Requirements and Rules of Application Promotion via Google Ads.

June 25, 2020

Welcome to the black box Google Ads Application Campaign. It is easy to set up and takes just 5 minutes: 2 minutes for campaign settings and 3 minutes to write some headlines and descriptions. Google claims that it is now super efficient and that it uses multidimensional signals to find your users etc.

But wait. There are some important things that you need to be aware of before you try it.

If your goal is in-app actions, you must be sure that you pass the right action. For example, if you pass a trial purchase with the value, this is not the same as a non-trial purchase because of the audience. You risk getting users who like trials but are not willing to buy at all. Do it if you have evidence that this in-app action lead to desirable result (e.g. purchase) or at least if you consider it as a test.
Use Firebase as the main source of your install and in-app conversions: Firebase and Google Ads work better together, sharing internal info with each other. You can use third party tracking for additional data that Google is happy to use, but these sources must be considered as secondary for Google. You should also be aware that at the moment (as of June 2020), you can only try tROAS beta bidding with Firebase.
You can send as many in-app events to the account as possible from Firebase in order to understand their value in the funnel. In other cases, it is neither good nor bad. Do it if you need to.
We don’t recommend applying MCC (manager account) conversions to other accounts, i.e., having the “Google Ads conversion account” status as “This manager”. This seems to make things easier, but it doesn’t. One of the main disadvantages is that an account will use only MCC data. Its own data won’t be taken into usage as conversions, and it limits your future experiments on sub-accounts. As for third party tracking, you can add this to your MCC and share Link ID through it, rather than making a new link in each sub-account. But for Firebase, you have to make a link with each sub-account every time. Make sure that your email has a status as a project owner at Firebase and admin access in the sub-account(s). The last thing is crucial. Even if you are an MCC administrator, you must add yourself as admin in sub-accounts! Only then can you link Firebase and sub-accounts and see the result in Firebase.
It is very, very important to add conversion of the install type (install itself or “first open”) as “Yes, Include in Conversions” in the Google Ads Conversion Settings.

