How building a simple automation tool helped us fight creative fatigue
“Creative is king.”
“Creative is the most essential UA lever in 2020”.
Creative this. Creative that. We get it — creative is important. What most people don’t understand, though, is that creative is hard. Really. Damn. Hard.
How did we get here?
If you’re a fan of the Mad Men series, you’re well aware that creative has always been a popular topic among marketers. Great creative is a conversation starter and something that makes great brands what they are. On the performance marketing side, however, creative used to be one of many levers that are used to optimize campaigns. An important one, obviously, but still one of many. What changed is that the recent user acquisition trends are giving creative an even more critical role, as other levers are getting deprecated. On Google UAC (Universal App Campaigns) we’re pretty much left with bid, budget and creative, and in the case of Facebook, we still have control over which audience we’re going after. However, there are rumors that Facebook is building a UAC-like product which will remove this lever as well.
What all this means is that UA professionals need to understand creative extremely well — how it gets produced, where and how a particular creative is being served and what the performance is in each placement, channel, geo, etc. Finding successful concepts and understanding what’s working will continue to be the most significant challenge as top-performing themes keep changing pretty much every quarter:
- 2017 was the year of playable and streamer ads.
- 2018 the year of video + playable end card combos and player vs. player videos.
- 2019 will probably be remembered as the year of misleading and provoking ads. Lily’s Garden anyone?
The frustration with creative comes in when, even if you find a successful concept, it only performs well for maybe a few weeks before the performance invariably plummets. This problem is known as creative fatigue — a behavior in which creative performance declines due to a particular ad being served too many times to a specific audience. This is the problem we at N3TWORK experienced early on when we started scaling Legendary: Game of Heroes, and this inspired us to design and build tools tailor-made to minimize the impact of creative fatigue on ROAS.
Let’s make more creative!
The most common response to creative fatigue is making more creative. However, for most gaming advertisers, scaling creative production efforts is both time-intensive and costly. This is due to the fact there are only a handful of vendors who understand performance marketing creative well, and finding and vetting them takes time. Additionally, this means that you have to spend at least a few thousand dollars with each one before you’re able to get the slightest idea of their capability. High production value concepts, such as TV-quality live-action ads, can even cost hundreds of thousands of dollars. Even if you have the budget, more expensive creative doesn’t guarantee better performance, which is the most frustrating part of the process.
There are a few reasons for that:
- Creative performance of new concepts is often hit-or-miss. While working on a new ad creative, we are trying to make the most educated bet possible but, in reality, most new concepts are statistically likely to fail.
- One would assume that the production value of creative plays an important role but, quite often, it has nothing to do with the performance. It can be quite the opposite actually as sometimes the simplest ads are the best-performing ones. I keep coming back to this Clash of Clans ad from 2013 that was all over the place at the time. Most brand advertisers would never do such a thing. However, creative fatigue forces performance advertisers to test as many silly-looking concepts as possible. Check this subreddit for more examples of this.
- This doesn’t mean you should not test more expensive concepts. As your project matures, it’s likely that your advertising will become more accessible as you start going after broader audiences. The risk profile increases this way (due to the production cost) but the potential reward also goes up as your addressable market grows with more accessible messaging. MZ’s successful partnership with Arnold Schwarzenegger that resulted in two Super Bowl ads is a great example of this.
- Finally, a lot of the creatives never get tested adequately on Facebook and Google due to under-delivery, which makes it hard to calculate the ROI of each.
With all this in mind, it becomes quite clear that scaling creative production requires a slightly more strategic approach than just trying to get as many creatives out there as possible. What we have learned over time is that the key to success is to have a more balanced creative process. This process should enable us to get the most out of well-performing creatives by extending their lifetime and, at the same time, provide enough new concepts that we could test in parallel with successful ads. This is how we at NSP (N3TWORK Scale Platform) ended up building the tool internally known as Draper — a creative management tool that helps us build, deploy, and track creative performance across all major channels.
