AI Tools Dominate Ad Buying: But are AI goals and your brand goals aligned?
The week the Wall Street Journal wrote about AI in ad buying with the provocative headline, “AI Will Soon Dominate Ad Buying, Whether Marketers Like It or Not.”
So I thought it was a good moment to take a beat, and outline some ways AI works for ad buying today, and ways it doesn’t.
- AI in ad buying is primarily leveraged in optimizing digital campaigns. For digital buys, the platforms’ AI tools automate bids and other optimizations. In the past year, Meta, TikTok, Amazon, Google and Pinterest have all released these tools, e.g. Google’s Performance Max and Meta’s Advantage+. Some industry pundits predict that AI buying agents will manage 80% of media buys by 2030.
- Google and Meta use AI for attribution, inclusive of providing media mix models.
- Audience insight providers use AI to educate brands on their audience’s media engagement. (among other insights) Technically, AI is used to create “synthetic audiences,” to rapidly scale.
But here’s the thing – AI means less control and less transparency. Even AI data scientists don’t exactly know how neural networks learn.
For example, using synthetic audiences necessitates a grain of salt. What if the synthetic audience hallucinates? Read more about the challenges here.
While ad buying platforms use AI to optimize within the platform, do you trust the algorithm to optimize to your business goals or their business goals? I’ve heard many peers complain about Performance Max’s optimization to clicks, and we all know clicks do not equal sales.
For media mix, do you trust Google’s solution to be fair to Meta? Do you trust Meta’s solution to be fair to radio? Taking it a step further, do you trust any black box, without visibility into how your data is weighted and modeled, to drive smart investments? Google’s Meridian is a statistical analysis technique using Bayesian causal inference. But what if a Bayesian model doesn’t fit your data set?
AI does make tasks a lot easier. Ask anyone using ChatGPT. Sometimes “easy” creates improvements. But, without control and transparency, sometimes “easy” makes things less reliable.