“I say your civilization, because as soon as we started thinking for you, it really became our civilization.” Agent Smith, The Matrix (1999)
Over the last two decades, I hired, trained, and worked with hundreds of brilliant media planners and campaign managers. During this time, the industry developed a set of principles for media planning and campaign optimization. My teams and I refined and successfully applied these strategies in a myriad of campaigns.
Now, having launched several AI products for media buying and optimization, I can confidently say that the old principles are not only ineffective but also harmful. If they are not discarded, it is only a matter of time before media executives realize that planners and campaign managers do more harm than good.
What is that “AI” I am talking about?
Generative vs. Predictive AI
Generative (Gen) AI garners all the hype, often becoming synonymous with AI in the public’s mind. This misconception stems from our daily interactions with tools like ChatGPT, Gemini, or Grok. These tools are valuable time savers. I use them to help me write standard Python code or summarize lengthy scientific papers before deciding to read them. Yet, as impressive as Gen AI’s advancements are, it merely imitates human-created content. By definition, it cannot produce original, groundbreaking work that surpasses the best human expertise. I chose not to use Gen AI for this article, as I believe I have something new and original to say here.
Predictive AI operates quietly in the background, often unnoticed despite our daily interactions with it. It powers personalized recommendations on platforms like Netflix, Spotify, and Amazon. It drives drug discovery, drives autonomous vehicles, and flies drones. Unlike Generative AI, Predictive AI doesn’t merely imitate humans. It frequently surpasses the best humans — and by a large margin. In 2017, AlphaGo defeated the world’s top Go player, mastering a game once thought to be AI’s ultimate challenge.
The accelerated computing that powers AI, along with AI applications, is said to follow Huang’s Law, where cost-adjusted performance doubles roughly every six months, far outpacing the two-year doubling of the old Moore’s Law. Modern Predictive AI systems are thus tens of thousands of times more complex than AlphaGo.
DSPs and ad servers leverage Predictive AI for targeting, bidding, and yield optimization. Having developed several such solutions, I was able to prove that they produce far better results than anything my teams and I could ever achieve as human media planners and campaign managers. Humans are simply outmatched. Relying on pre-AI principles when powerful Predictive AI operates in the background is like doing a complex surgery with a stone axe.
Here are the principles. You might recognize many of them.
The Old Strategies and Why They No Longer Apply
1. Know your audience; launch a precision-targeted campaign.
There are three problems with this outdated strategy:
- Most of the time, you do not know your target audience nearly as well as AI does. You just have a simplistic understanding of several key dimensions of your “ideal customers”: age, gender, income, interest, etc. The reality of what motivates consumers is much more complex. Predictive AI solutions calculate the probability that a user will convert based on hundreds of user characteristics. They can capture the complexity of user behavior.
- Even if you knew your best audiences, how would you reach them at scale? Your first-party data is invaluable—you know its source and can trust its quality. You should target (or re-target) the hell out of it. However, to acquire new users at scale, you often rely on third-party or second-party audiences. Can you trust their quality?
“But we can test the third-party audiences and see if they work.”
Easier said than done. When you target multiple segments in a single campaign—especially with a Boolean “AND,” like “Age = 30–50 AND Gender = Female AND Income = $75k+”—a single low-quality segment can tank the campaign, and pinpointing the culprit is nearly impossible. Other conditions may impact performance, making it difficult to establish a cause-and-effect relationship. A robust Predictive AI solution swiftly identifies which signals (first-party, second-party, or third-party) predict conversions and adjusts bids accordingly when the signals are present.
- (almost) Every user is valuable–at the right price. Of course, there are geographic and minimum age legal limitations. Besides those, every user should be “fair game.” You should not just bid on the best users. If User A is three times less likely to convert than User B, you should bid three times less on the former. With such a setup, you maximize scale and return on ad spend (ROAS). Predictive AI does this with ease.
2. Select the best-performing inventory. Aim for high viewability rates and utilize high-impact formats, such as video, interstitials, CTV, and native.
This practice would be useful if:
- The CPM bid price was fixed and
- You were the only one who knew that certain formats and premium placements perform well.
In an auction environment, other advertisers drive up prices for inventory that performs better on average, resulting in significantly higher CPM rates for high-quality inventory. Excluding inventory you deem “suboptimal” forces you to place higher bids on the remaining inventory to meet your campaign budget, leading to reduced efficiency. To optimize performance, bid across a wide range of inventory, proportional to the value of each impression for your campaign. A Predictive AI solution delivers this precision at scale.
3. Set frequency cap to prevent over-exposure.
Frequency capping (FC), like audience targeting and inventory selection, is another crude “human-era” tool now rendered obsolete. All else being equal, a user who has seen an ad five times is typically worth less than one who has seen it four times. But “all else” is never equal. A high-value user on contextually relevant inventory might justify a sixth, seventh, or even twentieth impression, while a less valuable user might not warrant a second. The solution? You guessed it–Predictive AI. It incorporates user exposure frequency as a model feature and lowers the bids correctly as the user sees more of the same ads.
4. Conduct an A/B test, then optimize.
Optimization via A/B testing involves launching randomized experiments and waiting for statistical significance to emerge before taking action. Experienced planners and campaign managers, especially those I’ve trained, are adept at these calculations. The math is straightforward for those familiar with it, but it’s time-consuming. Once again, Predictive AI outperforms this approach. Reinforcement Learning solutions dynamically allocate budgets to higher-performing options well before statistical significance is achieved, accelerating performance improvements. Elegant algorithms exist that optimally balance exploration (testing new strategies) and exploitation (scaling what works).
In conclusion, Predictive AI, which drives media buying and yield optimization, vastly outperforms human capabilities. It captures the complexity of user behavior and does billions of calculations for every ad impression—a feat that no human can match. Applying outdated “tried-and-true” principles alongside Predictive AI is counterproductive. Media planners, campaign managers, and their leaders would be wise to embrace this shift and adapt.
But is it all hopeless for the human Campaign Managers and Media Planners? Can we still add value?
As a matter of fact, yes! Stay tuned for my next post titled “Digital Advertising the Era of AI: Four Strategies That Still Succeed–and Have Become Even More Important”
Over the past 16 years, I hired and managed a couple of dozen digital campaign managers, both for guaranteed and programmatic campaigns. In addition, I interviewed many more campaign managers working for premium publishers, DSPs and Agency Trading Desks. Lastly, I interacted with dozens of Digital Advertising and Marketing leaders. Based on the numerous interactions, I came to the following conclusion. Manual (human) campaign optimization decisions are often made without correctly calculating statistical significance. As a result, many manual campaign optimization decisions have a strong gambling element.