Are auctions here to stay?

For most of its history, programmatic advertising has rested on one idea: every impression is sold in its own auction, in real time, to the highest bidder. It is an elegant model. It is also under more pressure today than at any point since it was invented. The question is no longer whether auctions are efficient, but whether they will remain the main way media is traded.

How we got to first price

In the early years of real-time bidding, most exchanges used a second-price auction: the highest bidder won, but paid one cent more than the second-highest bid. In theory, this encouraged buyers to bid exactly what an impression was worth to them.

Header bidding changed that. Once publishers let several exchanges compete at the same time, each exchange ran its own auction and then competed in a second one. Second-price logic stopped making sense across layers, and prices became hard to understand. The industry moved to first-price auctions, where the winner pays what it bid. In March 2019, Google announced the move of Google Ad Manager to first price, and the rest of the market followed.

The answer to first price: bid shading

First price created a new problem for buyers: if you pay what you bid, bidding your true value means overpaying. DSPs responded with bid shading, algorithms that estimate the lowest price likely to win and bid just above it.

Bid shading is now standard, and it is invisible to most of the people whose budgets it manages. A study commissioned by Cognitiv and conducted by Alter Agents, based on a survey of 251 US media buyers, estimated that misunderstood bid shading leads to about 6.6 billion dollars of wasted ad spend per year. Whether or not that number is exact, it shows that the auction is no longer a simple meeting of buyer and seller. It is a contest between algorithms on both sides.

The quiet rise of deals

At the same time, more and more programmatic money no longer goes through the open auction at all. Buyers increasingly use private marketplaces (PMPs), programmatic guaranteed deals and curated packages, where the price floor, the inventory and sometimes the audience are agreed in advance. Technically, many of these deals still run through an auction, but the competition is limited to a small number of invited buyers with pre-agreed terms.

The reason is not that auctions stopped working. It is that buyers want predictability and quality: known publishers, verified supply and fewer surprises. The open auction is still where most scale lives, but premium inventory is increasingly sold through narrower doors.

Regulators are looking at the auction too

Auction rules are no longer only a technical topic. In April 2025, a US federal court found that Google had illegally monopolized the markets for publisher ad servers and ad exchanges. On September 2, 2026, Judge Leonie Brinkema declined to order the sale of Google's AdX exchange and chose behavioral remedies instead. The exact obligations are still being finalized, but the case has made one thing clear: who runs the auction, who sees the bids and who sets the rules is now a legal question as much as a commercial one.

The real challenger: AI agents

The biggest change may come from somewhere else entirely. In October 2025, a group of companies including PubMatic, Scope3 and Swivel launched the Ad Context Protocol (AdCP), an open standard that lets AI agents on the buy side and the sell side communicate and transact directly. Its media buying protocol is designed to let agents negotiate and execute campaigns without real-time, impression-level auctions, while still working with existing DSPs, SSPs and ad servers.

A month later, in November 2025, the IAB Tech Lab released the Agentic RTB Framework (ARTF) for public comment. Instead of replacing real-time bidding, it brings agents inside it: companies can deploy their models as secure containers within another platform's infrastructure, for tasks such as pre-bid fraud detection or deal activation. One goal is to cut latency in a system where typical response times are still in the hundreds of milliseconds.

These two approaches point to two possible futures. In one, agents negotiate packages of media upfront and the per-impression auction becomes less important. In the other, the auction stays, but it becomes faster and smarter, with more intelligence running at every step.

So, are auctions here to stay?

Our view is yes, but not in their current form. A few things seem likely:

1. The auction will stay as the price discovery layer. As long as millions of impressions with different values are created every second, some form of competitive pricing will be needed. Agents can negotiate deals, but someone still has to decide what a single CTV impression in a given context is worth.
2. Fewer participants per auction. Sending every request to every buyer is expensive for everyone. The trend is toward smaller, better-targeted auctions, through deals, curation or prediction.
3. More intelligence before the auction. The most valuable decisions are moving upstream: which buyers should see a request at all, what floor makes sense, and whether the request is clean enough to send.

This is exactly how we designed PeopleAds. Before any auction happens, our engine verifies the request, predicts which buyers are most likely to bid, routes the request only to them and sets a dynamic floor. The auction still decides the final price, but it runs on a cleaner, smaller set of requests and buyers who actually want them.

The auction is not dying. It is being surrounded by intelligence. The platforms that win will be the ones that decide well before the bidding even starts.