We invested 250,000 euros in Facebook Ads: These are our 5 biggest learnings

The sneaker release platform Grailify has invested a quarter of a million euros in Facebook Ads and learned a lot in the process. I summarize the most important learnings and my tips for successful advertising on the platform in the article.

In May 2016 we launched Grailify. Our most important tool from the start was the Facebook group. This is where the building of our loyal, sneaker-mad community began. Accordingly, it was obvious to expand the community with the help of Facebook ads.

That was almost exactly seven years ago. Numerous different social media platforms have joined the Facebook group. With several million downloads, the Grailify app is now one of our most important traffic drivers.

But one thing has not changed to this day. The Facebook and Instagram ads are still one of our main advertising opportunities. With the help of almost exactly 259 different advertising campaigns, we have now exceeded the magic mark of 250,000 euros in advertising expenditure on the platform. Of course, many mistakes were made, but it is well known that you rarely learn anything without making mistakes. I would like to give you our five most important learnings today so that you don’t make the same mistakes.

The 5 most important learnings at a glance

  1. The mystery of the learning phase
  2. Facebook mag is simple
  3. Scaling is not easy
  4. Give freedom to the system
  5. Respond to advertisements

1. The mystery of the learning phase

The learning phase is a nightmare for many marketers. It’s a mysterious period in which the Facebook algorithm tries to create a perfect balance from the seemingly infinite number of variables. Not infrequently, the conversion costs shoot up to astronomical heights, so that one is inclined to deactivate the advertising campaign again just a few hours after the launch. We took some important insights from the learning phase with us:

The algorithm needs a certain freedom. Facebook itself suggests 50 “actions” within seven days when you start a new campaign. So if you advertise an app, as in our case, the algorithm should be given the opportunity to generate at least 50 downloads within seven days. How exactly did we deal with this?

  1. Set a realistic budget

In most cases, you know your desired CPO. For example, if you want to target an app download for a maximum of one euro, then the budget of the advertising campaign should be within a realistic range so that the algorithm can complete the learning phase effectively and quickly. 50 to 100 euros would probably be realistic here. This also limits the loss that you suffer in the worst case.

  1. Less is more

With the targeting and placement options, the Facebook algorithm already has many options for displaying advertisements. If you now throw numerous different creatives into play, the number of possible tests in the learning phase increases exponentially. For this reason, we always limit ourselves to a maximum of three advertisements in an advertising campaign.

Three active ads per campaign (click on the image for a larger view), © Grailify
Three active ads per campaign (click on the image for a larger view), © Grailify

This allows us to test a few different ideas without overwhelming the algorithm with an unnecessarily large number of advertisements, so that the learning phase can be completed as quickly and effectively as possible.

  1. Set and forget

This point is perhaps the most difficult. We also struggled with this for a long time in the beginning. It is not uncommon for the download costs to skyrocket for a new advertising campaign that is in the learning phase and one would like to deactivate the campaign immediately. But more often than not, prices fall just as quickly as they rose after the mysterious learning phase ended.

It’s important to realize that making manual changes to new campaigns usually just restarts the learning phase. Therefore, it is advisable to refrain from any changes in new advertising campaigns.

2. Facebook mag is simple

The greed for performance has often driven us to do unnecessary tests and learn the hard way. With our advertising campaign, we achieve average download costs of 0.60 euros for a new user? Very good! Can we also optimize the number to 0.50 euros? Usually the answer was no.

In the search for the “perfect” advertising campaign, we often found that we performed far worse with further tests than with the current campaigns. The truth is: We spent almost 100,000 euros on our ten most important advertising campaigns alone. The remaining 150,000 euros are spread over a whopping 249 campaigns.

Top ten advertising campaigns by spend (click image for larger view), © Grailify
Top ten advertising campaigns by spend (click image for larger view), © Grailify

Our rule of thumb is to have no more than three ad groups activated per campaign and no more than four creatives per ad group. This way we can keep a certain degree of clarity.

We’ve also found over the years that CPA for groups with 10+ creatives doesn’t differ significantly from ad groups with a few creatives. On the contrary, the more creatives are activated at the same time, the longer the learning phase takes and the higher the teaching costs.

3. Scaling is not easy

The concept of scaling has become a buzzword in the world of digital marketing in recent years. Everyone just wants to scale. It also sounds easy. You create a successful system and you get double the input, double the output. In reality, however, it doesn’t quite work that way.

Our experience has also shown that simply doubling the advertising budget does not immediately result in a doubling of conversions. Sure, the total volume of conversions usually increases, but usually at a significantly higher average cost. An excessive increase in the advertising budget usually ensures that the learning phase returns. Because if you increase the advertising budget of a campaign too quickly, the algorithm has to re-evaluate how many conversions it can achieve daily with the now significantly higher budget. In recent years, we have therefore mostly used one of the following two scaling options:

  1. Vertical scaling

Here we increase the campaign budget of a successful campaign by a maximum of ten percent on a given day. This ensures that the increase is just low enough that an unwanted learning phase is not triggered. Inevitably, this method will reach its limits at some point, so that, for example, the display frequency will gradually increase and the conversion costs will also increase slowly but steadily. Because of this, we tend to scale more and more horizontally.

  1. Horizontal scaling

Here, existing and profitable creatives or ad groups are duplicated and aligned for new target groups. For example, lookalike target groups, retargeting target groups, other demographic characteristics, people with different interests, etc. are conceivable. You increase your own budget, so to speak, by exploring new target groups with already successful advertisements.

4. Allow freedom for the system

In the beginning, we regularly racked our brains about defining the target group. We’ve done countless variations of different audiences with different demographics, different specific and less specific lookalikes, and countless tests beyond that.

Again and again we ran into a problem very quickly. The problem of ad frequency. If you create ultra-specific target groups with a few thousand or hundreds of thousands of people, the performance can be above average at first, but the ads will very quickly reach their limit because the same people are confronted with these ads over and over again. Due to the rapidly increasing frequency of ads, the performance drops quickly.

The solution? We give the system a lot of freedom by hardly restricting the target groups at all. We only give the system gender and age when creating new advertising campaigns; nowadays, we usually leave everything else to the Facebook algorithm. Thanks to the integrated pixels in our website and app, the algorithm has a pretty good idea of ​​the Grailify users.

In our experience, the performance does not suffer at all, on the contrary, the creatives have a longer lifespan and we save an enormous amount of time.

5. Respond to Ads

The campaign has successfully survived the learning phase, the performance is satisfactory and you can slowly but surely scale the campaign. This is how you want an advertising campaign. Gradually, however, the performance deteriorates for no apparent reason.

For a long time we just looked for the reasons in the advertising campaign. Often fiddled with the settings or the creatives in vain. But there’s one thing we’ve overlooked for far too long: ads on Facebook and Instagram are like traditional posts, meaning people react to and comment on those ads too!

Unfortunately, these comments are often negative or even completely irrelevant and below the belt. Welcome to the world of social networks. Unfortunately, we only realized late that advertisements also need moderation and that the performance of an advertising campaign can definitely be improved by reacting to positive and negative comments.


I hope you too were able to gain some new insights for your advertising campaigns from our own learnings. It is important for me to make it clear that these are our own experiences. This article is not to be understood as a blueprint for all advertising campaigns, but rather as an incentive to think about your own actions.

Source: OnlineMarketing.de by onlinemarketing.de.

*The article has been translated based on the content of OnlineMarketing.de by onlinemarketing.de. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!