Data is the backbone of AI models. But without high quality data, marketers are at risk of making incorrect assumptions and wrong decisions. That’s what Jane Ostler, Global Solutions Marketing and Thought Leadership at data and insights giant Kantar believes.
In this conversation, FutureWeek delves into the reasons why quality data is everything when it comes to using AI for marketers, why using AI models trained on the same data could create a ‘sea of sameness’, what some of the main barriers to AI-use for media companies are, and more.
What attitudes do you see marketers having towards AI?
The desire to use AI is there. From our research, 68 percent of marketers feel very positively about Gen AI and all of its potential uses. However, that’s balanced out with the fact that most marketers still feel unprepared for any kind of implementation and integration of AI into how they work.
We’re seeing experimentation. Marketers are trying to figure out ways of embedding AI into certain processes. Some of those will be workflow processes behind the scenes. Others will be more overt uses of AI, such as using it for effectiveness measurement, or for generating insights if you’re running a brand portfolio, for example. One thing you need to do in an organisation to build usage and trust in AI is to make sure that everybody has access to it and is allowed the time to play with it. The other thing that is required is a foundation of very high quality, reliable, timely and relevant data.
Why is relevant and timely data important for AI?
As a marketer, if you’re building your own models for AI, for whatever use case based on your own data, it needs to be reliable, because you’re going to be making decisions on that basis. And if we look at a future where a lot of businesses and marketing teams will start to look at the use of AI agents, having the foundation of high quality data is super important for that. If you’re allowing autonomous agents to go off and make decisions on your behalf, or even take actions on your behalf, you need to be confident that the data underpinning the training is right.
The thing our brands have at their disposal is amazing data, and it’s only the amazing data that, if it’s leveraged properly, can be used to continue to differentiate brands and to make them stand out for people. Because if everybody has access to the same tools and is operating with the same brief in the same category, the risk is you end up in a ‘sea of sameness’. So it’s actually the data that is your differentiator.
What are major challenges marketers have around data use?
We have a lot of data. We know the effort that is required to not only keep it up to date, but to make sure it’s cleansed regularly, to make sure it’s weighted. If you’re representing people’s views, it has to be a representative view of the population, and if we’re doing some pilots with synthetic data, that has to be based on solid data foundations.
From speaking to a few brands, where they are looking at how they organise their own data in-house to create their own tools, systems, etc, a smallish error can make a significant difference in the conclusion that you generate from that data, especially as we move to synthetic data and digital twins becoming more commonplace. You may amplify small errors that actually lead you to the wrong place. Most marketers obviously have got a handle on their data, but you need a governance and a compliance workflow to ensure that the data is always at the highest quality.
What is the ‘sea of sameness’ when it comes to AI-generated content?
It’s a hypothetical risk. If everyone is using the same data sets to generate the same ideas, are you all going to come out with the same answer? Image generators and video generators in the hands of the right people can be used in a highly differentiated and creative way. How you retain that creative spark, that sense of difference for your brand, has to be somehow worked into the training of an AI model.
There is the potential that you might just go online and be assailed with hundreds of thousands of ads, all of which lack differentiation. That’s why as well as data, you still need human input and human expertise to know where to steer the direction of travel for your Gen AI-created content.
These are all the big philosophical questions that we need to think about. It all comes down to the data and the humans who are handling it. I think it’s a risk. I think most marketers are aware of that as a risk, because you produce more content, and you still need it to stand out from the crowd somehow, either in the messaging, how it manifests itself, how it’s recognised, and how it persuades people. A better example of ‘the sea of sameness’ is when algorithms all optimise toward the same consumer preferences — so brands could lose out on differentiation and meaning.
How can brands differentiate themselves?
I think brands need to understand what the reaction is going to be at all stages of evolution in a creative process – don’t leave it to chance. So if you’re looking at an idea, what do people think about that? What do they take out of it? How can you tweak it to make it different from competitors? Or so audiences are more excited by it, or find it funnier?
In a very fragmented media world, and we’ve been speaking about consistency for years, it is important to make sure that your brand represents itself appropriately for the relevant platform. So ads need to be customised for every platform.
People have often talked about CMOs being like the conductor of an orchestra, and, in a way, they’re going to be even more like this in the age of AI agents. But what the difference between now and 20 years ago is that we have intelligence at every stage of the creative process.
Are marketers open to using AI?
I think it depends which part of the industry you’re talking to. When new ads come out that are clearly AI generated, or explicitly stated as being AI generated, there’s normally a bit of LinkedIn backlash from certain people. We can’t ignore that. How to harness AI and use it creatively, and master its use is actually what marketers and agencies need to get on top of, and the data is the foundation for that.
One of the interesting use cases, and we do this a lot at Kantar as well, is using AI for finding new innovation. Whereas five years ago, as a marketer you would whittle down some good ideas, get to a small handful of them, and then test them and see what would work as a new product. AI enables you to do that at scale. It enables you to take a whole host of ideas – and those can be digitally derived from things like search, social data, and trend analysis – and take them through a process of working out which ideas may be viable, and even what price you might charge for them.
And again, that all depends on the foundational data sitting behind it. Whether it be considering previous innovations in your company – how financially successful they are, what the reaction to them was, whether they were new products or trend-led etc. All of that data can be housed and used to train an AI model.
What are some of the barriers to adoption media companies are facing?
There are three stages: knowledge, having AI, and then using AI. In our ‘Fear or FOMO’ report, we actually identified that there is a gap in adoption. Even for the marketers who do have Gen AI tools, less than half are actually using it on a monthly basis. And this is where organisations and teams need to be encouraged to use AI, because they can generate use cases in their own right.
I think the other issue is you have to make AI ubiquitous. You have to give it to everybody. It’s a democratising technology. You might have an AI development or product development team who are working on these things day in, day out, but you have to democratise access to it. I think that’s something that organisations struggle with, because normally you cascade things down from senior people to more junior people, which takes a while. But actually, the way you can get the best out of AI and come up with the right use cases for your team, your department, and your organisation, is by giving everybody access and letting them have a go.
Do you think consumers have the same openness and receptiveness toward AI-generated content?
I think it’s similar for consumers, they tend to be positive about AI. It’s a lack of knowledge and lack of understanding that is holding people back. We looked at some ads that had been created using AI – either the script was written using Gen AI, or its video formats were wholly created using Gen AI, and then we used AI to test the ads to see whether they were effective or not.
We found that it doesn’t actually make that much difference if your ad is created using Gen AI or not, as to whether it’s effective or not. They’re either good or they don’t work so well. We have lots of benchmark data that tests ads against all sorts of different benchmarks. So we can tell brands what good looks like, and if an ad is underperforming or overperforming in certain areas.
What AI tools are you using at Kantar?
Probably the most relevant one is our ad-testing platform LINK AI, which is how we do creative testing – it’s powered by hundreds of thousands of ads. This is what the model has been trained on, and that works continuously. We refresh it on a regular basis.
It’s not ads just existing in one big database. The model learns from different ad platforms. What worked really well on an out-of-home ad or a TikTok ad is not the same as what works on television, for example. And the other thing that AI can start to help us to do is extract features from ads to see what works both in the short term for sales and in the medium term for brand impact.
So we can start to decompose the elements that are working in different ads. For example, it has celebrity recognition – it knows if you’re using Beyonce in your ad, so it can account for that in the measurement of the effectiveness and its prediction of what people will think. We also use AI for predicted eye tracking to see where people are looking, and for emotion detecting as well – emotion in advertising is massively under-leveraged still.
