❓ Frequently Asked Question or Initial Context
When supervising the human management of your platform, it is very common to ask yourself the following questions:
How is the service level of agents measured?
Why do metrics sometimes differ from what I see in other reports?
How do reassignments or anticipated responses affect my quality results?
The Agents: Service Level report was designed to answer these questions. Unlike the General Report (which measures volume and global states), this report helps you evaluate the real quality of service of your team, measuring efficiency from the exact moment the agent receives the chat in their inbox.
🤓 Explanation and Key Concepts
Service Level (SLA or NDS) is the percentage of conversations that met the expected response times, compared to the total conversations assigned to an agent.
To understand this report, you need to know how Atom calculates time:
First response time: The minutes that pass between the moment of assignment and the moment the agent sends their first message.
Average response time (between messages): The average time it takes the agent to respond to each subsequent message from the customer. (Note: If the customer sends multiple messages in a row compulsively, the system will only take into account the last one to start timing).
Attention duration: The total time from when the conversation is assigned until the last message is sent (regardless of whether the agent or customer sent it).
⚙️ Troubleshooting: Step by Step
How to review your team's performance?
Go to the side menu: Reports > Agents > Service Level.
In the Main tab, you will see the summary: Assigned conversations, Percentage attended, Overall service level, Average response time, and Sales achieved.
In the Additional tabs, you can dive deeper into: Individual or group performance, Attention categories, Reassignment times, and the complete audit.
✅ Analysis of Special Cases (Reassignments and Responses)
Response times can behave specially in the following scenarios:
📌 Scenario 1: Multiple assignments (Shifts) If a customer writes at 14:00, it is assigned to Agent A, but they never respond. Then, at 08:00 the next day, the supervisor reassigns the chat to Agent B, who responds immediately.
Result: The first response time reported for Agent B will be calculated from their assignment (08:00), not from when the customer originally wrote.
📌 Scenario 2: Responses before assignment Imagine a chat is being managed by the bot. An agent reads the chat and, without formally assigning it to themselves, sends a manual template to the customer. Seconds later, the conversation is automatically assigned to the agent.
Result: The system will not count that template as the first response. The system will record as "Valid first response" the first message the agent sends after the conversation has been formally assigned to them.
💡 Recommendation to avoid statistical noise: If you want to measure the pure efficiency of a specific agent, use the top filter "Reassigned?" and select "No". This ensures that only conversations that agent managed from start to finish are shown, without inheriting the wait times caused by previous shifts.
🤔 What does "N-A" mean in Service Level?
If you see the value N-A (Not Applicable) in any table, it means the conversation was excluded from the service level calculation for one of these two reasons:
The conversation had more than one agent (there was a reassignment in the middle of the process).
You have your quality metric configured based on "Average response time", but the customer never responded again after the agent's initial greeting (with no interaction, an average cannot be calculated).
📝 Additional Notes and Discrepancies
Sales Discrepancy: If you're missing sales in this report, check if you have the "Working hours" filter active. Sales that an agent closes or classifies outside their official shift will not be shown if this filter is active.
Valid classifications: This dashboard only includes conversations that were assigned to a human and classified by a human. Automatic closures performed by the bot are not considered when auditing agent quality.
Report export: The preview will show you a maximum of 2,000 records. For complete analysis in Excel, use the download (
.csv,.xlsx, or.json), which supports up to 1,000,000 records. You can also quickly export any chart using the "Export tab as PDF" option.
Measure the true quality of your operation! 🚀
By understanding how response times are calculated and how to clean your reassignment information, you will be able to set realistic goals for your agents, reward the fastest ones, and guarantee a first-class customer service experience. ✅