Sometimes, Talk metrics spike to extremely high values. What causes that? How can it be fixed?
When Zendesk Talk fails to calculate a duration for any reason, it stores the UNIX timestamp instead. That leads to some calls showing a high value for one of their metrics.
Talk metrics can fail for a number of different reasons, but they all end with the same timestamp behavior. Fortunately, this situation is rare, and it will keep getting rarer as our developers improve the systems behind Talk metrics.
There is no way to backfill or replace these values. The simplest option is to exclude them from your reports. These values are always extreme outliers, so you can filter them easily.
Numeric Range Filter in Insights
Here is an example report with a huge talk time spike:
After drilling in, it's clear that one call is skewing the results:
1553521820 is the UNIX timestamp for March 25, 2019, at 13:50:20 UTC, which is when this call ended. This call's talk time metric failed, so it does not have a valid value. It should be removed from the report.
You can filter out individual calls or tickets, but that would mean drilling in to find the IDs. Instead, you can do this quickly with a numeric range filter on the report:
- For the attribute, choose Ticket Id or Call
- Call is usually better for Talk metrics, but Ticket Id may be more useful in certain reports
- For the metric, choose the one with the extreme outlier
- This example is using talk time, but this type of outlier can happen on any Talk metric
- For the range, choose "less than or equal to" and enter a huge number (like 100000000)
This filter means the report will only include tickets/calls where the Talk metric is below the huge number. It should exclude extreme outliers without disrupting the rest of the report:
This type of error does not happen often. Accounts rarely have more than one or two calls like this in their whole history. A simple filter should be all you need to keep the outliers out of your reports.
If you start seeing a lot of these values, please contact our support team to investigate further. There may be something else going on.