On a bright September day, when a gentle breeze cools the warm afternoon sun, there’s nothing I love to do more than sit in the shade of a maple tree here in Vermont and optimize ad code. Sure, the subject may seem a little dry, but it’s actually a fascinating area of innovation and creative work. And no, I don’t actually optimize it myself under a maple tree, the product and engineers responsible for it here at CafeMedia are the best in the industry. More than 30 engineers and product people are focused solely on our ad code, which doesn’t include engineers responsible for data, UI, or our other products. The work this team does constantly delivers a great experience to readers and maximizes revenue for our publishers.
Before we dive in, it is important to understand the basics of modern programmatic ad technology. For a more comprehensive explanation, check out this article. The point is ads only show up on a site after bid requests are sent to a dozen or more partners, who in turn send them on to demand platforms used by marketers to buy ads, all in a chain of calls and responses that completes in milliseconds.
Innovation, Experiments, and the “Holdout”
Now here’s where it gets interesting. Every step of the programmatic advertising process has many options and configuration choices, each of which can significantly affect the outcome. This is because there are so many partners plugged in. They are each connected to demand sources in different ways, built with different technologies, and often set up to achieve different goals for the marketers on the other end who are trying to buy ads. Small changes to the setup can make a big difference in the results in terms of viewability, time on page, and revenue.
However, knowing which changes are likely to result in which outcomes is very difficult in such a complex and interconnected environment. That is why we built a system of continuous improvement based on data and A/B experiments. We can systematically apply statistical analysis and machine learning to a rapid pipeline of innovations and features. That way, we can prove what works, quantify by how much and endlessly tune the whole system to maximize publisher revenue.
We evaluate these experiments not just for each new release but against a “holdback” of ad code frozen as it was at points in the past, which we run on a tiny, but a statistically significant portion of traffic. That way, at any point, we can look at the performance of today’s ad code compared to a past version and measure our progress. How much has our ability to deliver RPMs, impressions, CPMs, or viewability improved? So even if the ad market’s overall RPMs go up or down, we have hard data showing how much more effective our latest ad code is than it was weeks, months or quarters ago. This is a pure measure of the code’s ability to deliver value, isolated from the macro trends of the industry and everything else.
Innovations and Results
So what are some of these innovations, and what are the results? Virtually any change to ad technology can have big effects that aren’t always easy to predict. Still, based on decades of combined experience from our team of experts, we have a good idea about the kinds of projects that are likely to improve revenue. And our tests against the holdback prove that they have.
Our overall ad code RPM performance has increased by over 13% from the beginning of the year. Video RPM performance has increased over 30%. (Keep in mind this is overall, and individual publisher results may vary. Also, this compares old ad code from January 1, 2022, to the latest release, both running today, with today’s challenging market environment.) Unfortunately, industry-wide RPMs have been soft in 2022. Still, with these innovations, publisher RPMs are up 13% from where they otherwise would be. It also means as the economy recovers and the ad market turns around, these innovations will supercharge growth and bring publisher RPMs to new heights.
The innovations generally fall into one of the following categories:
Ad format changes
Concept: These are the easiest to understand, though not necessarily the easiest to build and make successful. This category includes adding new ad formats or tweaking existing ad formats to drive better engagement from readers. Do readers tend to scroll fast past an ad or linger with it in view to drive viewability? Many of these innovations concern video, but format enhancements can be in display as well, rich media etc.
How it drives revenue: The idea here is if readers engage more, click more, watch more ads, stay on the page longer, keep the ad viewable longer, etc., revenue will go up. Of course, measuring many metrics and comparing them against control groups is important to ensure some side effects don’t eat away at the gains. But if more people engage more with an ad, in general, it will perform better because that’s what marketers are looking for.
Example: After testing several different solutions, we brought our mobile video player experience to the premium, top of page location. This premium design used by some of the most prestigious and trusted online publishers on the web is now available to independent publishers for the first time. Sites using the experience have seen an increase in video completion rates and 2–3% more overall ad revenue. It has been a big contributor to ad code RPM gains this year.
Ad performance or timing changes
Concept: “RTB” stands for “real-time bidding.” Small changes to the order, timing or speed of ad code execution and ad display can greatly impact its performance.
How it drives revenue: We have often seen gains by making the code small and faster, which can give a few more bidders the extra milliseconds they need to respond before the timeout closes the auction. Other times there are more complex effects between reader interaction and the timing of the code, which can increase viewability or other aspects of auction dynamics. When carefully tested in controlled experiments, these can reveal huge gains.
Example: We recently had a big win in video by adjusting the timing of the ad load based on readers’ scroll behavior. This was a small simple change to make, but identifying it took insight and rigorous QA testing to ensure there were no side effects. The result was a big boost to instream video revenue, contributing to the 30% lift in video RPMs against the holdback from January 1.
Demand integration/auction dynamics changes
Concept: Some techniques rely on basic auction dynamics. For example, adding more bidding partners tends to drive up “bid density”, meaning there tend to be more bids and, therefore more chances for higher bids. But there are many other tweaks that can have huge payoffs by tuning bidders that are already integrated, passing different parameters, and so on that can change the way the bidders behave individually or as part of the overall system.
How it drives revenue: Simple cases like adding a bidder with a lot of demand behind it are easy to understand, but the annals of ad tech are full of surprising and even counterintuitive changes that have big results. Because even though RTB and Prebid are standards, the particular implementations of each partner have their quirks. There have been cases where putting a parameter in ALL CAPS for a particular partner drives big gains and other unpredictable circumstances. More typically, gains are seen by passing optional data that particular bidders can use for better targeting.
Example: We made countless tweaks to the integrations with demand partners and added some new ones. Integration tweaks include adding more reader signals and ID partners, which help bidders understand the value of impressions and bid higher. Additional partners include our direct connection to the Trade Desk via our OpenPath integration. We were a partner at launch and among the first in the industry to deploy this new demand partner. We had many other successful examples of tweaking the demand integrations this year, most of which are trade secrets, so I can’t share them here. These innovations to auction dynamics and demand integration contributed almost a third of the more than 13% RPM improvement in our ad code since the beginning of the year.
We have the best and most sophisticated publisher ad code team in the world and constantly work to innovate for our publishers. The team is continuously testing new code, with fifteen to twenty ongoing tests (A/B experiments) at any given moment and averaging around 60 code releases per month. Each of these often leads to small increases in revenue; some bring large improvements. We don’t just test each change to ensure it is a winner; we also measure the cumulative lift of all the changes against a “holdback” snapshot of code from the past. This data proves the wins so far this year have been huge, more than 30% video RPM lift and more than 13% overall, compared to the code running January 1, 2022.
This year has been tough on the market, with the economy facing challenges and ad code spending in the market softening significantly. But the constant innovation and improvements in our ad code, which we are constantly measuring and proving through data and analysis, have significantly closed the gap created by the broader economy. These innovations are locked in and being compounded by new releases and improvements every day. They are setting our publishers up to do as well as possible during a tough market and are primed to reach higher RPMs than ever before as the market recovers.