Header Bidding vs Waterfall: Why the Old Way Leaves Money on the Table

For years, most websites sold their ad space through a waterfall: a fixed list of ad networks, asked one after another, until someone agreed to fill the impression. It was simple and it worked, but it has a flaw that quietly costs publishers money on every page view. Header bidding was invented to fix exactly that flaw.

This article explains both models without jargon, shows where the waterfall loses money, and helps you decide what your site should use.

How the waterfall works

In a waterfall, your ad server keeps a ranked list of networks, usually ordered by the average price each one paid in the past. When an ad slot needs an ad:

1. Network 1 is asked first. If it wants the impression at or above the price set for its position, it wins.
2. If it passes, the request drops to Network 2, then Network 3, and so on.
3. The first network that accepts gets the impression. If nobody accepts, the slot stays empty or shows a house ad.

The name comes from the picture: the request trickles down the list like water down a set of steps.

Where the waterfall loses money

The problem is that the waterfall is based on averages, while every impression is different. Network 1 might pay $1.00 on average, but for this particular reader, on this particular page, Network 4 might be willing to pay $3.00. The waterfall never finds out, because Network 1 said yes first.

That creates three recurring losses:

• Underpriced impressions. The first acceptable offer wins, not the best offer.
• Unfilled impressions. Each step adds a little delay, so long waterfalls often give up before reaching a network that would have bought.
• Stale ordering. The list is updated from historical averages, so it always reflects last month's market, not this second's.

How header bidding works

Header bidding replaces the queue with an auction. When an ad slot needs an ad, every connected network is asked at the same time. Each one replies with the price it is willing to pay for that specific impression. The highest valid bid wins.

It is called "header" bidding because the first versions ran from a script placed in the <head> of the web page, before the page's ad server was called. Today the auction can run in the browser, on a server, or through a mediation platform, but the principle is the same: everyone bids, the best bid wins.

Side-by-side comparison

A worked example

Say five networks are connected to a 300×250 slot. For one impression they would pay:

If Network A's waterfall threshold is $0.40, the waterfall sells the impression to Network A for $0.42 and stops. Header bidding asks everyone and sells it to Network C for $1.24, almost three times as much, for exactly the same reader and the same ad slot.

Across hundreds of thousands of impressions a month, that gap is the difference between a side income and a real one.

The honest downsides of header bidding

Header bidding is better for revenue, but a badly built setup can create new problems:

• Page weight. Running a separate script for every bidder in the browser adds code to every page. Too many bidders can hurt your page speed.
• Timeouts. The auction has to wait for bids. If the timeout is too long, ads appear late; too short, and slow bidders are excluded.
• Complexity. Managing adapters, versions and settings for each bidder yourself is real engineering work.

A mediation platform solves most of this by running the auction behind a single tag: one request from the page, a capped wait for bids, and one winning ad returned. You keep the revenue benefit without maintaining the machinery.

Floors: the setting that decides how the auction behaves

A floor price is the minimum bid you will accept for a slot. In header bidding, floors work differently from waterfall thresholds:

• A floor that is too low lets cheap bids win when there is no better competition, which is still better than an empty slot.
• A floor that is too high rejects bids you should have accepted and pushes your fill rate down.

The practical approach: start with a low or zero floor, collect at least a week of data, then raise floors slowly on your best-performing units while watching fill rate. If fill drops sharply, step back. Our guide to eCPM, RPM and fill rate shows how to read these numbers together.

Hybrid setups

Some publishers run a simultaneous auction first and keep a short fallback list afterwards, for example a direct deal or a house ad that fills anything the auction leaves empty. That is reasonable. What you want to avoid is a long waterfall as the main way your impressions are sold.

How to tell which model your site uses today

• If your ad settings contain a numbered list of networks with a price for each position, you are running a waterfall.
• If each impression is offered to several buyers and your reports show a winning bid per request, you are running an auction.
• If you only use one network's code and nothing else, there is no competition at all, and moving to mediation is the single biggest improvement available to you. Start with what ad mediation is.

How Advorex Media runs the auction

Advorex Media runs a real-time auction for every ad request: the connected networks - including Prebid Server and Prebid.js bidders - are asked together, bids below your floor are dropped, and the highest bid wins. Every auction has a deadline well under one second, so a slow network misses that auction instead of holding your ads back. You set a floor per ad unit from your dashboard, and your reports show impressions, eCPM and your own earnings, so you can see what the competition is doing for you. All of it runs behind one lightweight tag.

Key takeaways

• The waterfall sells to the first network that accepts; header bidding sells to the highest bidder.
• Because every impression is different, the waterfall regularly underprices your inventory.
• Header bidding raises both price and fill rate, but should be run behind a light setup so pages stay fast.
• Floors are powerful: start low, measure, and raise them gradually.