Real-Time Bidding and the Programmatic Exchange of Attention

A negotiation becomes an auction

Real-time bidding is the practice of buying and selling a single unit of digital ad space, one impression, through an instantaneous auction that runs while a webpage is still loading. The reference description used across the ad-technology industry calls the process similar to financial markets, and the comparison is not decorative. An RTB auction has bidders, a clearing price, and a settlement, all compressed into the time it takes a page to render.

That structure replaced something much older and slower. For most of the commercial internet's first decade, buying an ad meant a salesperson and a media buyer agreeing a price for a block of space, typically weeks or months ahead of when it would run. The price was set once, by two people, for a stretch of inventory neither could fully predict the value of in advance.

RTB collapsed that negotiation into a transaction no person ever sees. When a publisher's page loads, a request goes out to an exchange describing the available impression: the site, the placement, sometimes what is known about the visitor. Multiple buyers submit bids in response. The exchange picks the winner and places the ad. The full sequence, request, bids, placement, completes in milliseconds, finished before the page requesting it has finished loading.

The unit being priced is not a magazine page or a billboard. It is one person's attention, for one moment, on one screen, and it is priced fresh every time that moment recurs. Many of these auctions use what is known as a second-price format: the highest bidder wins the impression but pays only an amount just above the second-highest bid, a design borrowed from auction theory that predates the internet by decades. The winner does not have to guess how much value to leave on the table, which is one reason the mechanism could scale from a single company's own inventory into a general market covering advertisers and publishers who had never dealt with each other before.

Repeated at the scale of the modern web, that fact is easy to overlook. Billions of these auctions run every day, each one small enough to be invisible to the person whose attention is being cleared, and large enough in aggregate to be one of the more literal examples anywhere in commerce of a resource priced by algorithm rather than by a person.

The remnant trade that came before

Before the auction existed, publishers with more ad space than they could sell directly relied on ad networks, aggregators that bundled the leftover inventory across many sites and sold it onward in bulk. This was the remnant market: real, valuable, and largely opaque. A publisher rarely knew what a given impression on their site actually sold for once it passed through a network, because the price was set for a batch, not for the individual view.

An ad exchange is a different kind of intermediary. Structurally, it is a marketplace that connects the buy side, demand-side platforms representing advertisers, to the sell side, supply-side platforms representing publishers, and lets them transact directly rather than through a network's private bulk deal. Both sides can see the mechanism even where they cannot see each other's strategy.

RTB's contribution was to make that marketplace price every impression individually and in the open, rather than clearing a batch at one blended rate. The remnant trade did not disappear so much as get repriced, impression by impression, in a format closer to a trading floor than a media sales call.

The distinction between a network and an exchange looks technical, but it is really a distinction about who can see the price. A network buys inventory from many publishers and resells it under its own pricing, so a publisher effectively sold blind, trusting the network to pass on a fair share of whatever it collected. An exchange does not buy the inventory at all. It clears a trade between two named counterparties and takes a fee for doing so, closer to how a stock exchange operates than how a wholesaler does.

The efficiency case for this was not instant, but it did not take long to land. By 2010, the difference in outcome between static, bulk-negotiated buying and live per-impression auctions was clear enough to most advertisers that demand for programmatic buying accelerated across the industry, no longer a niche technique but the direction the whole market was heading.

Five months in 2007: building the exchange

The infrastructure that would eventually host this auction was assembled two years before the auction itself went live, and it was assembled fast. In April 2007, Yahoo bought the ad exchange Right Media. In May 2007, Google bought DoubleClick. In August 2007, Microsoft bought AdECN. Three of the internet's largest companies had each acquired a foundational ad exchange inside a single five-month stretch.

None of these companies were buying a live, real-time bidding market, because RTB as a per-impression auction would not exist for another two years. What they were buying was the plumbing: the exchange technology, the publisher relationships, the position to be the marketplace once a market formed around it. Ownership of the pipes came before the flow of trades that would run through them.

The consequence is structural rather than dramatic. Once the world's attention began to be priced through auctions running on this infrastructure, that pricing ran through exchanges controlled by three American companies, regardless of whether the advertiser sat in Mumbai, the publisher in Berlin, or the person whose attention was for sale anywhere at all. A German publisher's inventory and a South African advertiser's budget could both clear through the same American-owned exchange, on terms that company set.

That concentration cut both ways, in the way most infrastructure ownership does. It gave publishers and advertisers everywhere access to a liquid, transparent market for attention that the old bilateral sales call could never have scaled to, a genuine widening of who could buy and sell an impression on equal footing. It also meant the rules of that market, what counted as a valid bid, whose algorithm got priority, which data could travel with a request, were written by exchange operators who answered to shareholders in one country, not by any body representing the advertisers, publishers, or audiences the exchange served worldwide.

