How Blinkist Automated Their Best Performing Marketing Channels

Gessica currently leads the performance marketing team at Blinkist, a mobile learning startup that gives people the knowledge they need from nonfiction books to become their best selves. Gessica is also the host of the mobile marketing podcast Mobile Growth Nightmares.

Learn more from her Mobile Hero profile.

Automation is one of these hot topics everyone in mobile marketing is talking about. While it could seem like a pretty complicated subject (think data modelling, AI) this is something Blinkist has been doing for many years.

But why are you looking to automate something? Humans are great. They have intuition that is pretty unique and, so far, hard to replicate with a machine. We call it the “gut feeling”. However, there are many tasks where logic is the central element and machines are better at performing them than humans. I’m talking about long, laborious tasks where you apply a simple rule “IF…THEN…ELSE”. The nature of these tasks eventually leads to mistakes. It’s normal; we are only human.

In the past, the solution to this issue was easy: hire more people, smarter people. We ended up with big inefficient organizations filled with bored employees with limited growth opportunity. With automation, however, we could be more efficient in both growing a streamlined organization while applying technology to accomplish tasks that are better managed by technology.

Identify Tasks, Priorities and Time

In the past few years, Blinkist has grown rapidly and the performance team along with it. UA is one of our engines of growth. And while we have been doubling budgets year over year, the team has been growing at a slower pace. How did we do that? It’s simple. Once we realized that we were caught up in hands-on tasks with no time to dedicate to the big picture, we took the time to evaluate where channel managers were investing their time and identify tasks that could be automated. Our approach was as follows:

• Identify all the different tasks and the number of hours dedicated to each task.
• Order the tasks by time and look closely at the top 30% of the list.
• Exclude the tasks that require humans.
• Add a rank to each task that takes into consideration the time, complexity and opportunity.
• Evaluate the ordered list to identify opportunities for automation.

Thank you to Clément Halloo, Paid Social Lead at Blinkist, for providing this.

When applying this approach to performance marketing at Blinkist, we decided to focus our efforts on the top two channels of our marketing mix: Facebook and Outbrain (learn more about Outbrain in my MAU presentation).

For Outbrain, we decided to take channel management in-house. We found a tool to help manage Outbrain campaigns, even though it did not offer the bidding automation and publisher blocking features we were looking for.

With Facebook, making a decision was more difficult. There are hundreds of Facebook Marketing Partners (FMPs) available. I was already familiar with the three most popular ones. They are all excellent tools, especially for scaling ad campaigns when you’ve got limited internal resources to manage it, or you don’t like using the bulk upload of the business manager. These services usually, charge a fee based on the budget invested. Yet while evaluating these tools we realized that while these tools offer hundreds of features, we only required a few of them. And the cost was simply too high for our needs.

Decide What Features to Automate

Based on the methodology explained above, we ultimately choose one specific task to automate – campaign bidding. Bid management has always been a time-consuming task. Depending on the budget and channel, we could easily be managing bids every day. Sometimes even multiple times a day.

After we identified the task to automate on what channel, the person currently managing this task detailed their process to manage Facebook ad bids. We documented the process using a simple logic tree, applying “IF…THEN…ELSE” rules. Here is an example of how the budget automation logic could look:

Include Business Intelligence and Data Engineering

Once we defined a set of rules, we then involved our Business Intelligence (BI) and data engineering departments.

First we needed the core of the calculation. This can be done internally or using an ETL tool that will help you manipulate the data and present it in a way that makes sense to you. For example, if we are talking about bidding, store a list of ad IDs with a boolean value that defines if the bid is changing or not and the new bid value. While you can do this yourself, especially if you are using an ETL tool like Matillion or Fivetran, I recommend including your BI department.

Second, you need to send your data to the marketing provider, in our case Facebook and Outbrain. In order to do this, our data engineering team developed an API connection. If you have a lot of marketing channels to automate it may make sense to build a dispatcher – depending on the platforms it will connect with different APIs and upload the right bids.

Automation Timeline and Launch Planning

If you expect to start saving time from day zero, you are mistaken. We usually start testing automation just after sending data to the marketing provider. Our channel manager will select a couple of tests and campaigns and go over the suggested bids to make sure that everything works well. Then we will manually upload the bids.

The next step is to fully launch the automation. It’s a good idea to start with one or two campaigns and slowly add the remaining campaigns. You don’t want to automate all your campaigns at the same time in case something goes wrong.

Time Saving Advice

We made a lot of mistakes. So here are some suggestions to help you save time:

1. The logic you develop is not going to be perfect the first time around. Make sure you revisit it after launch, especially during the first month or if you have big scale budgets. Checking the competition is also a good idea, you may need to modify bids much faster.
2. Make sure that your API supports enough changes at once. Some platforms only allow you to have a certain number of changes.
3. Print and display the results of the calculation somewhere. If you use a data visualization tool, mirror the result of your table and ask the channel managers to have a look.
4. Set up an alert system that can send you a recap of all the daily changes. These alerts should notify you of the new bids that were uploaded effectively or alert you in case something failed.
5. Occasionally the data pipelines will fail. It can happen. Make sure you have people that can take over the decision making in the short term.
6. There are some edge cases where humans can provide more strategic choices, so make sure that there is a way to overrule the automation.

Results and Conclusion

We are pleased with the results of our automation efforts so far. Our channel managers can finally focus on the next iterations of our bidding machine and new things to optimize while we significantly decreased human error.

While automation can seem like a complicated topic, start by breaking it down into small steps and work together with other departments to make it happen. Our job as marketers is changing and if you want to keep up with the developments of our industry, make sure that you and your UA team understands the logic and need for automation. And maybe it’s time to take that coding class you’ve been thinking about.