARB calls its TV Player Report. The key point here is that the census data are consistent from web player to web player, channel to channel and can be independently verified so that it’s an even playing field.
Secondly, the tags enable us to keep to a minimum the installation of software on our panellists’ devices. We can also isolate our panellists within the census data and extract their viewing behaviour, so we can see it on the same basis as their traditional TV viewing.
The best of both worlds
So, we have high quality, representative panel data on individuals - and we have access to large scale, device-level behavioural data (RPD and/or web player data) to measure volumes of viewing, even for niche content. Our ultimate goal is bringing these data sets together into a fully integrated hybrid measurement system, so that the measurement system is based on the best of both worlds. Saying hybrid is one thing, and we’re aware that the word hybrid has been over-used and on occasions mis-used. Actually doing it, building a fully operational system that combines the complex data integration mathematics with a production system that delivers integrated ratings data reliably, every day, is a huge challenge. But we’ve done it for the Total Video service we are delivering for the Dutch TV market. We’ve built and we’ve delivered it, and we’ve done it in a way that is transparent and open to assessment by the industry.
So while experience tells us TV (Total Video) Audience Measurement is complex and challenging, we can confidently say we have the technology and methodological blueprint to deliver it.