Back to blog
epom ad server background
ad server abstract visualization

Two Ways to Optimize Ad Serving: Waterfall vs. Header Bidding

Jan 30, 2020  ·  8 min read
Kate Novatska
Kate Novatska, Marketing expert at Epom

Ad trading via auction was the best decision ever made. No matter which price you put, there is always someone willing to pay more. But, what is the best way to not mix up the types of programmatic bidding and get prime impressions? The ultimate waterfall vs. header bidding comparison is waiting for you below.

5 December 1766. London's Pall Mall is full of British noblemen in anticipation of the first Christie's auction in history. Soon, a former print warehouse will open its doors, welcoming all who arrived to celebrate the best art pieces from England and beyond. They are all put on display and set for sale, but with one condition: who ever has the highest bid gets the item.

Christie's auction first saleroom

The appeal of art auctions has not decreased over time — it's quite the contrary. Christie's grew into the largest auction house with 10 salerooms and a presence in 46 countries. The auction proved to be the best way to sell something that doesn't have a definitive price.

In 2009, miniature pieces of commercial art, advertising creatives, started to be traded via auction to optimize publisher's inventory. With one major difference: real-time bidding auctions are completely automated.

In the early days of ad tech, ads were served through a daisy-chaining process aka waterfalling, where the bid of each advertiser should meet the minimum price set by the publisher. As technology advanced, the programmatic advertising ecosystem shifted to header bidding auctions, in which the highest bidder wins, just like what happened at Christie's.

What is Daisy-Chaining or Waterfall Tags?

A waterfall auction is the old school way of ad serving. It came from the same era as brick cell phones and disk drives. The waterfall auction was slow, yet an effective solution to increase publisher's fill rates while modern algorithms were unavailable. Elaborating a bit more on the waterfall definition in advertising, I'd say the following:

Waterfalling, or daisy-chaining, is a method to sell the publisher's inventory in which the publisher sets the priority for each advertiser or ad network they're connected to. They also specify a price floor, i.e. the minimum accepted price for this ad placement. The inventory slot is sequentially offered to demand partners in order of priority. Once someone meets the price floor, the impression is sold.

Waterfall auction in an ad server

Inventory is passing the route from top to bottom, similar to the water in the river flowing down the cliff to the basin. That's why it's called a waterfall auction, although it's not quite an auction in its nature. The inventory is often going to the one who is first in line, not the one who bids the highest. This was effective in the early days of programmatic advertising, but not now, when we have RTB and header bidding as a product of its evolution.

What is Header Bidding Auction?

Comparing header bidding vs. waterfall, the first difference is that header bidding is an automated auction in the truest sense of this notion. Publishers exhibit their inventory at several ad exchanges, which in turn allow multiple demand partners to bid on those “lots” simultaneously. Unlike the waterfall, in header bidding the highest bid wins.

Programmatic header-bidding auction

That's only a superficial explanation of what header bidding is. To understand how header bidding works, you need to learn more about how exactly all parties interact with each other.

The core element in this type of RTB is the header code. Publishers place it in the header of their website. Each time someone visits the website, the browser requests ad serving on a particular page.

  1. A browser or server sends the request to a supply-side platform — the tool that allows publishers to sell inventory programmatically.
  2. An SSP handles received data and pushes the request forward to an ad exchange — the open marketplace which serves as a center of trade between supply and demand partners.
  3. An ad exchange forwards it to a demand-side platform — the advertiser's piece of software used to launch and manage their programmatic campaigns.
  4. A demand-side platform analyzes publisher's data and matches the inventory to the advertisers with suitable targeting options.
  5. Advertisers simultaneously bid on their targeted audience. The ones who bid the highest price, get the placement.

Publishers can connect their websites to several SSPs to maximize fill rates. The whole process takes 100-200 milliseconds. An SSP can initiate up to 10 auctions while the page is loading (assuming the page load speed is 1 sec).

RTB vs. Header Bidding in Programmatic

Real-time bidding and header bidding may be easily confused with one another. They are often used as synonyms today and that's not necessarily wrong. A modern RTB often involves header bidding technology, so both the RTB and header bidding concepts may be applied to explain what's happening inside your DSP.

Still, RTB has a wider definition. If header bidding is not implemented by the publisher, the programmatic platform will use an older ad tag model to sell inventory. So, an RTB protocol can be based on both ad tags, header bidding, or even an SDK (software development kit) depending on the situation.

First-price vs. Second-Price RTB in Programmatic

Header bidding auctions in programmatic can be first-price and second-price. A second-price auction is the initial model of the RTB process, where the highest bidder doesn't pay the price equivalent to their bid, but $0.01 higher than the bid of the second bidder.

A first-price header bidding auction is more beneficial for publishers, as the highest bidder pays exactly the price equivalent to their bid. It's the simpler framework that allows maximizing yield from each inventory sale. Some experts like Verizon's Media Deepti Phatak, believe that first-price auctions give an edge to advertisers as well, as this model is more transparent and abolishes hidden fees & manipulations inherent to second-price auctions.

