Learning Phase refers to the beginning period of ad delivery when the system continuously explores new customers to help the Ad group reach the most suitable audience. This is an important part of the ad delivery life cycle, as the data during this stage is used to train the system to help better optimize delivery, and offer the best performance.
It’s important to stress that the learning phase is an experimental process.
During this phase, the CPA may fluctuate, but will become more stable as the learning data accumulates. According to historical analysis, usually an oCPM ad group can expect a stable CPA after achieving 50 conversions. The ad group will have passed the learning phase at around this time.
Therefore, the results you get during this phase are likely to be more unstable than the results you will get once TikTok Ads has had time to optimize your campaigns. So don’t panic if your ads don’t perform as well as you wished. This testing phase is pretty standard, and once TikTok Ads has collected enough data about your campaign, it will start delivering your ads more stably.
However, being more stable does not mean that the cost for each conversion will never fluctuate. As illustrated by the graph below, we expect smaller CPA fluctuations after passing the learning phase, but occasionally there will be conversions obtained at a high cost (like the red dot) due to the competitive nature of auction ads.
If there is a fluctuation in CPA on a day when there are few conversions, it's likely you will observe an abnormal daily CPA. The CPA will adjust towards your target when new conversions are obtained at a lower cost.
If your ad group is generating less the 10 conversions a day, we recommend using a longer period to track CPA. (Historical data shows that calculating the average CPA for more than 10 conversions will generally offset abnormal fluctuations.)
Based on the number of conversions your ad group generates, here's the timeframe you should use to track CPA.
Period to Track CPA
6 to 10
3 Day Average
7 Day Average
Here is an example to illustrate how it works:
Let’s say you want to create a campaign to increase installs for your new e-commerce app. You set your cost per install bid at $10 and starts running your ad. As your ad starts delivering, it enters the Learning Phase. TikTok Ads is going to try and find out who in your target audience is most likely to click on the ad and install your app.
During this phase, your cost per install may fluctuate anywhere from $5-30 as users start to install your app. As you get 50 installs, the Learning Phase passes and your cost per install becomes more stable. With your campaign now delivering stably, you are on your way to building your business and growing your app's user base.
Achieving 50 conversions is the most significant indicator of passing the learning phase.
After you pass the Learning Phase, you do not need to do anything else. The system will start delivering more stably now that it understands your ads and audience better.
If an ad group finds it difficult to obtain at least 20 conversions within the first 10 days, there is a high chance that this group will not pass the learning phase.
If you did not pass the learning phase, try optimizing the creative, broadening the target audience, or increasing the bid and trying again.
It is suggested that the advertiser does not make any adjustments which may have a negative impact to the data accumulation (lowering the bid or the budget, deleting creative or narrowing the target audience), during the learning phase.
If the delivery performance is far below expectation during the learning phase, the advertiser could try to increase the bid, optimize creative, or broaden the target audience.
If the CPA performance is far below expectation during the learning phase, it is suggested that the advertiser should not conduct any adjustment until achieving at least 20 conversions. Please be cautious that any improper adjustment during the learning phase may affect the ability of the system to explore, which could trigger further fluctuations in estimates and CPA.
Historical data analysis suggests that most ad groups which fail to pass the learning phase are unlikely to pass it without some adjustments. Advertisers are recommended to optimize the original ad group before trying again to pass the learning phase.