The Hidden Auctions That Set Online Ad Prices Just Got a Math Upgrade
Smarter ad auctions could mean cheaper ads and less wasted marketing money.
When you load a webpage, an auction runs in milliseconds to decide which ad you see. Advertisers no longer bid by hand — they set a goal like "get $4 of sales for every $1 spent," and software called autobidding does the rest, adjusting bids in real time. This system runs a huge chunk of the free internet, so the rules governing it matter more than most people realize.
The new paper looks at something researchers had mostly ignored: what happens when software is bidding on both sides of the market at once. The authors found a stark negative result. In the toughest scenario, where nobody knows the true value of anything and everyone is chasing maximum value, nearly every fair and balanced auction design collapses — the outcomes can get wildly inefficient. In plain terms: when machines negotiate with machines, the usual safeguards stop working.
The good news comes in two parts. First, an old, simple design called McAfee's Trade Reduction mechanism performed far better than expected under goal-based bidding, keeping waste to a constant, manageable level. Second, when the platform has some statistical information about what things are worth, the researchers designed a new auction that achieves the best possible outcome whenever at least one side is using goal-based bidding. That is a genuine surprise, because a famous 1983 proof (the Myerson–Satterthwaite theorem) says perfect efficiency is impossible when both a buyer and a seller act purely in their own interest.
So what should you do with this? Nothing yet. This is a theory paper headed to an academic conference, not a product launch — no platform has announced it will use these rules. But the direction of travel is clear: as AI systems take over buying and selling on ad platforms, ride-hailing apps, and ticket markets, the math underneath those marketplaces has to be rebuilt. Getting it right could mean lower ad costs for small businesses, less money burned on irrelevant ads, and healthier payouts for the websites you read for free.
- Autobidding (software that bids for advertisers automatically) now runs most online ad auctions, on both the buying and selling side
- In the hardest case, nearly all fair auction designs break down — but an old, simple design called McAfee's Trade Reduction still performs reliably
- The researchers built a new auction that reaches the best possible outcome when at least one side uses goal-based bidding — something a famous 1983 theorem said was impossible
Why It Matters
Better ad auction rules could mean cheaper ads for small businesses and more revenue for free websites.