First-price vs. second-price auctions

Is RTB Always Programmatic?

Some novice users also struggle to distinguish programmatic vs. RTB. Most programmatic sales are based on RTB, but that's not the only way of programmatic ad serving. There is such a thing as programmatic direct, where the ad placements are already reserved for specific advertisers, yet the ad serving is done via automated algorithms.

RTB vs. direct vs. programmatic

Is Waterfalling Still Viable in 2020?

If you're going to exclusively use waterfall as your method to automate your media buying, then it's a resounding NO. There is no reason in 2020 to lose a share of your revenue and give priority to low-paying demand partners.

When the first advertiser doesn't meet the publisher's price floor, an ad server passes a request back to the next in line. And if, say, the first three advertisers bid below the floor, the waterfall may increase ad serving latency.

The only viable case when a publisher can use waterfalling is when they need to organize their equally paying direct partners in a specific order. Also, publishers may prioritize some advertisers over others and start with waterfall first before getting involved in a real-time bidding auction.

Ad networks often use an ad server and its waterfalling feature for premium placements, while programmatic tools are utilized for regular and remnant inventory. It doesn't mean that you can't find premium partners while trading ads via RTB; you just can't be 100% sure. Everything is automated and based on the advertiser's targeting preferences, not on their overall credibility.

Header Bidding — Effective Solution for Publishers

Considering how header bidding works, this method of RTB advertising increases healthy competition and reduces time spent on campaign management. Thus, publishers get a clear advantage of making more money for a smaller time investment.

Summarized header bidding benefits for publishers:

  • Header bidding ensures that inventory is sold to the highest bidder, while waterfall sells it to the first-in-line advertiser.
  • Header bidding happens in real-time, while waterfall embraces a sequential auction model that increases latency.
  • Header bidding requires no ad server adjustments, while waterfall tags need to be constantly updated.
  • Header bidding implies a lot of versatile demand sources, waterfall implies only sources you already know.
  • Header bidding ensures a better fill rate, while waterfall sometimes fails to sell inventory on time.

It may seem that header bidding only matters for publishers, as it's a publisher's method of data transmission, but that's not quite true.

Let's review some benefits for advertisers brought to the table by header bidding:

  • Header bidding gives advertisers access to more traffic sources, while waterfall involves them in auctions organized by partner publishers.
  • Header bidding allows bidding for premium inventory, while waterfall is usually used to sell remnants.

Header Bidding Disadvantages

It's not hard to understand who is more handsome in a header bidding vs. waterfall couple. Yet it doesn't mean that RTB and header bidding, in particular, are flawless. Header script may still cause latency in ad serving. It's also slightly more complicated to implement than a simple ad tag.

Another issue is duplicate bidding, because your bids sometimes compete against each other. This happens when advertisers use several DSPs — on average, 3.9 DSPs at a time. Publishers often connect to dozens of SSPs to maximize reach. As a result, two or more different DSPs offer the same inventory. To avoid this, you can use a white-label DSP, where you unite ad campaigns in one interface and trade with custom SSPs, therefore avoiding duplicate bidding.

Client-Side vs. Server-Side Header Bidding

Both waterfall and header bidding suffer from the latency disease. Is there any cure for it?

In an older, client-side process of header bidding, it is a user's browser that sends ad requests to an SSP, which causes most of the issues mentioned in the paragraph above.

Server-side header bidding implies that the publisher's ad server sends ad requests to an SSP by itself. The page load speed may improve by 40% on average. However, most publishers experience declining profits. Server-side header bidding lacks control, transparency, and complicates user identification, as most data about an RTB auction is hidden within a server.

Client-side header bidding offers better access to data, ensures more precise cookie matching, and is recognized as an industry standard. So it's hard to say which one is clearly winning here. That's the battle between latency and transparency, and everyone in ad tech has their own priority.

Conclusion: Header Bidding vs Waterfall Comparison Table

Feature Waterfalling Header Bidding
Automation Lack of automation, constantly needs manual adjustment. Needs only one manual action, then becomes automatic.
Core Element Standard Ad Tag Header Tag
Winning Bid First bid that meets price floor The highest bid
Priority Advertisers are prioritized in a specific order. No priority, all advertisers are equal.
Process Sequential Real-time
Latency High Still persists, yet eliminated in server-side header bidding
Fill Rates Low to Medium High
Revenue Lower than could be Optimized
Optimization Manual Automatic
Competition Low High
CPM Doesn't matter much, what matters is priority. Matters a lot, the higher the better.

Header bidding is a new standard of programmatic ad serving, while waterfall is becoming more and more obscure. Still, it doesn't mean that waterfalling will irrevocably fade into oblivion. Some of the publishers that prefer direct deals will likely use it as a way to prioritize their direct deals. Overall, though, you should take a closer look at header bidding if you haven't implemented it yet.

Want to trade ads via header bidding? Get RTB bidder in Epom ad server or sign up for Epom DSP.

Rate this article

(3 ratings, average: 5 of 5)