User acquisition through RTB channels can be challenging. It’s a data-heavy science that looks for trends and patterns that work. Without this data-driven focus, you just guess at what’s effective. If you are just starting with mobile programmatic it can be tough for you to find and convert new users. With this guide the Kayzen Customer Success Team (KCST) intends to provide you with the best practices in the context of media buying and campaign optimization which ultimately will help you to find and convince new users to install your app.
First, we start with basics on how to start a new RTB campaign on the Kayzen platform. Second, we discuss UA tactics (specifically whitelisting / blacklisting). And lastly, we go through creative optimization.
To set up and run campaigns on Kayzen, an order needs to be created first. See the following document for details on how to create an order: Orders. Additionally, it is required to have Suppression Lists: historical and dynamic (both are required):
Historical Suppression List - all device IDs who have installed your app in the last 3-12 months
Dynamic Suppression List - based on MMP integration, select All Users in Custom Audience. This will only capture new/active users from the time MMP integration took place.
Once you are done with setting up an order (and creating your suppression list audience) you can navigate to the Kayzen Dashboard to create your very first RTB campaign. For detailed instructions refer to the following document: RTB_Campaigns.
UA Tactics - Whitelisting / Blacklisting
In this section the KCST provides you with effective UA tactics which will ensure that you have a successful onboarding/start on our platform.
For illustrative purposes let us suppose you are running RTB campaigns on Kayzen to find and convert new users for your new mobile game.
Steps to follow:
1. Identify similar categories / app groups
First thing you need to do is to identify lookalike similar categories / group of gaming apps for which your target audience would show great interest. For instance, if your game falls into adventure genre, good way to start would be gaming apps in Epic category. Once you have several categories figured out you need to source bundle IDs of gaming apps in those categories.
What is App Bundle ID?
App Bundle ID is the unique identifier for an app on all Ad Exchanges (unlike App ID which is an alphanumeric ID for an app and is specific to the Ad Exchange).
You can do your research and assemble app bundle IDs for your desired categories. Alternatively, you can get in touch with Kayzen Success Team. Apart from bundle ID lists, our team will provide you with available inventory breakdown across all ad exchanges (based on the app bundle IDs).
2. Structure media inventory
Let us suppose you have 2 game categories, Epic and Adventure and 100 app bundle IDs per category. After you receive the inventory breakdown list from KCST it would look like the following:
It is always a good idea to structure your inventory: create separate campaigns per category and per ad exchange. In doing so, you will be able to observe and spot trends specific to each category and ad exchange. Such approach will maximize your UA efforts; it will enable you to get learnings fast and put them in use by optimizing your campaigns accordingly.
Thus, it is highly advised to create your RTB campaigns in the following manner:
Campaign 1: campaign_adventure_Appodeal
Campaign 2: campaign_adventure_Unity
Campaign 3: campaign_adventure_Mopub
Campaign 4: campaign_epic_Appodeal
Campaign 5: campaign_epic_Unity
Campaign 6: campaign_epic_Mopub
§ In order to take advantage of the highly sophisticated Kayzen algorithm, it is highly recommended to start CPI optimized campaigns; unless your goal revolves around branding and need to get visibility in front of broad audiences - in this case CPM optimized campaigns are recommended.
Once you get your initial learnings (usually in 2-3 weeks), you will be able to optimize your campaigns so that given the advertisement budget, available inventory and your goals you get maximum results with minimum spend.
3. Golden Rule
After running your campaigns (min. 10-14 days) long enough to derive meaningful decisions based on accumulated data, it is time to deep dive into campaign reports.
Download your campaign reports (for detailed instructions see Reports) and start accessing your placements’ performance.
Generally, the best performance indicator of your placements (CPI optimized campaigns) would be end to end conversion rate (eeCR), at the same time taking into consideration the scale. Depending on your goals you may also judge performance solely based on either CTR or CR.
eeCR = installs/impressions(%)
CR = installs/clicks (%)
CTR = clicks/impressions(%)
Steps to follow:
1. Identify the best performing placements based on your threshold eeCR.
§ What is the threshold of eeCR?
There is no given threshold of eeCR that you can benchmark. It depends on your goals and price of the inventory. For instance, if average CPM stands as 12$ and your target CPI is 6$ your eeCR threshold would be 0.2%. Beyond this, you would fail to meet your goals.
2. Group your placements in the following manner:
Best performers: eeCR is equal to or greater than your threshold eeCR.
Bad performers: eeCR is less than your threshold eeCR.
Low spend placements: eeCR not available (installs=0) with few impressions (see details below).
3. Blacklist the placements below your threshold eeCR. (Alternatively, you can start a new campaign and whitelist the placements higher than your threshold eeCR).
4. Increase your target CPI (15-20%). With doing so, you give more room to Kayzen algorithm and ensure you get max scale out of these placements without failing to meet your targets.
5. Start a new CPM optimized campaign on low spend apps and set target CPM higher (10-30%) than your average CPM for the original CPI campaign.
§ What is a low spend app?
