Estimated Metrics

When a metric's definition includes language that says, "This metric is estimated", this means the metric is derived through statistical sampling or modeling, and not purely or only by aggregating or counting the unit underlying that metric. You will see this disclaimer in TikTok Ads Manager for metrics that measure actions or events where we have partial data or when we know that data might be missing. Here are a couple of techniques we use to estimate metrics:

• Sampling is the selection of a subset of data from within a larger dataset to estimate the characteristics of the entire dataset. This technique is used when it is difficult, resource consuming, or, in some cases, impossible to collect information about the entire dataset due to the size of the data. And because the sample is representative of the larger dataset, we can calculate metrics for the entire dataset with high accuracy. For example, the Reach metric we have in TikTok Ads Manager uses sampling for estimation.

• Another technique we use is modeling, where we use data from multiple sources to estimate activity that is difficult to measure directly. For example, purchase events may be difficult to track because of signal loss.

Both modeling and sampling techniques help estimate metrics with a reasonable level of accuracy and these can be used by advertisers to gauge their ad performance, and as a directional guide for business planning purposes.

Estimated Metrics include (but are not limited to):

• Reach

• Frequency

• Cost per 1000 people reached

• Average Watch Time per Person

• LIVE Unique Views

• Audience Estimation (learn more here).

• Cross-campaign Reach

• Reach by data dimension