In the words of Oscar Wilde, nowadays, people know the price of everything and the value of nothing. Of course, one must bear in mind the famous Abraham Lincoln quote, “Be careful of quotes you find on the internet.” Still, Wilde’s comment applies perfectly to the ongoing transparency debate around programmatic: people know the price of everything and the value of nothing.
Now, before I go on, let me say this. I’m not advocating for agencies to rip off their clients. I’m not saying any form of opaqueness is a good thing. It’s very clearly a bad thing. But there is a fundamental piece of this debate that’s missing and that is the value of inventory.
If inventory is worthless – as in, it’s not viewable, it’s being seen by the wrong eyeballs, it’s being exposed to the devices of people that have shown no purchase intent, etcetera – it doesn’t matter what you’re being charged for it, because you shouldn’t be charged for it at all. It’s actually worth nothing.
In recent months, we’ve heard that the programmatic supply chain is “murky at best, fraudulent at worst”. The truth is, there will always be some level of murkiness because there is in every medium. We’re not 100% sure that the readership numbers associated with newspapers are accurate, just as we can’t say for certain television viewership is completely spot on. We don’t live in a world of absolutes.
We’ve had people say that the programmatic well has been poisoned. That programmatic doesn’t work, that it doesn’t return ROI. That’s a bit like saying, “Advertising doesn’t work. I never buy products that I see advertised.” It’s a sweeping generalisation and couldn’t be more bollocks unless it was wrapped in a sack and slung between the hind legs of a large dog.
What we do know is this: sophisticated marketers who employ mathematically robust models and approaches to attribution are seeing tangible results from programmatic. They don’t have this pressing need to focus on transparency because the models are very clear about what is and isn’t effective when it comes to return on investment.
The best of these models rely on machine learning. The average brain can handle about seven variables, however, using machine learning in conjunction with programmatic, you can run campaigns that apply up to 200 variables that positively or negatively impact an outcome. If you line up those 200 attributes all at once, the computer will bid much more for inventory that meets them because it knows the value. And that’s what we get paid to provide our clients: value.
Right now, everyone’s getting their knickers in a twist about the viewability but again, if you’re applying the right models, viewability isn’t an issue. If a website has consistently poor viewability, and you aren’t generating conversions from placements there, a system that relies on machine learning will quickly divert the ads elsewhere.
Similarly, if you’re optimising to a conversion or a sale, if there’s fraudulent inventory being served up, the computers will quickly establish that the inventory isn’t generating a sale and won’t serve to that website, that format, that time of day, or the myriad of parameters you have set. And the more factors it learns, the more likely it is to buy valuable inventory that’s transparent, at a better average price.
But this takes time. A learning computer needs a minimum of a week, sometimes two, to really learn. And during that time, cost per acquisition bids can be high because it has got to run so many tests. In today’s world of short-termism, often people aren’t willing to wait. If your acquisition rate hasn’t improved in a week, it’s not a disaster. Often times, you simply need to give the computer longer to learn.
The fact of the matter is that we have trained media planners and buyers for years to do this job. They make mistakes and they learn from them but they’ll never be able to do it at the speed of computers which means machine learning optimised programmatic is on track to eclipse their abilities. Any media person that thinks they’re better and smarter than a computer is in for a really rude shock. It’s a bit like driverless cars. Despite the fact that they are statistically proven to be safer than cars driven by humans, they have one accident and people want them off the roads.
We have to accept that programmatic overlaid with some form of machine learning will absolutely be the future of media buying. If people disagree with that, they need to get their heads read. Right now, we don’t have access to the power of machine learning beyond the digital environment but that is changing. It won’t be long before we’re applying these methods to more traditional mediums such as television, outdoor and radio. And when we do, we better line up the right target to shoot at.