RTB Explained: What Actually Happens in the 100ms Before Your Ad Shows

The anatomy of a bid request

When a user opens an app with ad inventory, the publisher's supply-side platform packages the impression into a bid request: the app, the placement size, the geography, the device, and whatever privacy-compliant signals are available. That request fans out to exchanges and DSPs simultaneously.

Your DSP has roughly 100 milliseconds to decide: is this user worth bidding on, and at what price?

How the price gets set

Most in-app inventory now clears in first-price auctions — you pay what you bid. That changed the game: naive bidding overpays, so modern DSPs use bid shading models that predict the minimum price likely to win and bid just above it.

The intelligence layer is what separates platforms. A model trained on your post-install outcomes knows that a puzzle-game player in Toronto at 9pm is worth $4.20 to you, while a similar-looking impression is worth $0.40 — and bids accordingly, millions of times a day.

Why win rate is a health metric

Win rate — the share of auctions you enter that you actually win — tells you whether your bids are calibrated to the market. Too low and you're starved of inventory; too high and you're likely overpaying.

Healthy campaigns typically win enough to learn continuously without clearing every auction. Your DSP should report this transparently, per source, so you can see exactly where budget is competitive and where it isn't.

What this means for advertisers

You don't need to run the auction yourself — but you should demand to understand it. Ask your DSP how bids are priced, how shading works, and whether you can see win rates by exchange. The answers reveal whether you're buying intelligence or just access.