Last week, I wrote about the relevance of radio ratings in the current state of the market. I noted that Xperi, with the DTS AutoStage product, can produce large-sample data for in-car listening.
Further, I covered a paper given recently at an audience measurement conference in Copenhagen that described cooperation between radio broadcasters and their supplier in Switzerland to develop a new system that will cost around half as much as their current system.
One thing I did not touch on was artificial intelligence (AI). AI is working its way into our business just as it is in many other industries. AI services can create spec spots in seconds, giving salespeople more ammo when meeting with clients. SonicTrek recently released a set of four AI formats.
But what about research? The term for using AI in research is “synthetic data.” In other words, the research company uses AI personalities trained on real human data to offer insights in hours instead of weeks.
One company, Toluna, is presently selling AI research services in multiple countries. Would you like your proposed ads to be tested quickly? No problem, because Toluna doesn’t have to recruit anyone. Their “synthetic AI personas” can see, hear, and react to dynamic stimuli like TV ads and online video.
Before you think, “they’re just making this up,” AI models have to learn from humans. In Toluna’s case, their AI personas, more than one million of them, are based on first-party data from 79 million people. The company claims to have “rich demographic, psychographic, lifestyle, and consumption attributes” from their data.
It’s easy to see how an AI system, whether from Toluna or a competitor, would be worth trying for a company that needs to test, in Toluna’s words, “ideas, claims, messages, flavors, or other product features at unprecedented scale and speed.” As I write this, the service is available in nine countries, including the US, with six more on the way.
For the moment, synthetic data and personas are not being used for audience measurement, at least not that I’m aware of, but it wouldn’t be hard to imagine this happening at some point in the not-too-distant future. If we can be certain of one thing, it’s that technology improves over time. Think about cell phones and how far the device has come in terms of capabilities.
With all the competition in the AI space, why couldn’t a research company, whether an incumbent or a new one started specifically for this purpose, come up with a large-scale set of synthetic AI “people” that would yield audience estimates?
Let’s play a mental game with that for a moment. We know how hard it is for Nielsen to recruit PPM panelists and diarykeepers. That’s not to fault Nielsen, as no company or even the US government can easily recruit survey respondents in 2025. Why not AI?
If we think about Nielsen for a moment, the company has data from hundreds of thousands or maybe even millions of unique respondents in recent years. The company could match their listening and viewing data to other available first- and third-party information about these people to create the synthetic personas. Privacy issues? Probably not. Audience measurement is about aggregation, not individual responses.
To some extent, this isn’t new. We’ve been modeling data since the dawn of audience measurement. TV used genres of shows to model missing data from TV diaries. Radio has ascribed days from filled-in days in a diary to missing days for over 20 years. There are all kinds of edits and models used to deal with missing data in audience measurement. And if you think about it, weighting to populations is a form of modeling. Individual meter and diary data are weighted to make the survey look like the population.
Further, AI could take some of the “crazy” out of the system. A little AM in the top three because two heavyweight meters listened all day? A country station beating the local urban formats among Black listeners? An HD2 with a seven share? My guess is that won’t happen with synthetic data.
AI companies are moving quickly, with trillions of dollars being invested in chips, data centers, and people far smarter than me who can design the systems. These systems will get better, and at some point, someone, whether Nielsen or a new competitor, will figure out that an AI system can deliver near-real-time data that makes sense.
The investment in the design will likely be expensive, but the goal of nearly eliminating the recruitment, collection, and processing of human-generated listening and viewing data is very tempting. I don’t want to suggest when, but for my money, it’s inevitable.
Will advertisers and agencies push back? I doubt it, because the numbers they have today are just approximations anyway. AI ratings will probably smooth out fluctuations, which means fewer complaints about audience delivery.
Maybe this is just a crazy dream, but who would have imagined all the technology we have available to us today? Don’t bet against it.
Let’s meet again next week.
Barrett Media produces daily content on the music, news, and sports media industries. To stay updated, sign up for our newsletters and get the latest information delivered straight to your inbox.
