All Access Pass: Dissecting Header Bidding With OpenX’s Qasim Saifee

What Really Happens When You Use Header Bidding

Header bidding (also referred to as tagless bidding) might seem as mysterious as it is exciting. Sure, it’s driving mad revenue for all your publisher friends, but what are these whispers about editing source code, latency and bizarre implementations never seen before in digital advertising? Oh dear, oh my!

Not to fear – Qasim Saifee, OpenX SVP of Monetization Platform, is here to hold your hand through a comprehensive exploration of header bidding and how it enables publisher demand partners to evaluate inventory before the impression even hits the ad server and dynamically enter competitive bids. In addition to walking through the tech basics, Saifee expounds on the benefits for publishers, implementation challenges, where private marketplaces fit and how header bidding is paving a bold future for programmatic transactions.

G: So, is there a layman’s explanation for how header bidding works?

Q: The term can be daunting. Think about it this way: Traditionally, a publisher’s ad server has access to all of a publisher’s inventory, and traditionally monetization partners only see a portion of that inventory – whatever the ad server decides to allocate to that partner based on settings the publisher has put in place. When we invented this technology in 2013 (patent pending) we wanted to give the publisher the ability to allow an exchange partner to see 100% of its inventory and compete on a dynamic basis for every impression rather than at a static price.

In a traditional tag-based implementation, a publisher will traffic a demand partner’s tag in their ad server at a fixed price, which may be an average across all impressions but certainly isn’t reflective of each individual impression’s true value. Based on that fixed price, the ad server will allocate some portion of the inventory to that tag.

In a header implementation, the publisher places a piece of code in the header of their page that calls out to our exchange and we send back a dynamic price. That price is obviously going to vary per impression – it may be $2.50 for one impression, it may be $3.90 for another. For the impression we valued at $2.50, we pass in that price into the publisher’s ad server. We have a number of tags already in the publisher’s ad server targeted at various price buckets, including one for $2.50. We’ve now dynamically told the ad server how much we’re willing to pay for that impression and can compete against all other demand that exists within the publisher’s ad server.

We may value the next impression at $3.90, and again we’ll pass that price into the publisher’s ad server. We’ll have a different tag targeted at that price point, so that impression will compete at $3.90 against all the other demand in the ad server. So you’re going to have a tag for every different price bucket — we typically do them in 10-cent increments all the way up, but that can be configured by the publisher.

G:  Do I really want my demand sources seeing all my inventory? I thought as a publisher, I’d want to play a little coy – I don’t want you to know everything I have.

Q: One of our fundamental beliefs here at OpenX is that a publisher is best served when all quality demand is competing simultaneously for all inventory. So when you silo off part of your inventory and only expose it to certain demand, you are basically restricting competition, and that ultimately drives down price, particularly in this environment of programmatic demand often seeking specific audiences.

A prime example of this is DFP and AdX without any header bidding. If a publisher is using DFP and has Dynamic Allocation turned on, AdX is getting last look at all the inventory. Within that environment, each impression is limited to AdX demand. With header bidding, a publisher can insert non-AdX demand into the system on an impression-by-impression basis, which immediately creates more competition for every impression.

We do not think it’s to a publisher’s detriment to allow competition for every impression because our demand is diverse, and you have a different set of buyers targeting different audiences and users. If you introduce all of your inventory to our demand, you’ll often find that there are pockets of inventory that you’ve been historically undervaluing.

G: Is this just for display, or are we talking video as well?

Q: Header bidding can be applied to all formats and screens, though it’s primarily a web-based implementation at this point.

G: Is changing the source code the only way to implement a tagless solution?

Q: It is not the only way, but it is generally the preferred way because it’s the most operationally efficient. Publishers place a piece of code in the header to feed the value into the ad server that is why it needs to sit outside the ad server. Typically that means sitting on the publisher’s page. The alternative is to sit inside the publisher’s ad server and have the call to our exchange happen within the ad server. However, that ends up requiring two calls to the publisher’s ad server — one call to get a price from us, and another call to pass that price back into the publisher’s ad server. The second call introduces reporting challenges for the publisher and a higher risk for latency.  

