I had just torn my favorite puffer jacket, a Northface Thermoball, and had spent the whole day looking for an upgrade. Should I go utilitarian with Arcteryx? Extreme with Canada Goose? Flex with a Moncler?

Halfway through my frantic search, I saw a Google ad on Hulu. It was probably the most well-placed ad I’ve seen in my life. The ad was about how Google Shopping was the best place to search an aggregated list of puffer jackets, and get the best deal possible once I had made my choice.

But this got me thinking: someone had invested time + effort to create this ad, right? What would have happened if I had been looking for… wool coats? or socks? or anything else that wasn’t exactly what Google had already created an ad for?

note: the following is in no way definitive, and comes from a uninformed idea of how ad serving works. This post will likely get improved over time as I learn more after talking to some friends who are more embedded in the bleeding edge of AI + Ads.

Current Architecture for Ads: Pattern Matching

This is how I imagine current Ads Systems work

However, this architecture has a couple constraints.

  1. If no one creates ads for the system, we have nothing to pattern match a user’s behavior to
  2. The maximum “favorability” score of an ad/user matching could still be low. For example, extremely niche and picky audiences will likely never have an ad that perfectly matches up to their wants.

Knowing this, how can we fill in the gaps?

Creating Ads with Guaranteed High Favorability

Instead of relying on ads creators to create ads, which is process-intensive, manual, and slow to adapt to market changes, we can create ads on the fly for users.

Hypothesis: The current/future state of AI can enable on-demand ad creation that can perfectly cater to a users interests, even if the system has no knowledge/ads of the niche beforehand.


WORK IN PROGRESS BELOW THIS POINT

Here’s a prototype that a friend of mine and I tested out.