Moreover, be sure to set up “Yes” only for ONE conversion of this type. Even if your goal is in-app actions, be sure to check that the install or “first open” conversion is set as “Include in Conversions”. This is crucial because tracking of install/“first open” is a signal for Google Ads that it’s working correctly; that it will bring real users. If you use in-app events from Firebase, you must set up “Yes, Include in Conversions” only for “first open” from Firebase (there is no such thing as install from Firebase, just first open) as install conversions. If you use AppsFlyer in-app events, “first open” must be from AppsFlyer etc.
Don’t use any value for install/“first open” and other non-purchase conversions. Value is only for purchases!
You can include purchases in the “Conversions” column like installs (“Yes, Include in ‘Conversions’”). It won’t ruin anything, but don’t do it just for the sake of hygiene. Instead, use Segment -> Conversions -> Conversion action to look at this data within a campaign.
Using Segment -> Network (with search partners) is also very helpful to understand how your App Campaign works across Google Networks like Search, GDN (Google Display Network) and YouTube.
There are only two options for targeting (countries and languages) and only two recommended combinations of these options to set it up right: you either take 1 country AND all languages, or you take various numbers of countries AND 1 language. Sometimes, as an exception, a huge region (several countries) with all languages can make sense too, but it must be tested.
You can also target several small countries, but you will need to join them to a region (and use 1 language!) if the campaigns separated by geo do not acquire enough data (events per day).
If your account is new, we recommend two ways to get off to a good start: either by firstly running a worldwide campaign to get a fast, low-cost and good (= big amount of data) account history; or by taking a big region that is the main target for your business with the tCPA (target cost per action) bidding. This may cost, but data will be valuable. Only then can you start campaigns by regions/countries.
If you have several installs or in-app action campaigns on different geo, and budget is the issue (say, part of it has a “Limited by budget” notification, a very important notification for this type of campaign; see paragraph № 23), merge these campaigns to allocate a budget in one campaign for best bidding in the auction. Plus, the campaign will have additional data to learn from.
These days (as of summer 2020), there is no such thing as remarketing within an App Campaign, but one day we are going to see this. You can still do remarketing using Search Campaign and app engagement ads, and you can also exclude your users from showing ads, but you have to be whitelisted for this beta. Ask your representative about this. Once again, you need Firebase for future exclusion of the audience and remarketing.
Try to start App Campaigns right before the weekend starts. Auctions are not that hot these days, so App Campaigns can get good results more easily.
You can start both install and in-app action campaigns with similar targeting, since the profiles of those who are likely to install your app and make in-app actions are different. These audiences are different, so you won’t compete with yourself.
At the start, an App Campaign learns, and it learns in no less than 7 days. It also depends on the lookback windows of conversion and the amount of data that it has acquired since the start, which is why it can be about 14 days of learning. Be ready to have your high CPA (cost per action) this week or two (maybe even higher than you want it to be). Please do not change anything except the budget if Google asks you to.
We recommend waiting another 7 days to see how it really works after learning. To sum up, the best case is 14 of learning and 7 of working.
UAC must gain at least 10 in-app conversions, which you set up as a target, or 100 installs a day or 300/3000 in a month. The keywords are “at least”. It is a minimal requirement. We recommend 15–20/day as a minimum. Having a big enough budget is very important.
But let’s consider what you need to do before setting up the budget, assuming that you know your average CPI (cost per install) and CPA by each geo. So, your tCPI (target cost per install) and tCPA for a Google Ads Campaign must be at least (!) 20% higher than your average for the same geo. If you have no account history (app events) because you have just started to promote apps, make sure that you set the tCPI and tCPA 1.5 times higher than the expected one. You need this to get your campaign to more auctions (and to have enough events, of course).
Another important thing is that your tCPI and tCPA are actually bids. The more you set these up, the more reach you have. Despite the existence of account history and our previous advice, we recommend setting up an even more radical tCPI/tCPA for every new campaign, 2–3 times higher than your average for specific geo, in order to share more profit auctions. It will cost, and (once the starting learning is over) only after every 100 conversions can you reduce your tCPI/tCPA by 20%, until the tCPA/tCPA is about 1.2 times the average CPA and CPI for the same geo. As a result, there will be a high chance that you won’t lose your place at a profit auction, having a much lower tCPA or tCPI. We find that setting up a campaign vice versa is not that effective. Starting from your actual CPC or 20% more than this, and increasing your bid by 20% will give you more traffic with a much higher tCPI/tCPA.
But don’t decrease your bids too much yet. App Campaigns really don’t like it when bids are decreased. So, the minimum bid should be near the market average (for your type of business) – ask for benchmarks from your Google representative.
Ensure that for the iOS operating system, you set up your bid 1.5 times that of Android. The iOS audience is considered to be richer, which is why everyone who starts promoting their apps wants them first and why its auction has higher bids.
To sum up, if your average CPA by geo is $10, and your account has events history, your Android bid (tCPA) must be at least $12 (20% higher), and the iOS bid should be $18 (1.5 times higher than Android). If there is no history, Android should be $15, and iOS should be $22.5. However, the following is recommended for every new campaign regardless of account history: Android: $20–30; iOS: $30–40 (2–3 times). This is not a universal rule, and if your app costs $1, having a tCPA that is 20% bigger can lead to difficulties with the reach. Try to get market averages for your type of business and use it as tCPA.
