Rob Webster is Founder of Canton Marketing Solutions. He’s worked in the adtech industry since 2001 and is NDA’s monthly adtech columnist.
Just as flares go into and out of fashion so AI has had several runs as being the be all and end all feature to justify marketing technology budgets. It first reared its head in marketing paid search in the middle of the noughties before moving primarily to programmatic display early in the programmatic era.
Now it’s back in many forms and promises to be one of the features of the age. Used well AI can be transformative and there is no doubt the best marketing programs utilise AI extensively. However, two decades of experience have shown me that all too often it can be snake oil – used only to line the pockets of the tech companies founders and of little real world value to the advertiser. How can the modern marketer tell one from the other?
First a (true) story. About a decade ago the programmatic (display) revolution was really kicking into gear. I was running an agency technology team at the time and one of the gatekeepers to programmatic technology for the agency’s clients. Buying platforms with undisclosed margins were lining up to tell me that to run programmatic display well – you needed an algorithm built by a brain surgeon, astronaut, nuclear engineer or some other brainbox with no marketing experience. Such companies had done very well by combining media lunches with a rhetoric that tried to bamboozle the marketer that only with their brain-box-driven AI could you hope to succeed in programmatic.
When I asked why this was the case I was usually at first given some canned rubbish on the power of AI and who invented the algorithm. My pushback, that these companies were poor at taking on marketing strategy, were poor at explaining why things worked, had no concept of incrementality, were taking a lot of margin off the table and that these things were important, often led to me being politely accused of being a luddite in the face of the AI future. On several occasions, but one memorable one in particular, they would wheel in one of their “brain surgeons/rocket scientists” to teach me the error of my ways. On every occasion, but again one in particular sparks would fly in these sessions, I have heard the particular debate described many times as a train crash.
Before I go further into that particular story I should tell you that I am a huge fan of AI when it is aligned to purpose. It is the future of not only our industry but our world. As a big, popular science level reader of AI (ie not that smart) I know that there are too major kinds of AI, there is the specific AI and there is general AI. General AI does not really exist yet, it is general in that it can be applied to anything it chooses to teach itself. As and when it exists the same AI could wipe the floor with you at chess then work out the optimal way to run the economy, design an amazing vaccine and maybe even predict the future.
The sci fi drama series Devs using a quantum computer is a lovely example of a general AI and right now it’s the preserve of science fiction. Only humans on this planet have a high-level general AI and even then it’s debatable. Instead what humanity has developed in the last 10 years particularly is some truly amazing specific AIs that have changed their field. From language translation, picture and voice recognition, chess (and any board game going), driverless cars, aeroplane autopilot, financial trading – the list goes on and on – and yes it very much includes marketing. What these specific use cases have in common though is a very specific purpose and definition of success. As we shall also see, they also have access to all the relevant information, which is vital.
Back to my debate with the Brain Surgeon/Rocket scientist. It was a long winded back and forth that went into a lot of detail but can be fairly summarised as the following. My first major claim was that my trading team, in certain circumstances, even without getting into margin, would be able to beat his AI because we had information he didn’t. His counterclaim was that in the world of financial trading many thought this to be the case, but that today in financial trading you always bet on the trades done by the machine. My killer point to him though was that this is not financial trading, it’s marketing, success is less well defined and buyers and sellers can know things that the AI does not without it being a case of insider trading.
The whole point of marketing is to have an advantage and to communicate it to the market, my team for example know for this client when the sale periods are, know when the TV campaigns are going live, have experience of what external factors influence demand (sports games for this particular client) and have an intuitive knowledge of incrementality from having run experiments. They also have their own, modest algorithms to help them do so. All of this knowledge COMBINED with a great AI would be amazing but that if its a choice between one or other we would take the marketing skill all the time. I would absolutely let my team use an amazing AI self-serve (with a disclosed margin) but would not without more proof be able to hand over the budget to his business. My Brain Surgeon/Rocket scientist interlocutor had nothing to say to this and toddled off muttering with smoke coming out of his ears. I think I heard him muttering that it’s just not logical / does not compute or some such.