Certain apps get low impressions and no installs, so it is not possible to judge the performance based on eeCR. For instance, based on our example (CPM 12$/CPI 6$/eeCR 0.2%), the impression threshold would be 500 impressions. In other words, we would categorize apps that get below 500 impressions and 0 install as low spend apps (500 impressions, 1 install would be also a low spend, however we would have included that in our best performers as eeCR equals 0.2%). You can adjust the impression threshold as you desire with an upper cap of 900 impressions - for instance, it would be up to you to include apps with 700 impressions and 0 installs in low spend or bad performers’ group.
6. Run a new CPM campaign and observe performance. Adjust target CPM further so that placements get enough impressions. (Alternatively, you can start 2 CPM campaigns – Group 1: high CPM, Group 2: low CPM. This way, you can adjust target CPM more efficiently so that you would be able to observe performance and at the same time avoid overspending).
7. Once you identify best performing apps, update the original CPI campaign and pause/delete the current CPM campaign(s).
4. Kayzen auto-blacklist tool
Kayzen Datascience blacklists bad performing apps once it has enough learnings (2-3 days). See full Algorithm_FAQ.
Here, we discuss the ways you can take advantage of Kayzen auto-blacklist tool (ABT) and internalize it in your campaign optimization. Generally, ABT comes into play when placements get enough impressions to conclude that its really bad performing app, however there might be cases when apps get blacklisted under 200-300 impressions (this criteria depends on CPM, target CPI and eeCR). In this case, you can start a parallel (derivative) CPM optimized campaign and whitelist auto-blacklisted apps which you think got judged prematurely. You need to check ABT status (in every 3-4 days) and keep the CPM derivative campaign in sync with the original CPI campaign. So that if auto-blacklist status of an app is removed in the CPI campaign you remove the same app in the CPM campaign.
If you believe that certain placements are definitely not working, it is better to perform manual blacklisting, so as to make sure Kayzen Datatscience is not running them again after a while. Sometimes, when auto-blacklist status expires, the algorithm starts bidding aggressively on these placements and if these placements are high-traffic loss making kind, you risk burning the daily budget in vain.
TIP #1: If winning rate of your RTB campaigns is low and you fail to get enough traffic to meet your goals, try to increase your target CPM/CPI/CPC.
TIP #2: Remember that conversion rate of cheap inventory is relatively low. Thus, sometimes it is a good idea to increase your bid value and observe; this way you will be able to spot the trends in the context of inventory price and conversion rates. In doing so, you will find the right balance between the two and optimize your campaigns accordingly.
TIP #3: Successful CPI optimized campaigns may sometimes stop delivering. Time to time the algorithm tends to shift spend/focus from one group of placements to another (within the same campaign), which may result in temporal drop in scale, especially if the other group of placements are not converting as good as previous one.
In such cases, you can whitelist placements which performed really good in the past but algorithm stopped/reduced spending properly on these placements.
i) Create a temporal CPM campaign (or CPI)
ii) Pause/delete the temporal campaign once the scale is back to normal in the original campaign.
Creative optimization plays an important part in UA. On the Kayzen platform you have an opportunity to try out different type of creatives and tailor them to your specific campaign goals.
§ On Kayzen, it is possible to run one creative type per campaign. You can organize your campaigns based on different types of creative. For instance: campaign_1_Playable, campaign_1_Video, campaign_1_Banner etc.
KCST advises you to start out with rewarded inventory: Playable or Video. The fact that rewarded ads have higher engagement rate compared to other types will provide you with a smooth start in terms of getting initial results/learnings.
§ Rewarded inventory is a type for which users need to opt-in to watch the ad - having been incentivised to do so by being offered a reward such as an additional life in a game or in-app credit.
In terms of performance between rewarded playable ad and video, playble ad tend to outperform video with higher CTR and scale (eeCR more or less the same). For instance, if you run 2 campaigns, 1 video and 1 playable (clones - except creative type), you would achieve similar eeCR on average; however, the scale would defer. You will achieve greater scale with playable compared to video ads.
For detailed information about creative specifications and guides see Creative_Specs_Guidelines.
A/B testing is a great method to optimize your creatives. You can test your ad variations to see which ad brings in more converting users. Well-planned A/B testing can make a huge difference in the effectiveness of your UA efforts. Narrowing down the most effective elements of your creatives and then combining them can make your campaigns more successful.
Steps to perform successful A/B testing:
The first thing to do when planning an A/B test is to figure out what you want to test and simultaneously you should have a clear idea of the results you’re looking for.
Once you know what you will test, make a list of all the variables you’ll test. For example, CTA button, headline, graphic mechanics (relevant for playable ads), descriptions, mottos etc.
Test one variable at a time.
Use a creative you are currently running as your control to benchmark the performance.
A/B Tests need to be run simultaneously to account for any time variations. For instance, you can’t test one variation today and another tomorrow.
Your A/B test will need to run at least 7-10 days. This is required to get the sample size necessary to determine a winner.
Keep testing regularly, the effectiveness of your creative variations can change over time.
See the following document on how to perform A/B Testing on the Kayzen platform: A/B_Testing.
Once you are done with testing and you get initial results, it is time to make sure that those results are real and not a random chance. You can use Kayzen Statistical significance calculator to calculate the statistical significance of your results.
For more details see the following document: A/B_Testing.