Auctions: first price vs second price

Two buyers want the same impression. One is willing to pay 5 dollars, the other 3 dollars. The first one wins, but how much should she pay: 5 dollars, or just a bit more than 3? That single rule, which sounds like a detail, decided billions of dollars of ad spend, triggered one of the biggest shifts in programmatic history, and features in the US antitrust case against Google.

A second-price auction is like an eBay proxy bid: you enter the most you would pay, and the system only charges you slightly more than the next person. A first-price auction is like a sealed-envelope bid for a house: you pay exactly what you wrote, so you have to think hard about not overpaying.

Second-price: the early standard

In a second-price auction, the highest bidder wins but pays the second-highest bid plus a small increment, often one cent. In theory this is lovely: it is in each buyer's interest to bid their true value, because bidding higher never makes you pay more than necessary. For its first decade, most programmatic advertising ran this way.

In practice it broke. With header bidding, the same impression passed through several auctions in a row, each with its own rules. An SSP could run a second-price auction and then submit the result to another auction. Some sellers set hidden or soft floors, so the "second price" might be a number the buyer could not see. Buyers began to suspect that "second price" did not mean what they thought it meant.

First-price: pay what you bid

In a first-price auction, the winner pays exactly its bid. It is simpler and more transparent: a bid of 5 dollars costs 5 dollars. Around 2017 and 2018, major SSPs including AppNexus, Index Exchange, OpenX and Rubicon Project moved to first price. In 2019 Google Ad Manager followed, completing the move to a unified auction in which AdX, Open Bidding partners and header bidding line items competed in a single first-price auction.

Bid shading: the buyer's answer

If you pay what you bid, you should not bid your full value. Bid shading is the algorithm that lowers a bid to roughly the price needed to win. DSPs, including Google's DV360 and The Trade Desk, built shading models that learn from the bid landscape: what clearing prices look like for each type of impression, publisher and time of day.

1. Value the impressionThe DSP's model says this impression is worth 5.00 dollars to the advertiser.
2. Estimate the competitionHistory suggests similar impressions usually clear around 3.20 dollars.
3. ShadeThe DSP bids 3.40 dollars, trading a slightly lower win rate for a much lower cost per win.
4. LearnWin and loss feedback updates the model, so shading keeps adapting as competition changes.

OpenRTB 2.5 and later let sellers send back the winning price and a minimum bid to win after the auction, which gives shading models better data. Sellers, in turn, use dynamic floors to stop shading from pushing prices too low. The ongoing push and pull is a good example of auction dynamics: rules change behaviour, and behaviour changes the rules.

Floors and pricing rules

A floor is the minimum a seller will accept. With first price, floors got more sophisticated: many SSPs now set per-impression floors with machine learning. When Google moved to first price in 2019, it also introduced Unified Pricing Rules, which required publishers to set the same floor for all buyers in Google Ad Manager instead of letting them set lower floors for rival exchanges. Publishers complained this limited their control.

Where second price survives

Some search and social platforms still use variants of second-price or generalised auctions, and some in-app and CTV marketplaces mix models. OpenRTB's "at" field lets each request state the auction type, so a buyer always knows which rules apply to that specific bid request.