Please don’t fall into the trap of taking the wrong things away from this. I am not saying AI is bad and I am not saying that I am smarter than a Brain Surgeon/ Rocket Scientist. On the latter I am very clearly not in their league. However in these examples I do know more about marketing. Indeed crucially I know a lot about marketing and a useful amount about AI and algorithms which as a combination of skills absolutely destroys someone who knows a lot about AI but nothing about marketing. Crucially these companies fell over by not trying to combine AI with insight and marketing knowledge, by not empowering marketers but looking to try to replace them (and make a big profit). As for the value of AI, AI is aligned with purpose and the best information is marketing nirvana – yet only with this alignment. To be clear I am the biggest advocate possible of AI aligned with data and purpose however I am also the biggest detractor of poor AI sold as a reason unto itself.
No one should buy anything because of AI, only because of the benefits AI brings.
To examine this further let’s remember the great examples of AI success. Language recognition, chess, self driving cars and the like. In these cases AI is a truly magical technology when applied to a purpose where success can be easily defined. In the stock market it is turning a profit. In chess it is winning a game. You can take any successful AI area and apply this. Crucially too in those areas there are not many externalities, things the AI does not know about. This is why an AI can beat you easily at chess but may not necessarily be great at betting on the horses (at least not yet) as it needs more data.
In marketing, AI buying algorithms have some real problems marketers need to remember. First among which is the poor standard of measurement and incrementality throughout the industry. AI is usually tuned to poor measurement systems version of attribution. Since these measurement systems over value the bottom of the funnel last click activity so does the AI. Since these measurement systems can be confused by cookie bombing (low frequency ads shown on low cost sites, often unviewable) so the AI amplifies this problem. It is for these reasons and more than when you looked at the buying of the “Brain surgery” type managed DSPs of the previous era their actual delivery was very often 90% bottom of the funnel retargeting. Let’s be clear, you don’t need AI to make retargeting work, the fact that the user has already shown an interest in the product by visiting the site is enough. Yes it might make it a little more efficient but the incrementality on that is very very low.
For all these reasons they should remember Webster’s first golden rule of AI.
Golden Rule of AI number one: Whenever a marketer is sold an AI solution without being told why it works, its purpose, its margin and a way of validating it just walk away.
Black box AI is more likely than not to be snake oil. Only when you can validate the output adequately should you entrust time and media spend. For whilst the days of programmatic networks are gone AI without purpose is not. A new breed of solution is emerging, some amazing and some not so. It is vital that this rule be followed for marketers to spend their money only on what they can see with their own eyes is true value.
For there are two more Golden Rules of AI.
Webster’s Golden Rule of AI number two. Incrementality is king. Where AI can be adequately directed at a legitimate marketing purpose (incrementality), with sufficient sophistication and with sufficient data it will be incredibly effective and efficient.
Webster’s Golden Rule of AI number three. When evaluating which AI to test, look for those that have the most access to data and can best be directed to marketing purpose (incrementality).
For we are truly in the era of privacy and AI now where the best marketing will rely on AI. Great uses of AI in our industry for buying are surprisingly rare but they do exist (for my money the best uses have long been in paid search (outside of brand) and shopping search). Lookalike targeting using AI can be phenomenally powerful providing we can validate its not just retargeting by another name.
Marketers must be able to evaluate how to distinguish between the good and the bad, to understand the true value of what they are using. Being able to test different approaches and understanding how aligned to real world marketing benefits and incrementality is vital to success. Being able to do this in an automated and omnichannel way is the journey the best marketers and companies are on over the next era. When brain surgeons/rocket scientists sit by side (as equals) with marketers and collaborate to get the best results exciting things can happen.
Even better when they build modular platforms that marketers can use applying their experience to the power of data.
