Looking to the past with an eye on the future, Simon Bird offers up a different perspective on the demise of third-party cookies.
Our industry has a fetishism with new. We love new ideas and by virtue of not being new, established ideas and tactics are too often forgotten about or thought of as inferior. The demise of third-party cookies presents us with an opportunity to remember this and make the next decade of digital marketing better than the first one.
There is a wide variety of opinions about what the future of a no third-party cookie marketing world might be like. Some opinions are positive, others more negative, and, obviously, making predictions about a future state that is still unclear is fraught with uncertainty. What is quite consistent across the industry however is the rise in talk about context being important and the value of some form of attention metrics. They are essentially talking about the same thing – quality matters. Metrics such as half an of an ad in view for one second counting as ‘viewable’ sets a rather low bar for quality for digital advertising but it’s made worse by forgetting, or ignoring, the role of context and attention. We have an opportunity to address this.
One of the foundations of programmatic buying has been the thought that someone’s eyeballs (or ears) are exactly the same, regardless of the environment. For people old enough to have worked in the industry well before ad exchanges this thinking was always flawed. For the same reason wine tastes different when poured out of a different bottle, advertising works differently depending on where it is delivered. The more we unpack the first ten to fifteen years of digital marketing, the more other flaws are exposed. Many of the worse ones are around analytics and optimisation.
Pathway analysis, attribution modelling and real time optimisation sound far more scientific than they really are. The assumptions they are built on are far too deterministic, marketing is a game of probabilities not certainties and any effective analytics or modelling needs to reflect this. It doesn’t take a degree in statistics to see the flaws. Even in a small market like New Zealand, for a typical campaign the number of potential pathways of digital channel and message are so numerous that even the ‘best performing’ pathway will probably only deliver circa 10% of conversions. Furthermore, even if we did find a magic pathway you cannot control someone’s digital channel usage or pathway anyway.
Whilst real time bidding is enormously useful, real time optimisation makes little sense. It is in essence analysing random data noise – the drivers of human decision making are hundreds of thousands of years old and don’t change in real time, the major media placements don’t get substantially better or worse in real time either, people’s online behaviour does change in real time but because this activity is so inconsistent from one minute to another it usually doesn’t provide a stable enough signal to make real time optimisation worthwhile. You can optimise offers and messages and products in real time but optimising across longer timeframes will almost always deliver more reliable signals and more effective optimisation.
Attribution modelling models only digital channels. It doesn’t include important factors such as price, effect of non-digital channels, competitor activity or broader drivers such as weather, traffic or a global pandemic. In addition, most attribution modelling, because of the methodology, models in ad fraud, whereas other modelling methods, such as, econometric modelling, model fraud out. Despite significant advancements in ad fraud detection tools, the estimated amount of money marketers are losing to ad fraud globally is now estimated at over $30 billion dollars a year. Inadvertently, attribution modelling is contributing to this.
Lastly coming back to third-party cookies it has now been shown that the accuracy of this data is subpar. Even simply identifying male (or female) has been shown to be less accurate than chance – as shown by Neuman, Tucker & Whitfield in their 2019 academic paper. Their research shows the average accuracy of identifying someone as male from a selection of data brokers to be only around 42 percent.
This isn’t really a surprise as it’s inferred data from online behaviour but it is problematic if ad content is customised ‘knowing’ the sex of the target audience. How, and if, a message might be customised if we knew the target might be female would likely be different that if we knew for certain they were female. Furthermore if, alongside sex, we include another marker such as age or sporting interest the accuracy of the data reduces further still. Clearly once data gets too inaccurate customisation eventually becomes ineffective and expensive.
I am not suggesting that digital marketing itself is inferior or inherently flawed. The last 15 years of technological evolution has made our industry far more interesting to work in but the increased complexity has also given us new ways to make mistakes. We have persuaded ourselves that using technology and being data led was being more ‘scientific’ but in our enthusiasm for new ideas, tactics and technologies we completely ignored other parts of being scientific such as scepticism toward new concepts. Scientists know they are prone to biased thinking about their own work so they use double blind studies and work is peer reviewed before being published. In many ways embracing new ideas quickly is one of our industry’s greatest strengths, but from time to time it pays to be a little more sceptical.
Looking back over history it has always taken time to learn how best to use, think about or measure new channels or technologies. Given we have made mistakes with pretty much every channel it would be somewhat surprising if we have not done this with the new technologies of the last 10-15 years.
The first ads in press, radio and television were all rational and information heavy as we were yet to find out that most decision making is intuitive and subconscious. The first books on how advertising works compared it to how a salesman works and were very much about persuasion via a rational delivery of information and all analytics was only about direct sales. Our first attempts at the what and why of the dot-com economy were enthusiastic but also proved to be a little suboptimal.
Overtime we learned new things. We now have a much better understanding of how the brain works such that we can now make effective advertising with no information in it at all and we understand how valuable, and predictive of future sales, simple metrics like top-of-mind salience can be. We also now know that people are happy to pay a premium for brands simply because they like them and that good branding is so strong it can even influence product quality perceptions.
Last week I read an article all about the making of the iconic Apple ad ‘here’s to the crazy ones’. Despite being over twenty years old it remains a great example of powerful TV advertising and one that would never have been made in the early days of TV. It talks about the crazy ones, the people who challenge the status quo and change things… …in many ways it’s a summary of the types of people that work in our industry.
The passing of third-party cookies is an opportunity for us as an industry to challenge the status quo and change things. Let’s admit that we made a few errors in how we approached digital marketing. Let’s not make these mistakes again using third-party cookies in disguise. Let’s challenge ourselves to find ways to use the amazing canvas that today’s technology has given us to make a more enjoyable eco system for users, one that doesn’t use data to chase people around the internet serving ads for products they have already bought. Let’s use analytics that help grow business not analytics that serve ads to people who were going to buy anyway. Let’s break down the false constructs of digital and non-digital. Let’s think about what we know about how people actually make decisions and embrace the complexity of human behaviour and imprecision that comes with it.
We need to stop thinking deterministically and start thinking probabilistically. We need to remember that most people, most of the time are not buying most products, they are not on a customer journey. Amazon cart abandonment rates are really high, and these are people displaying buying intent via search and are one click away from purchase. No amount of data will ever get close to us being able to serve the right message, to the right customer at the right time. We can use data to be slightly righter, some of the time. The rest of the time we should remember marketing is mostly about future persuasion not direct conversion and use messaging and analytics accordingly.
And lastly, you’d never hear a physicist say gravity is so last century or a mathematician say calculus is four hundred years old and a bit irrelevant now. Science builds from the knowledge of what has gone before. It does change when new information is found but it doesn’t throw old theories simply because they are old.
So, if we really do want to be more scientific as an industry, and get better at using new technologies, let’s try and stop inferring that new is good and old is bad.