None of this was hidden at the time. Trade coverage of the three deals read them, correctly, as each company racing to secure a piece of infrastructure it expected the future of online advertising to run through, not as three unrelated corporate purchases that happened to land in the same year. The race itself is a small window into how the geopolitics of a medium usually work: the decisive positioning often happens years before the public, or even most of the industry, notices that a new market has formed around what was bought.

September 2009: the auction goes live

The pipes started carrying live trades in September 2009, when Google rebuilt its DoubleClick Ad Exchange to run real-time, per-impression bidding. Competing ad networks could plug their own bidding algorithms into the exchange and bid against each other for each individual impression as it became available, rather than buying access to a fixed slice of inventory in advance. This is the point where the comparison to a financial exchange stopped being a figure of speech and started being closer to a description.

Demand-side platforms and supply-side platforms took on roles that map fairly directly onto exchange brokers: DSPs execute bidding strategy on behalf of advertisers, SSPs manage the sell side on behalf of publishers, and the exchange in the middle clears the trade. Attention had become the underlying asset, and DSPs and SSPs had become the brokers who traded it, on an exchange that took its own cut of every transaction that passed through.

The protocol Google used to open its exchange to competing bidders, OpenRTB, was not proprietary to any single company; it became a shared specification that other exchanges adopted too, which meant a demand-side platform built once could bid into more than one exchange rather than needing a separate integration for each. That openness at the protocol layer sat awkwardly next to the concentration at the ownership layer. Anyone could learn to speak the language. A small number of companies still owned the rooms where the conversation happened.

From $23.5 billion to $57 billion: attention as an asset class

The scale of what this auction now moves is measurable in the United States alone. US real-time bidding spend rose from roughly $23.5 billion in 2018 to roughly $57 billion in 2023, more than doubling in five years. That growth did not happen because advertisers discovered a new place to spend money. It happened because the auction displaced the older, direct-negotiated sale as the default way digital ad inventory changes hands, no longer the exception to it.

The economic shift underneath that number is the one this series keeps returning to under different names: a resource once allocated by relationship and negotiation became a resource allocated by price, discovered continuously, in public, by algorithm. Attention is the resource here, and the auction, not a salesperson, is now the mechanism that sets what a given moment of it is worth.

This produced real efficiency. An advertiser can buy exactly the impressions worth buying, at a price discovered in the moment rather than guessed at months in advance, and a publisher can sell exactly the inventory worth selling instead of bundling it off to a network at a discount. Granular, live price discovery is a genuine improvement over the opacity of the remnant trade it replaced.

It also concentrated the value capture at a new layer. Every trade that clears through an exchange pays a toll to whoever owns the exchange, whether the advertiser and publisher notice it or not, and that toll accrues at the auction's chokepoint, not at either end of the trade. A market built to be efficient for buyers and sellers was, at the same time, built to be lucrative for the operator standing between them.

The mechanism also did not stay confined to the display banner it started with. The same live-auction logic extended over time to video, audio, and eventually connected-television inventory, each format bringing another slice of previously negotiated ad buying under the same real-time pricing model. What began as a way to sell leftover banner space on ordinary web pages became, in effect, the default clearing mechanism for a growing share of the world's advertising, across nearly every screen a person's attention might land on.

The auction's next owner

What happened in 2007, infrastructure built by a small number of American companies before the market that would run on it fully existed, is not unique to advertising. A comparable pattern is forming around the systems that now decide whether a business is named at all: the AI answer engines a growing share of buyers ask instead of typing a search. Those systems are, again, owned by a small number of companies, and much of that position was settled before most businesses had a way to measure whether they were being cited by them.

The currency is different. RTB priced a moment of attention by auction, in dollars, per impression. An answer engine does not run an auction for the answer it gives; it decides, by its own process, whether to name a business at all, and there is no bid a business can place to buy its way into that answer. What carries over is the position: whoever controls the exchange, or the engine, sets the terms under which everyone else's attention, or visibility, gets priced or granted.

The throughline across both eras is the one this history keeps returning to. Whoever owns the dominant medium of an age, the auction that prices attention or the engine that decides who gets named in an answer, ends up setting economic terms for everyone who depends on being seen through it, no matter where that business, publisher, or buyer happens to sit in the world.

One plausible outcome, and it is a scenario rather than a forecast, is that a market resembling RTB eventually forms around AI-answer placement, complete with intermediaries, standard protocols, and a small number of owners setting the rules, the way one formed around ad placement after 2007. Nothing in the record so far confirms that outcome, and the mechanism, if it arrives, may look nothing like an auction. What the RTB history does confirm is the earlier pattern such a market would have to pass through first: whoever builds the exchange tends to own it for a long time afterward.