View a markdown version of this page

AI agent traces using Contact search and Contact details - Amazon Connect Customer
Enable AI agent trace detailsPermissions for AI agent tracesNavigate to AI agent trace detailsAI agent performance metricsAI agent trace details

AI agent traces using Contact search and Contact details

You can use AI agent traces to review, investigate, and improve your automated customer interactions. Drill down into AI agent trace details from the Connect Customer analytics dashboards, or directly through Contact search. The AI agent trace details appear on the Contact details page and include the following information:

Enable AI agent trace details

Complete the following steps to verify that AI agent trace details are enabled for your Connect Customer instance.

Note

Currently, Connect Customer doesn't support S3 buckets with Object Lock enabled.

  1. Open the Connect Customer console at https://console.aws.amazon.com/connect/.

  2. Verify that your automated interaction logs are saved to the S3 bucket you configured for call recordings. If call recordings are not yet enabled for your instance, enable them now:

    1. In the navigation pane, choose Data storage, Call recordings, Edit, Enable call recording, and then create or select your S3 bucket.

  3. In the navigation pane, choose Flows.

  4. Under the Amazon Lex Bots section, select Enable Bot Analytics and Transcripts in Amazon Connect. Choose this option to log a full transcript of the Amazon Lex portion of the customer's experience. The transcript and traces is then available for you to read on the Contact details page.

    Note

    If you previously enabled Bot Analytics and Transcripts in Amazon Connect (prior to June 5, 2026), you must disable and re-enable this setting to activate the AI agent traces feature.

  5. Under the Automated interaction logs section, select Enable Automated Interaction Logs. This enables you to view Flow details, Lex bot, and AI agent traces on the Contact details page.

Permissions for automated interaction logs

To keep customer data secure, you need to set up permissions for granular control over who can access automated interaction logs. Access is gated by the following security profile permissions:

  • Flows and Flow modules – View permissions: These permissions are required to see flow and module specific data on the automated interaction logs.

  • Analytics and Optimization - Automated interaction voice (IVR) transcripts (unredacted) permissions: These permissions are required to access logs of the IVR interaction such as keypad inputs in response to IVR prompts, transcripts of Lex interactions, AI agent traces and more.

You can drill down into AI agent trace details in two ways:

  • From the AI agent performance dashboard: Navigate to the AI agent performance widget and drill into the AI agent of your interest. This automatically redirects you to the Contact search page, pre-filtered to the AI agent and time range you selected. You see all contacts handled by that AI agent. Choose the Contact ID hyperlink to open Contact details.

  • From Contact search: Search for a contact directly through Contact search. For more information, see Search for completed and in-progress contacts.

Contact search page showing contacts handled by an AI agent.

On the Contact details page, under the Automated Interaction tab, toggle Show flow & trace details to show or hide automated interaction details for Flows, Lex bots, and AI agents.

Note

Please allow up to 30 minutes after a contact terminates for the Automated Interaction Log to become available, and the toggle to be activated.

Toggle to show flow and trace details on the Contact details page.

The following image shows an example of AI agent trace details on the Contact details page.

An example of AI agent trace details on the Contact details page.

AI agent performance metrics

The following AI agent performance metrics are available for your contact:

  • Completeness score: This metric measures the proportion of sessions where the Orchestration AI agent responses fully address all parts of customer requests rather than providing partial or incomplete answers. Value is between 0-1, where 1 indicates complete responses across all sessions.

  • Faithfulness score: This metric measures the proportion of sessions where the Orchestration AI agent responses remain faithful to the conversational context, including messages and tool call results. Value is between 0-1, where 1 indicates perfect contextual fidelity.

  • Goal success rate: This metric measures the proportion of sessions where the Orchestration AI agent successfully resolved customer issues. Value is between 0-1, where 1 indicates successful resolution across all sessions.

  • Tool utilization accuracy: This metric measures the rate of correct tool utilization by AI agents, including proper parameters and selection. Value is between 0-1, where 1 indicates perfect utilization.

Note

This set of metrics becomes available after 24 hours and is included as part of the Connect Customer AI.

AI agent trace details

Flows

  • Choose the flow and block hyperlinks to open the flow designer in a new tab, enabling you to quickly follow along with your flow.

  • Select Play to play the specific prompt within your audio recording file.

    Note

    If no audio recording is available, the Play option does not appear.

  • Quickly identify where errors have occurred, including customer timeouts or Lambda function errors.

Lex bot

  • The hyperlink redirects to the Get customer input block configured in your flow.

  • You can see where bot intents are detected and resolved, with the Action outcome status.

AI agents

  • You can choose to expand all trace details using the (+) icon, and collapse all trace details with (-) icon. You can also manually refresh your traces.

  • Choose an AI agent invocation activity to see AI agent trace details.

AI agent invocation activity showing trace details.

  • When your AI agent calls a Prompt (executed through an inference span), the AI agent is invoking an LLM through Amazon Bedrock. The trace details show model reasoning and tool call information.

  • Knowledge base citation reference is available under the Prompt (inference span) under Span details, when configured and available. To ensure you have your citations configured, see Knowledge base retrieval configuration.

  • For each tool call, you can view the Tool metadata along with the input parameters passed into the tool call.

  • Latency for each span is displayed on the right to each activity.

  • Choose the AI agent or Prompt hyperlink to navigate to the AI agent configuration page or the AI agent prompt page, respectively.

AI agent prompt and inference span trace details.
AI agent span details with knowledge base citations.

The following labels may appear in the trace:

  • ESC: Stands for Escalate. Shows you when the AI agent escalated to a human agent.

  • ERR: Stands for Error. When the span received an error status. A status description provides the error details. This can include errors such as:

    • AI barge in: The end-customer interrupted during AI reasoning or response generation, causing the model inference to stop before the AI agent completed its response.

    • Timeout: The system waited for a response or result from that operation, but it exceeded the maximum allowed time window.