Webtrends aggregates a massive amount of data from Facebook ad campaigns that we’ve either run or white-labeled for other agencies. We decided to take a look at some of data that these campaigns have generated to see what trends would emerge. The result is our latest white paper, Facebook Advertising Performance, Benchmarks & Insights
Our sample group examined 11,529 campaigns serving 4.5 billion impressions that logged 2.2 million clicks. It represents approximately 11% of the Facebook advertising data we’ve collected since June 2007, as we’ve chosen to look at only the most recent snapshots. We wanted to talk about methodology in this and share insights on how large campaigns are run.
Facebook provides reports for Responder Demographics and Responder Profiles in the self-serve UI. These are in addition to the Advertising Performance reports that you may normally run to get performance break-outs at the campaign and ad level. The demographic reports break out impressions and click share by pre-defined age-range buckets, gender and geography. That’s why you see these particular age cuts in our reporting.
Whether you actually segment your ad campaigns along these lines or not, these break-outs still show in your reporting. Sophisticated advertisers use this insight to determine what segments are getting the best traction (and indirectly, the lowest cost per fan).
You can, however, get demographic performance by setting up a number of ads multiplied by these same buckets or other more granular buckets. These numbers will often align with Facebook’s responder reporting. Where we’ve broken out demographic data, we’ve primarily used responder reporting data rather than targeted segments from ad-level data. Note that Facebook provides this responder data at the campaign level, so you cannot de-aggreagate at the ad level.
In spite of this disadvantage, we believe that these stats were more appropriate to use than ad-level data from our own segments because Facebook’s eCPM algorithm will skew inventory to what generates a higher CTR when you specify CPC as your targeted metric. This means that Facebook will choose inventory on your behalf in order to get more clicks (if you’re telling them you want clicks). Facebook doesn’t currently have negation targeting, so you’re not able to specify that you don’t want app traffic or that you want to exclude particular interests.
You may notice in your own Facebook ad reports that the number of unique impressions and unique clicks is significantly lower than the served impressions and reported clicks. Facebook doesn’t have frequency capping yet, which means a certain pool of very active users may consume a large share of your inventory. This was a major problem for Google for in the first five years of AdSense, with publishers (those who have inventory to sell) often complaining that 90% of their inventories were being consumed by 10% of the users. This significantly reduced CTR — after an individual sees an ad 10 times, the next 100 impressions won’t have an impact.
Of the 4.5 billion impressions, we served 523 million unique impressions, meaning that on average, we served nine ads to each user. The ratio overlap is actually higher than what it initially appears, since a campaign could have 50 ads that theoretically served just once to a user, but it still counts as 50 unique impressions. Of the 2.2 million clicks, 2.1 million were unique, meaning that users rarely click on the same ad more than once. Facebook does have a click fraud algorithm that removes duplicate clicks and bot actions, but the logic is proprietary, so it’s not possible to gauge the exact impact.
The overall CTR of 0.050% may be higher than the Facebook average for US based traffic because of a heavy proportion of ads being targeted to fans and friends of fans. As we’ve discussed, adding social endorsements dramatically increases CTR, often by more than double. Thus, the CTR you should use as your benchmark should be based not on the average CTR for our study or your colleague’s campaign, but on the proportion of your ads that have social impressions and/or are targeted at fans.
Comparing CTRs on Facebook to general display advertising is difficult. For example, retargeting or branded ads will drive a much higher CTR than remnant inventory on any site. The size of the ad, placement, and number of ads on the page also contribute heavily to CTR. Facebook’s inventory runs in a standardized format, which makes it easier to compare one Facebook creative against another.
Timing is also key, particularly when it pertains to breaking news. When word of Michael Jackson’s death first hit, we had Facebook ads up within 10 minutes for a social search engine. The CTR on those ads were in excess of 1% during the first 12 hours, but the declination occurred swiftly as the news lost its immediacy.
With highly targeted ads, it’s not uncommon to get 0.200% within the first few hours, with ad burnout taking place within 72 hours. The more precise the target, the better the initial CTR, but a smaller audience also means that the ads burnout with fewer impressions. If there frequency capping, ads could live longer, but until it’s in place, we’ve recommended using friend-of-fan targeting and non-fan targeting to keep ads alive. The friend-of-fan targeting continues to feed in fresh users to target as new fans join. The non-fan targets ensure these ads don’t hit users who are already friends.
Several advertisers believe that blind multiplication of ads are a sophisticated approach to test ads and prevent burnout. This is akin to a job hunter blasting a resume to hundreds of employers in the belief that the law of large numbers applies. Our data demonstrates that with blind multiplication, ads do not evenly receive traffic.
In a followup post, we will analyze the distributions of popular and unpopular ads to demonstrate how heavily Facebook favors just a few ads. We’ll also show how many ads a campaign should have based on factors such as total spend, estimated size of audience and value of a fan.
For now though, please sound off in the comments with any questions you might have about the data that we’ve shared.