This topic explains how to query business data of Real-time Conversational AI in the console.
Overview
Data compass provides a multi-dimensional, visual, and real-time quality monitoring service for agents in Real-time Conversational AI. It helps you quickly understand your application's health, pinpoint issues, and resolve anomalies. This aggregated data helps you analyze quality and operations to manage your business efficiently.
Dashboard
Use the real-time data on the Dashboard for quality analysis. This data-driven approach helps you gain precise insights into your agent's performance, allowing you to continuously optimize the service experience. For example, a high average response delay indicates a slow agent response. You may need to optimize code logic or expand hardware resources. A significantly low connection success rate in specific business scenarios might suggest a problem in the business process design. This issue may require targeted adjustments.
Procedure
Log on to the IMS console and choose Intelligent Production > Real-time Conversational AI > Data Compass.
Select the agent and its corresponding client that you want to analyze, and then select a time range.
Use the Average Response Delay metric as a reference. If you notice that the latency during a specific time window far exceeds the average, you can view the details of that call for quality analysis.
On the Call Entries tab, locate the call and click Single Call Analysis.
On the Single Call Analysis page, you can analyze Stream Quality and Terminal Performance.
Metric analysis guide
Metric | Definition | Analysis guide |
Average response delay | The time between a user finishing speaking and receiving the first part of the agent's response in a conversation turn. | A high value indicates a slow agent response. This may be due to low algorithm efficiency or insufficient server resources. To resolve this issue, optimize your code logic and expand your hardware resources. |
Average connection delay | The delay from when a user initiates a call to when the call is successfully connected. | When it exceeds the normal range, it may indicate access gateway congestion, poor network quality, or issues with the session allocation algorithm. To resolve this, optimize the gateway configuration, expand network bandwidth, and readjust the session allocation strategy. This shortens user wait times and improves the interactive experience. |
Connection success rate | The number of successful connections divided by the total number of call attempts (including failures). | If the connection success rate for a specific business scenario is significantly lower than others, investigate whether the business process design is causing connection conflicts. You can then adjust the process architecture accordingly. |
Audio stuttering rate | The rate of audio interruptions experienced by users. Audio stuttering rate = Total audio stuttering duration/Total audio duration. A stutter is recorded when the terminal fails to receive audio data for over 200 milliseconds. | If the rate is higher than the industry standard, it is likely caused by low audio codec efficiency, high network packet loss, or improper buffer settings. To resolve this, upgrade the audio codec, optimize the network transport protocol to reduce packet loss, and adjust the buffer read/write strategy to ensure smooth audio streaming. |
Video stuttering rate | The rate of video interruptions experienced by users. Video stuttering rate = Total video stuttering duration/Total video duration. A stutter is recorded when the terminal fails to receive video data for over 600 milliseconds. | When the rate increases, it may be because the video resolution exceeds the agent's processing capacity, the transmission is affected by network jitter, or the video rendering module has insufficient performance. To resolve this, select a resolution that matches the agent's hardware, use adaptive bitrate technology to handle network fluctuations, and upgrade the rendering engine and optimize the rendering process to ensure stable and smooth video playback. |
Single call analysis
In addition to identifying abnormal calls on the dashboard, you can also analyze a specific call by obtaining its instance ID. The ID is returned when you call StartAIAgentInstance to initiate a call.
Procedure
Log on to the IMS console and choose Intelligent Production > Real-time Conversational AI > Data Compass. Click the Single Call Analysis tab.
Search for the instance ID to view charts and analyze the cause of a specific exception:
Stream Quality
Terminal Performance
Call Content
Events
