Real-Time Bidding

What happens in a real-time bid

Real-time bidding is an automated auction held for a single ad impression in the interval between a page starting to load and the ad slot rendering. The sequence: ad opportunity → bid request → buyer evaluation → auction → ad render — all typically inside a hundred-odd milliseconds.

Step by step:

1. The opportunity. A user opens a page or app screen with an ad slot. The publisher's stack — possibly via header bidding, an SSP, or both — packages what it knows: slot size, page context, geo, device, available identifiers, floor price.
2. The bid request. That package goes out to demand sources as an OpenRTB request (the IAB's protocol that standardizes the conversation), fanning out to dozens of DSPs.
3. Buyer evaluation. Each DSP matches the request against active campaigns: does the user fit targeting? what does the predicted conversion rate imply this impression is worth? Bids are computed per impression, not per placement.
4. The auction. Bids return within a strict timeout (late bids simply lose). The auction — historically second-price, now predominantly first-price — picks a winner, applies floors, and returns the ad markup.
5. Render and record. The creative renders; impression, viewability, and spend events flow back to both sides' reporting.

The detail that explains most of programmatic's behavior: pricing happens per impression, per user, per moment. Two visitors on the same page see different ads at different clearing prices because different buyers valued them differently.

First-price auctions and why they matter

The industry's shift from second-price to first-price auctions (completed across major exchanges around 2019) changed bidding logic fundamentally. In a second-price auction, bidding your true value was optimal — you'd pay just above the runner-up. In a first-price auction, you pay what you bid, so naive truthful bidding overpays. DSPs respond with bid shading: algorithms that estimate the minimum bid likely to win and bid closer to that than to full value.

Consequences worth knowing:

• For buyers, shading quality is a real differentiator between DSPs; poor shading quietly taxes every impression.
• For publishers, first-price made floor prices more strategic — a floor is now a direct negotiating position rather than a backstop.
• For everyone, "auction price" stopped being a stable reference: the same inventory clears at different prices depending on who shades how aggressively.

Auction price versus total cost

The clearing price is not what the advertiser ultimately pays per unit of value. Stacked on top: DSP platform fees, data costs, verification fees, and on the sell side the SSP/exchange take. The spread between advertiser spend and publisher net — documented repeatedly by supply-chain transparency studies — is why supply-path optimization exists as a discipline. When comparing channels or partners, compare cost per verified outcome, never raw clearing prices.

Not all programmatic is open auction, either. The deal hierarchy runs: open auction (anyone bids) → private marketplace (invited buyers, often with floor or priority) → preferred deals (fixed price, no exclusivity) → programmatic guaranteed (reserved volume at negotiated price, transacted through the same pipes). Much premium inventory never touches the open auction at all.

Latency: the physical constraint

Everything above happens while a user waits. Timeouts cascade through the chain — the header-bidding wrapper waits perhaps 1–1.5 seconds, the SSP gives DSPs a couple hundred milliseconds, the DSP's internal models get a fraction of that. The engineering consequences:

• Bids that arrive late are discarded — a slow but high bid loses to a fast mediocre one, so infrastructure quality directly affects both buyer win rates and publisher yield.
• Publishers tuning header bidding trade demand breadth against page speed; every added partner widens the auction and lengthens the tail latency.
• Mobile networks make all of this worse, which is one reason app monetization leans on mediation SDKs with different timing behavior.

Common mistakes

• Assuming every programmatic deal is an open auction. PMPs, preferred deals, and programmatic guaranteed move large budgets through RTB infrastructure without open competition; analyzing "programmatic" as one bucket misreads both price and quality.
• Ignoring latency. Buyers blame targeting when their bidder is simply slow; publishers add partners until timeout losses eat the theoretical demand gain.
• Confusing the auction price with total media cost. Fees between buyer and seller can be a large share of spend; clearing-price dashboards systematically understate true cost.
• Treating bid requests as truth. Request fields are publisher-declared; verification (ads.txt checks, post-bid measurement) exists because declarations can be wrong or fraudulent.
• Optimizing bids without optimizing supply paths. The same impression arrives via several routes at different effective costs; a perfect bid through a wasteful path still overpays.

FAQ

How fast does an RTB auction actually run? End-to-end budgets typically allow roughly 100–300 milliseconds for the auction itself, inside a page-level patience window of a second or so. Individual DSPs often must respond within ~100ms including network time.

Is RTB the same as programmatic? No — RTB is the auction protocol; programmatic is the broader category of automated, pipe-driven buying, including guaranteed deals that have no auction. All RTB is programmatic; not all programmatic is RTB.

Who actually runs the auction? It can happen in several places, sometimes layered: the header-bidding wrapper on the page, each SSP/exchange internally, and the ad server's final decisioning. One impression can legitimately pass through multiple sequential auctions before rendering — each with its own rules and fees. The programmatic path walks this chain end to end.

What data is in a bid request? Slot and page context, device and geo, privacy-permitted identifiers, and declared publisher info. Privacy regulation and identifier deprecation are steadily thinning the user-specific portion, pushing valuation toward contextual and first-party signals. Publishers operating in GDPR markets must transmit a valid consent signal for each impression to unlock behavioral demand.

Why did first-price replace second-price? Header bidding created parallel auctions whose results fed into other auctions, making "second price" ill-defined and enabling hidden fee games. First-price is less elegant for bidders but transparent: the winning bid is the price, auditable by everyone in the chain.