You have probably been told that the best practice is to have a daily budget at least 100 times the tCPI and at least 20 times the tCPA. Why? Because the system needs a minimum number of target conversions (10), but it’s better to have 15–20. That is why there are such scalers. Once again, the keywords are “at least”. It is a minimal requirement. In actual fact, if you have a tCPA ~ $10, you only start from $200/day, and this budget will be used to get as many conversions as possible; i.e., you just aim to get 20 conversions with this CPA (200/10 = 20). You can be sure that it will spend all the starting budget! Moreover, in reality, Google will demand up after a day or two, to 3–5 times your budget (possibly even more). In our example, it will be $1000/day or more because aiming to get 20 conversions doesn’t necessarily mean you’ll get that many, and Google knows it.
Usually, if Google initially wants you to raise the budget 2–5 times (remember, if our starting budget is $200, it will raise it from +$200 to +$1000), then, yes, it will definitely spend all the new daily budget. But when Google asks again for it to be raised 1.5–2 times (from $1000 to $1500–2000), it is highly likely to take all of this too. Sometimes, the campaign asks for a new budget, gets it and then reduces spends by 50% or more, so the daily costs for our example could be $500–600. That’s neither a bad nor a good sign. The campaign may still work fine, but it could mean that it misses something or there are changes in the auction. So, what is the point of raising it as Google wants?
Well, if you don’t raise it within a week of the demand, Google Ads will automatically lower your bid, trying to get as many users as it can. The consequence of this is that your campaign can decrease spend dramatically, especially if it doesn’t have much historical data. It may also affect the number of conversions, which is crucial for this type of campaign, making it work ineffectively. The only way to fix such cases is to start again.
Restarting, by which I mean copy-pasting existing campaigns to make a new one, is a problem because it will have to be learned once again. That means a very high CPA at start and so on. Again.
Another thing to mention is the ad assets. You can add up to 5 headlines, 5 descriptions, 20 images, 20 videos and 20 HTML5. That means a Search Network with 5 x 5 = 25 combinations and a Display of 5 x 5 x 20 = 500. Let’s take the last one. To learn what works and what does not, UAC must show each of it a comparable number of times, plus have enough data for statistical inference, and this is one of the reasons why the bid-budget ratio is important and why budgets need to be scaled so dramatically: if your tCPA is $10, your ad asset has 500 combinations; it must have a minimum budget of 10 x 500 = $5000 in order to test all variations, or you will have a situation where one set of combinations has more data and another less. You can think of CPA as a mini-budget for each combination in this case. If your daily budget is small too, Google will have a hard time getting useful information. Yes, not all managers from Google will tell you that your starting budget must be 50 times bigger than the tCPA because it’s hard to sell. Our advice is that you should either be ready to set a big budget at the start or not use too many creatives.
If you want to show a video ad on YouTube, your bid must be big enough, and the video should be at least 15 seconds (landscape format). Those that are less than 15 seconds, with a portrait or square format, are more likely to be shown in the GDN. YouTube and GDN will still do their best to work with all formats.
Another thing to note about creative content is that you should use disclaimers if you promote something that doesn’t guarantee specific results; otherwise, you’ll get misleading content and Google won’t show you exactly where. Furthermore, only one of your descriptions can include an exclamation point, and it can use it only once.
If you want to scale or lower a bid and make budget adjustments to ensure campaign stability, you must restrict this to no more than 20% per day and only if there are 100 conversions after it; once you did it, you have to wait until it makes this number of conversions to scale or reduce once again.
Sometimes, campaigns start but just fade out, and they significantly reduce the daily costs, spending only half of the usual (not necessarily daily) budget. There are a lot of reasons for this: your creatives have faded out; limits set by policies; a change of network/auction or the start of a new one; a bug or new competitors coming onto the market. Our advice is to get help from your manager asap. If you don’t have one, try to increase the bid by 20%, but no more. The only thing to remember is that this type of campaign is tendentious. If everything is good, it will make itself even better; if it is bad, it will get worse. So, if increasing the bid doesn’t work, your last chance to reanimate your campaign is to double your bid. It’s risky, but restarting is worse. You should only restart if this doesn’t work (and do so by copy-pasting).
If you want to test new creatives, it is better to make another ad group. This will take all the performance data from the campaign.
If your campaign is all right, but you want to have more reach to improve the outcome, we recommend firstly testing another ad group with new creatives in order to get a higher CTR (click-through rate). It’s well known that the higher the CTR, the lower your bid will be due to the quality score. This reduces the actual CPA if the CR is the same. Your second step is to increase the bid to make a higher target CPA. Repeat the first step etc. With this sequence, your goal is to hold onto a new CPA that is close to what you had when you started increasing but with a much higher bid.
If your goal is a purchase (and especially if this doesn’t happen that often), test different in-app actions as optimization, not only the purchase itself. Use action that is at least correlated with the purchase or has a high probability of leading to it. This may be registration or the n-th session etc.
Conclusion: Be sure to set everything up correctly before you start, i.e., the right in-app event/conversion, good creatives (that look at least attractive) and a bid budget that is cost-effective but won’t ruin your marketing budget. The most important thing is to be sure that you are ready. You must be cold-blooded. You have to be mentally prepared to spend a lot and to wait at least 14 days (a minimum of 7 days for learning, plus another 7 to get results). Fasten your seatbelt and push the start button. Remember, if you are not ready to increase the budget when Google asks you to, or you stop it before the 7 days of learning or right afterwards (to avoid suffering a high CPA), you are definitely making a mistake and wasting money that you won’t get back. Increase the budget and wait. This is more likely to yield better results than stopping it. Yes, there is no guarantee, but this is the nature this type of campaign.