What is real-time bidding (RTB)? A plain-English primer

Every time a web page loads with an ad on it, a small auction happens that you never see. It opens and closes in roughly the time it takes to blink. Dozens of buyers are invited, they each name a price, a winner is chosen, and the winning ad is painted onto the page — all before the rest of the content finishes rendering. That auction is real-time bidding, and understanding what real-time bidding is is the fastest way to understand how most digital advertising gets bought and sold today.

This is a primer. If you are new to ad tech — whether you run a site that shows ads or a brand that buys them — this post explains RTB from the ground up: the moving parts, the flow, the tight time budget, and why the industry abandoned the older way of doing things.

What real-time bidding actually is

Real-time bidding is the practice of auctioning a single ad impression, one at a time, in real time. The unit being sold is not a campaign or a block of inventory. It is one impression: one ad slot, on one page, loading for one specific person, right now.

The older model sold inventory in bulk and in advance. A salesperson negotiated a price for, say, a million impressions on a site’s homepage, and those impressions were delivered over the following weeks. RTB inverts that. Instead of pricing inventory ahead of time, it prices each impression at the exact moment it becomes available, by asking the open market what that specific impression is worth to whoever wants it most.

That shift — from selling inventory in advance to auctioning it impression by impression — is the whole idea. Everything else is mechanics.

How RTB works: request, bid, auction, creative

The clearest way to learn how RTB works is to follow a single impression through its lifecycle. There are four steps.

The request. A person opens a page. The ad slot on that page signals that an impression is available. That signal becomes a bid request — a standardized packet of information describing the impression — which is broadcast out to potential buyers.

The bid. Each interested buyer evaluates the request and decides, in a few milliseconds, whether to bid and how much. A buyer might bid on this impression because the user matches a target audience, the page is brand-appropriate, or the buyer simply has budget to spend. Buyers that want it return a bid — a price, plus the ad they would run if they win.

The auction. All the bids are collected and compared. The highest bid wins. This auction is the heart of the system, and the rules that govern it (more on those below) determine what the winner actually pays.

The creative. The winning buyer’s ad — the creative — is returned and rendered into the slot. To the person looking at the page, it simply appears alongside the content. They have no idea an auction just happened.

That is the full loop: request goes out, bids come back, the auction resolves, the creative renders.

The 100-millisecond budget

The remarkable part is the speed. This entire programmatic auction typically has to complete in around 100 milliseconds — a tenth of a second.

The budget is tight for a simple reason: the auction is racing against page load. A person should not sit and stare at an empty rectangle while buyers deliberate. If the auction takes too long, the impression is wasted — the slot fills with something default, or nothing at all, and everyone loses. So the system imposes a hard timeout. Buyers that do not respond in time are simply left out of that auction.

A hundred milliseconds is not much time to make a financial decision. Within that window, a buyer has to receive the request, look up everything it knows about the user and the context, run its bidding logic, calculate a price, and send a bid back over the network. This is why the buying side runs on heavily optimized infrastructure: the cost of being slow is being excluded entirely.

First-price vs. second-price auctions

Once the bids are in, the auction needs a rule for what the winner pays. There are two classic rules.

Second-price. The highest bidder wins but pays one cent more than the second-highest bid. If you bid $5.00 and the next-highest bid was $3.00, you win and pay $3.01. The appeal is that it encourages buyers to bid their true value without fear of overpaying, since the price is set by the competition, not by their own bid.

First-price. The highest bidder wins and pays exactly what they bid. Bid $5.00, win, pay $5.00. Simpler, and more transparent — there is no hidden gap between what you offered and what you are charged.

For years, second-price was the default. But as the supply chain grew more complex and harder to audit, the industry shifted decisively to first-price auctions; by the early 2020s, first-price had become the standard across the major exchanges. First-price is easier to reason about — what you bid is what you pay — though it pushes buyers to think carefully about not overbidding, a practice known as bid shading.

Who’s who: DSPs, SSPs, and ad exchanges

RTB has a vocabulary, and three acronyms do most of the work. Each represents a role in the marketplace.

SSP (supply-side platform). Works for the publisher — the site or app with ad space to sell. The SSP packages up available impressions, sends out bid requests, and tries to get the highest price for each one. Think of it as the seller’s agent.

DSP (demand-side platform). Works for the advertiser — the brand or agency that wants to buy impressions. The DSP listens to incoming bid requests, decides which ones are worth bidding on, and submits bids on the advertiser’s behalf according to its targeting and budget rules. The buyer’s agent.

Ad exchange. The neutral marketplace in the middle where the two sides meet. The exchange receives impressions from the supply side, runs the auction, and connects the winning bid to the impression. It is the venue, not a participant.

The shorthand: SSPs bring the supply, DSPs bring the demand, and the exchange is the floor where they transact. Most real-world setups layer more intermediaries on top, but those three roles are the foundation.

A word on header bidding

One refinement deserves a mention because you will hear the term constantly: header bidding. In the early days, a publisher’s ad server would offer an impression to demand sources sequentially — ask the first, and only if it declined, ask the next, and so on down a ranked list. This was the “waterfall,” and its flaw was that a buyer further down the list might have paid more but never got the chance. Header bidding fixed this by soliciting bids from many sources simultaneously, before the ad server makes its call, so every buyer competes for every impression at once. More competition, higher clearing prices, and far better visibility into what inventory is truly worth.

Why RTB replaced the old way

The older model — direct-sold deals negotiated by salespeople, and the waterfall that filled whatever they did not sell — had two structural problems. It was slow, requiring human negotiation for inventory that exists for only an instant. And it was inefficient: a publisher could not easily know whether a directly negotiated price was the best price the market would bear, and an advertiser could not buy precisely the impressions it wanted without buying a lot it did not.

RTB solved both. By auctioning each impression to the open market at the moment of availability, it lets price discovery happen automatically and continuously. The publisher gets the highest bid the market will offer for that exact impression. The advertiser pays only for the specific impressions that fit its goals, and skips the rest. No salesperson required, and no guessing about value — the auction reveals it. That combination of efficiency and precision is why RTB went from a niche experiment to the dominant way display advertising is bought and sold.

Where the auction goes next

The mechanics described here grew up around the web page and the mobile app screen — surfaces where an ad slot is a rectangle waiting to be filled. But the auction itself is more general than that. At its core, RTB is just a fast, standardized way to discover what an impression is worth and match it to a buyer in the moment. There is no fundamental reason that logic has to stay bolted to a banner.

As new surfaces emerge — video and audio streams, and increasingly conversational and streaming interfaces where content arrives as a response rather than a page — the same request-bid-auction-creative loop can be adapted to fit them. The slot changes shape; the auction does not have to. It is worth watching which of these mechanics carry over cleanly and which need rethinking. (Elo, founded this year, is one of the early efforts exploring what an auction looks like when the surface is a conversation rather than a page.) If the terminology in this post is new to you, a glossary of conversational advertising terms is a good next stop.

For now, the takeaway is simple. Real-time bidding turned the ad impression into a tradable unit and built a tenth-of-a-second auction around it. Once you see that pattern, you start seeing it everywhere — and you start to wonder where it shows up next.