How are response times calculated in the Service Level report?
This article explains how response times are calculated within the Agent Service Level Report, why in some cases the time may appear as 0, and why the platform uses the median as the main metric.
❓Frequently asked question or initial context
Why do some records show an average response time of 0?
Why do manually calculated times differ from those shown in the report?
How to correctly interpret times outside of business hours?
🤓 Explanation
Responsetime refers to the time it takes an agent to respond to a customer message once the conversation is assigned. This time can be displayed as:
First response time
Average time of all responses
However, in some cases the time appears as 0 or is not displayed. This can be due to various factors:
⚙️ Troubleshooting: Step by step
📌 Case 1: Average response times equal to 0
Reason:These conversations do not record a customer response after the agent's initial message.
🔍Frequent scenario:
These are conversations initiated from flows triggered via API (for example, automatic campaigns or integrations), in which the client does not interact. Since there is no second intervention from the client, a response cycle is not generated, therefore, the system cannot calculate a subsequent response time.
📌 Result: the report marks a response time of 0.
📌 Case 2: Doubts about times outside of business hours
Reason:The service level report uses the median and not the average as a calculation basis, which can generate differences compared to manually performed analysis.
🔍Real example:In a review carried out between March 1 and 15, a difference was observed between the manual calculation and the value of the report. This was explained because the manual calculation used the average, while Atom uses the median.
✅ Possible solutions
📌 Why is the median used and not the average?
The choice of the median responds to criteria of quality and statistical precision:
Lower impact of extreme values:If there are conversations with very high or very low response times, the average can be distorted. The median represents the central value and is not affected by these atypical cases.
Lower impact of extreme values:
If there are conversations with very high or very low response times, the average can be distorted.
The median represents the central value and is not affected by these atypical cases.
Better representation of typical behavior:In skewed data sets (for example, many fast responses and few very slow ones), the median gives a more faithful view of typical behavior.
Better representation of typical behavior:
In skewed data sets (for example, many fast responses and few very slow ones), the median gives a more faithful view of typical behavior.
Independence of distribution:It is not necessary for the data to be distributed symmetrically or normally. The median adapts well to any distribution.
Independence of the distribution:
It is not necessary for the data to be distributed symmetrically or normally. The median adapts well to any distribution.
Easy to interpret:Indicates the midpoint of the ordered times, offering a clear and consistent reference.
Easy to interpret:
Indicates the midpoint of the ordered times, offering a clear and consistent reference.
📌Do you want to use the average anyway?
You can do so by exporting the report in .xlsx or .csv and calculating the average directly in Excel or Google Sheets.
📝 Additional notes
Cases without customer interaction (such as automatic flows without response) do not generate valid response times. This is expected and does not represent a system error.
Response time is only calculated when there is a customer message after assignment.
Times outside of working hours are also included in the calculation, unless explicitly filtered by working hours.
The system only considers responses from the agent assigned at the time, ignoring previous interventions if there were reassignments.