G: What are the typical obstacles for publishers implementing header bidding?

Q: Inserting the code directly on the page can be a hurdle for some publishers. Publishers will occasionally make changes to their page, which is why we work in concert with them. We test pages to make sure there is no impact from our header implementation resulting from their code changes.

Most importantly, we’ve made sure that our header solution runs asynchronously with the rest of the content on the page, so it doesn’t hold up loading the page. It doesn’t influence anything other than the ad that we’re going to serve.

After we insert the code we need to set up a number of tags within the publisher’s ad server for various price buckets. We tend to do it in small increments to ensure dynamic pricing which traditionally would create a lot of work. To make that process more efficient, we’ve written scripts that go into a publisher’s ad server and automatically set up the necessary line items on behalf of the publisher, taking that operational burden off ad ops.

G: How do you optimize header bidding performance?

Q: We have a team of yield analysts assigned to each publisher and their focus is optimizing yield on the publisher’s behalf. Think of them as an extension of the publisher’s monetization or overall yield team. We are constantly pulling levers to ensure that performance is as accretive to a publisher’s overall yield as possible.

G: Can you have Deal IDs and private exchanges run through header bidding?

Q: Absolutely. Marketers are excited by the prospect of PMP functionality against header inventory because they know they’re getting access to 100% of the publisher inventory, including inventory that historically went to a publisher’s direct-sold clients. Both marketers and publishers set up a PMP deal the exact same way they would traditionally, and we execute it through the header solution.

G: How many demand sources could you theoretically run through a header bidding setup, or would you need to set up multiple header implementations for each demand source?

Q: This is becoming a topic that’s being widely discussed across the publisher community. While we’ve seen publishers implement multiple header implementations simultaneously, there are certainly more challenges in doing that. One is the potential for increased latency as a result of the incremental partners. Another challenge is duplication of demand. We recommend that publishers implement unique, diverse sets of demand simultaneously. In fact, we offer a solution for publishers interested in managing multiple header providers that efficiently deploys and manages the code of each provider, controls latency across providers, and automates set up.

G: Is there some reason why publishers would say, “No, no, header bidding isn’t for me?”

Q: Header bidding makes less sense for publishers that are entirely sold out on a direct basis, and don’t have as strong a need for programmatic demand. Outside of that, every publisher that we’ve set live has seen overall yield increase, not only because we are becoming a more significant demand partner as we get access to all of their inventory, but also because of the pricing insights we give them.

Think of it as a price check on every impression, giving publishers a sense of what each impression is worth to the market and therefore additional insight into how to allocate it appropriately. As we’ve learned with programmatic channels, all impressions are not created equal, and there are tremendously varying prices that marketers and buyers are willing to pay for each one. Historically, the buy side holds much more information than the publisher in terms of the value of inventory. So getting that price signal is incredibly powerful and valuable for publishers.

G: This seems to be building toward something. Where is header bidding going next? What’s the bigger plan?

Q: We often contrast direct sales with what happens in the programmatic channels, where historically buyers have been given ultimate choice but no certainty. They’re able to bid on individual impressions and determine the price per impression, but they don’t know which impressions they’re going to get access to bid on. In traditional direct sales, you have the exact opposite situation. You have ultimate certainty, but you have limited choice. To a large degree, you’re at the mercy of the publisher’s ad server. Ultimately the publisher’s ad server determines which impressions you’re going to get.

Header bidding enables publishers to offer marketers both choice and certainty simultaneously. It gives marketers access to 100% of a publisher’s inventory, but it also gives marketers the ability to choose which impressions they want because they’re buying through a programmatic channel. If you layer in the sophisticated forecasting that’s required at the audience level, you can deliver programmatic guarantees.

That is one of the most powerful things you can offer to a marketer. You give them both choice and certainty, and all of a sudden their willingness to pay is higher than either direct sales or programmatic today.