The future of data measurement

When looking at TV, Connected TV, and everything else, you need big data as much as small data or panel-based data. You want to ensure that every view of a show has representative data, and panels alone cannot guarantee that, and neither can big data by itself. For big brands like CBS, this is not an existential issue, as we are the most extensive TV network in the US. But, when you get smaller and smaller, this representation becomes a core issue. Therefore, you need big data along with smaller samples, and you must cross-check constantly.

During Covid-19, we saw two things happening: First, panel-based measurement completely broke down. The incumbent had a particular way of recruiting and engaging people, requiring in-home visits to ensure that all TVs were wired. These visits were no longer feasible.

 Second, TV viewing – including linear TV – was increasing significantly, but the incumbent measurement was not reflecting this. Therefore, CBS and others supported a whole range of companies such as Ispot, VideoAmp, Comscore, and 605.tv to get new measurement numbers. It is a very competitive environment right now. With small data, you want representation in these panels. You want everyone watching your programmes and buying your products and services to be counted and represented. For example, it does not work for the US if you are not counting Hispanics, who eat food, buy cars, and invest in Telco like everybody else. I am talking about broad demographics and proper representation. You also need big data in a fragmented world to capture all viewing. I have worked on this for a good ten years, and you need both robust panels and big data. You also need to know the challenges and opportunities of each.

As media sellers and media providers, it is essential to have data and analytics experts in-house. This might not be what some partners or vendors want to hear, because it is their job to sell their specific offering. However, we must ensure that we take their valuable insights and supplement it with further intelligence. Because only then will we be measuring the entirety of a programme or the whole of a campaign.

I want to tackle this question differently. Let’s zoom out for a second and see it from our client’s point of view. We can then acknowledge our experience as a marketer and talk about them in that context. When I talk to partners or a top CMO, I always ask what we are trying to solve. The fundamental question for our clients is how to spend their money across media. People will then tell me that we must control reach because they are experiencing over-frequency.

We also have another critical customer – the viewer. We must talk our customers and understand their pain points at a deeper level. For example, are they complaining about seeing the same ad on TV and the internet or an ad being over-frequented on the web? We need to solve the right viewer concern.

So back to the question. We, the television companies, came up with our solution through a consortium called Open AP, which makes our data available to the advertiser in the same way as the GAMAM. It can be ingested into their formats, algorithms, APIs, etc. However, we are controlling what data gets represented and how. This benchmark enables us to be the masters of our destiny instead of being the victims. It is very empowering. You cannot equivalize impressions; a half-second viewable, below-the-fold ad on a digital provider of choice is not the same as a Super Bowl spot or an NCIS spot. Whether watching it on OTT or linear TV, one cannot equalize sight, sound, motion, good production values, and context. I want to be clear; we are against the efforts towards equalising impressions – 30 seconds is not the same as one second, and quality and context matter.

AI is only as intelligent as the data you feed it. You must constantly watch it, improve it, and ensure that better data comes into the marketplace. It is up to you to verify that AI keeps improving and is not full of biases. There are plenty of cases of discrimination; for example, when hiring new recruits, some companies only hire people like themselves because their resumes are most like the ones already in the companies’ databases. It is a matter of representation.

There are cases where AI benefits if the data fed into the AI is representative. For example, when you have a recommendation engine – “you’ve liked this; therefore, you will like this too.” However, this means that you will only act within that ecosystem. You will not talk about shows outside of that ecosystem or books that a provider does not sell. But recommendation engines work fine when you are in that ecosystem, and they get better over time.

This is because so much of it has become commodified, which is where AI works. We used to call lookalike modeling another area in which AI performs well (assuming that the data is representative) by identifying lookalike audiences, segments, and so on. But it cannot solve attribution, and it cannot yet solve measurement. That is why you need the panel and a large amount of data because the AI will be highly biased.

One last thing I would like to mention is that I have been hearing a lot of talk about AI and creativity. That does not work. AI will not help you write a song, make a better television show, or anything like that. Humans are the lead in the creative process. However, there is a role for AI in creativity if you know what you are doing. And again, you need to ensure that you have intelligent, analytical people on your teams who can look across market research, big data, AI, and panel data. Because then you’ll know where its liability lies and where you can use it.

Radha Subramanyam, Chief Research and Analytics Officer, CBS Corporation & President at CBS Vision