Common data issues and anomalies

Brand Check is a complicated tool and sometimes you might find data that doesn't look right. More often than not, there's a straightforward explanation; this document will help you understand what's happening in these situations.

Not detecting the moment I searched for

This is usually because the moment is of the wrong type. For example, if the word or phrase is visible on-screen but you searched for something spoken in the audio.

To resolve this, check your search matches the moment type you're looking for.

Total Views doesn't match my platform data

In the CSV export, each row is a moment, so common operations (like summing the number of views) can lead to inaccurate data because the same viewer might be counted across multiple moments.

To resolve this, avoid performing sum operations on the data and if you need to extrapolate asset metrics always calculate an average (mean).

Fewer viewers at the end than the asset's VTR

It can be confusing when Brand Check says fewer views reached the end of the asset than the external platform reported as the VTR (View Through Rate). Again, this is caused by treating a table of moments as if it were assets.

Avoid this misinterpretation by remembering that you're looking at one row per moment and that the point at which the measurement was taken isn't always at the start or end of the moment (for example, a visual moment might measured after 1s to allow the viewer to process it).

Moments are captured at the wrong time

Brand Check looks for moments at an interval of 0.5s and any detections will be rounded to the nearest 0.5s. That means very short moments can sometimes be attributed to the wrong timecode but it will always do its best to represent the data as accurately as possible.

Numbers don't match when values are low

This mostly affects new assets or assets with extremely low views. Because Brand Check does so much in the background with analytics to ensure the data is accurate and usable, it isn't always 100% accurate when the source data is limited.

For example, the retention chart uses very complicated maths to ensure it accurately models user behaviour, but if an asset only has a handful of views it can sometimes be skewed simply due to limited data.

This issue will resolve itself when the asset gets more views and a few dozen should be enough to be extremely confident in the data.