What is real-time bidding (RTB)?

The sequence

1. A user opens a page or screen, and an ad slot becomes available.
2. The publisher's [SSP](/resources/glossary/supply-side-platform-ssp) builds a bid request describing the opportunity — the surface, the format, the context, the floor.
3. The request fans out to an exchange and on to many DSPs.
4. Each DSP values the impression against its campaigns and returns a bid, or declines.
5. The auction resolves and the winning creative is returned.
6. The ad renders. Total elapsed time, typically around 100 milliseconds.

First price and second price

The industry ran on second-price auctions for years: the winner paid a penny above the second-highest bid, which in theory makes honest bidding optimal.

It did not survive contact with intermediaries who could see both sides and adjust. Most of the market moved to first price, where the winner pays what they bid. That is more transparent and puts the burden of not overpaying back on the buyer, which is why bid shading — algorithmically bidding slightly below your true valuation — became standard practice rather than an edge.

Where RTB strains

The model assumes three things that are becoming less true: that an impression can be described adequately in a structured request, that a user can be identified well enough to value, and that a hundred milliseconds is available.

Identity has already gone in most contexts, which is why contextual signals have regained importance. Description is the binding problem on conversational surfaces — a bid request has no field for what the user is trying to do. And the latency assumption fails outright when the surface is generating a response rather than loading a page.

RTB is not going away on the inventory it already serves. It is simply not the mechanism that will clear conversational inventory without significant extension.