If you’ve followed me for a while, you know that I’m not a big fan of ad-supported business models. It’s not just that I find ads annoying; I also see them as economically inefficient. I’m excited to see technology favoring ad blockers in their arms race with advertisers.
But I was recently thinking about an ad-supported model that I could get behind — a model that isn’t adversarial towards users. Here’s a rough sketch.
My proposed approach builds on explore-exploit, a paradigm used by most modern recommender systems — including for advertising. Explore-exploit systems manage the trade-off between exploring the space of payoff opportunities and exploiting the payoff opportunities they discover.
Underinvesting in exploration causes systems miss out on new opportunities. Overinvesting in exploration prevents them from sufficiently exploiting the opportunities they’ve discovered. Finding the optimal trade-off between exploration and exploitation is an active area of computer science research.
Putting Users First
Explore-exploit systems generally optimize for business metrics, such as revenue or conversion, rather than for user happiness. Indeed, ad-serving systems have an adversarial relationship with users — as evidenced by the proliferation of ad blockers.
But it doesn’t have to be that way. The explore-exploit paradigm suggests a way to engage users cooperatively rather than adversarially. Exploration is a form of investment to obtain signals that can be subsequently exploited. What not implement explore-exploit in a way that the same users who enjoy the benefits of exploitation also pay for the exploration?
Let’s look at how this approach could work for a content recommendation system. When new content arrives, we don’t know whether users will find it relevant. So we place that content in the exploration funnel: a subset of users engages with the content in order to obtain the signal from which everyone will benefit.
We can build an ad-supported model around this interaction. Instead of segregating ads from organic content, we treat ads as new content to throw into the exploration funnel. Participating in exploration is the price users pay in order to benefit from the exploitation of signals derived from that exploration.
But what about advertisers? Since advertisers benefit from users seeing their ads, they pay for the exploration work. If their ads are worthless, they’re wasting their money. But if their ads are relevant to users, then a small investment quickly allows them to earn their quality signals.
Google’s Susan Wojcicki once said that “advertising should deliver the right information to the right person at the right time”. I was skeptical at the time. But if we subject ads to the same evaluation as any other new content, we can build a non-adversarial model. And it makes economic sense to have advertisers pay for users to perform the exploration work.
Tear Down This Wall!
Replacing the wall between ads and organic content with a unified explore-exploit model won’t be easy. Users have come to expect this wall, and various laws and regulations require distinctive labeling of ads.
Still, I think it’s an idea worth exploring. If ads are intended to be relevant content, they should be evaluated as such. Let’s move beyond the adversarial model to one where advertisers and users cooperate in a goal of delivering the right information to the right person at the right time.
Let’s tear down this wall.