Creativity can’t be automated
Before we started productizing Draper, we had to understand and define what’s possible and how technology can help us scale creative production while improving performance and fighting creative fatigue. There are a lot of bold claims about using machine learning and artificial intelligence for building creative, but in 90% of cases, the claims are false. Because, simply put — creativity can’t be automated. There are things that technology is great for (e.g. working with large data sets) and in this particular case could help us automate the boring task of speeding things up on the creative front:
- Creative Resizing — an extremely manual and time consuming task
- Localization — uploading a CSV file with translations to programmatically generate localized variations of ads
- Creative naming — making sure that length, type, format, language placement are always passed on the click and consistent across all networks
- Automated deployment to multiple ad networks so media buyers don’t have to spend too much time moving creative between different ad accounts.
- Making decisions on which creatives to run, make or iterate on based on the data from multiple campaigns/networks/placements.
None of these sounds as cool as having ML build your creative, but each one of these can provide a pretty significant improvement to your creative performance.
Draper to the rescue!
Once we had a better understanding of what could be done, the NSP team started building Draper, our automated creative generator tool. The initial goal was to build a simple tool that would let us create multiple variations of the best performing creative concepts by making smaller changes to the original concept.
Let’s assume one of the best performing video ads is a movie trailer-style ad that features one character and its skills, a short gameplay session, and ends with an end card. If we dissect this ad into three segments — character, gameplay and end cards as shown in the picture below — Draper can easily create multiple variations of the same concept with small alterations. This allows us to generate multiple variations of the same, top-performing concept, and in an ideal case, double its lifetime.
Creating these segments is usually easy and significantly less time-consuming than building a new ad from scratch and, given that these ads get automatically pushed to different channels and ad accounts, there is very little additional work that needs to be done by media buyers. Bottom line, Draper helps quickly create variations of winning concepts and extend the lifetime of a successful creative.
Extending the lifetime of a creative concept
So what happens after these ads get pushed? Well, they breathe “new life” into the existing creative as they are different enough for algorithms to think it’s new, but similar enough to maintain the excellent performance of the original concept.
Out of a dozen variations, Facebook and Google usually focus the spend on one or two and, while there is still some decay in performance with each new variation (check the graph above), the average performance of the creative (red dotted line) is still higher than if we just ran one version of that ad (gray dotted line). Creative performance in this case is measured by IPM (installs per 1000 impressions).
Unfortunately, this trick doesn’t work forever but gives you more time to find the next successful concept.
Noob vs. Pro Case Study
The most notable example of the Draper concept that N3TWORK ran in the past is the “Player vs. Player ‘’ creative that most of the performance advertisers in gaming are quite familiar with. It’s one of those “hit-or-miss” creatives that was created because we wanted to test more “meme-looking” ads around that time. The creative took off pretty quickly and became the best performing creative we had had in a while. This incentivized us to push the creative further, which just sped up the creative fatigue process. Already a week later, we saw the original concept start underperforming, which is why we decided to leverage Draper to create dozens of variations.
The initial step was to build a creative template that would let us quickly generate variations. We started with a simple head to head template that just included two different gameplay segments and their titles. Later on, we augmented that template to include localization, richer visual effects as well as real player footage (vs. just using titles).
By the end of its lifetime, we tested around 30 different concepts of “Noob vs. Pro” creative. While the performance of the initial creative started declining after only ten days, we managed to successfully run this concept for another two months and spend a total of $1.9M just on Facebook.
Here are some other interesting metrics for this creative:
CPI reduction ~ 30%
The highest IPM creative concept for over three months
86% higher D7 ROAS compared to other creatives that we ran in the same period
63% higher ROAS to date compared to other creatives that we ran in the same period
$1.85M total spend against this creative concept
Move fast and break algos!
Draper’s approach to creative automation does have some limitations, as there is only so much you can do with this type of template. The future goal of Draper, as part of the NSP platform, is to support multiple creative templates similar to the ones that DCO (Dynamic Creative Optimization) platforms such as Celtra or Sizmek have. This will likely change the creative process on our end as we might start focusing more on designing templates and segments vs. entire ads. Once variations are created, Draper should automatically generate and deploy multivariate tests and promote the winners to the existing campaigns.
With the increased importance of creative, I believe some level of creative automation, similar to Draper, is necessary for succeeding in the hyper-competitive performance marketing space. Combined with our other automation tools such as Brewster and Markowitz, the goal of the automation layer should be to speed up the execution and help media buyers move and learn fast while, at the same time, freeing up the creative people to focus on being what they are great at — being creative.