# Amazon Connect Contact Lens

Canonical: https://warmtransfer.net/knowledge/amazon-connect-contact-lens

Last verified: 2026-09-30

AWS documentation states that Amazon Connect Customer is the current name for the product previously called Amazon Connect[^30]. Conversational analytics provides contact center analytics and quality management capabilities, including conversation analysis, performance evaluations, agent screen recording, contact search, and monitoring of live and recorded conversations[^51].

## Platform capabilities and architecture

Conversational analytics is built on Amazon Bedrock, and AWS implements automated abuse detection for the service[^2]. The system analyzes interactions between customers and agents, or customers and conversational AI, across voice, chat, and email channels[^4]. 

Service endpoints for conversational analytics retain the legacy naming convention, using the `contact-lens` hostname prefix (for example, `contact-lens.us-east-1.amazonaws.com`)[^11]. Real-time analysis can be retrieved through 2 polling APIs: ListRealtimeContactAnalysisSegments for voice contacts and ListRealtimeContactAnalysisSegmentsV2 for chat contacts[^32]. To avoid API rate limits, streaming through Amazon Kinesis Data Streams is available as an alternative[^32].

Default service quotas are established at the account level and can be adjusted:
- 300 concurrent real-time calls with analytics[^8]
- 200 concurrent post-call analytics jobs[^8]
- 200 concurrent chat analytics jobs[^8]
- 10 concurrent post-contact agent conversation summary jobs, shared across voice and chat[^56]

## Flow configuration and channel behavior

Enabling conversational analytics requires 2 ordered steps: first enabling the feature on the instance, then placing and configuring an analytics block in a flow for voice, chat, email, or a combination[^10]. In new flows, the Set recording, analytics and processing behavior block replaces the Set recording and analytics behavior block, though existing flows continue to support the earlier block[^3]. The earlier block supports only voice and chat contacts, sending email and task contacts down its Error branch[^24].

Voice analytics requires call recording for both the agent and the customer[^34]. When post-call analytics is configured for voice, analysis runs after both the conversation and After Contact Work end; AWS states this mode yields the highest transcription accuracy[^25]. Selecting real-time voice analytics delivers both real-time insights during the call and post-call analytics once the conversation and After Contact Work conclude[^33]. Conversational analytics supports calls with up to 2 participants; AWS notes that quality degrades with additional participants and recommends disabling the feature on multi-party or third-party calls[^60].

When chat analytics is enabled, the system provides both real-time and post-chat analytics without a separate post-chat-only enablement option[^6]. Real-time analysis begins once any block enables it, and subsequent blocks in the flow cannot disable real-time chat settings[^7]. Every processed chat message counts equally for billing once conversational analytics is active, even if specific features do not run on every message[^5].

Email analytics categorizes contacts, redacts sensitive data from transcripts, and generates contact summaries[^9]. Because email is asynchronous, it does not support real-time analytics or sentiment scores[^9].

For voice calls, analytics settings are overwritten by each successive Set recording and analytics behavior block; if a later block omits post-call analytics, no post-call analytics is generated[^1]. If a contact transfers to another agent or queue, a new contact ID and contact record are created, requiring another block with analytics enabled to maintain data collection[^59]. Queue-to-queue transfers copy the conversational analytics configuration over to the transferred contact automatically[^31].

## Sentiment analysis and post-contact summaries

Sentiment analysis is active by default whenever speech or chat analytics is enabled in a supported language, and administrators can disable it across agents and customers by unchecking Enable Sentiment Analysis in the flow block[^52]. Each speaker turn is categorized as positive, negative, or neutral[^53]. Call periods receive a score from -5 to +5 derived from sentiment streaks and frequency, and the contact's overall score is the mathematical average of those period scores[^53].

Generative AI post-contact summaries produce at most 1 summary per contact, and some contacts receive none[^55]. To display these summaries on the Contact Control Panel (CCP) for voice contacts, flows must enable recording for both agent and customer, enable speech analytics using the Real-time and post-call analytics option, and select Post-contact summary under Generative AI capabilities[^54]. If a summary fails to generate, the output file includes 1 of the reason codes `QUOTA_EXCEEDED`, `INSUFFICIENT_CONVERSATION_CONTENT`, `INVALID_ANALYSIS_CONFIGURATION`, `FAILED_SAFETY_GUIDELINES`, or `INTERNAL_ERROR`[^58]. `INSUFFICIENT_CONVERSATION_CONTENT` occurs when any participant provides fewer than 1 voice utterance or 1 plain/markdown text chat message[^58]. Post-contact summaries do not support granular redaction; when granular redaction is configured, the summary replaces all detected PII with `[PII]`[^57].

## Sensitive data redaction

When default redaction is enabled, conversational analytics identifies personally identifiable information, replacing it with `[PII]` in the transcript; administrators can also select specific PII entities to target[^36]. Redaction is applied after the interaction concludes: when the call disconnects for voice, or when the contact ends for email[^44]. Redaction operates on post-call and chat analytics, but is unsupported for real-time call analytics[^41]. AWS cautions that redaction may not capture all sensitive data, recommends reviewing output, and states that the feature does not fulfill statutory de-identification standards such as HIPAA[^40].

Enabling redaction produces a redacted analysis file by default, produces an unredacted analyzed file only if both redacted and original transcripts are configured, and generates a redacted voice recording where sensitive audio is muted with silence[^38]. The original analyzed file is the sole location where complete voice conversation data persists; deleting it eliminates all records of the redacted text[^42]. All raw, redacted, and unredacted assets are stored in instance Amazon S3 buckets under `Analysis/Voice`, `Analysis/Chat`, and `Analysis/Email` directories, accessible through the Amazon S3 console[^43].

Redaction behavior can be assigned dynamically per contact using a Set contact attributes block or Lambda function to pass the case-sensitive `redaction_option` attribute (`None`, `RedactedOnly`, or `RedactedAndOriginal`) along with a supported language attribute[^35]. When a user profile holds permissions for both redacted and unredacted viewing, the administration console displays redacted content if redaction was active on the flow and unredacted content otherwise; both versions are never displayed concurrently in the web UI[^37]. Redaction is supported in English (en-AU, en-IN, en-IE, en-NZ, en-AB, en-ZA, en-GB, en-US, en-WL), French (fr-CA, fr-FR), German (de-DE), Italian (it-IT), Portuguese (pt-BR, pt-PT), and Spanish (es-ES, es-US), but is unsupported in ja-JP, ko-KR, zh-CN, de-CH, and hi-IN despite post-call analytics availability in those locales[^39].

## Rules and categorization

Conversational analytics rules evaluate interactions across 5 analysis event triggers: post-call analysis, real-time analysis, post-chat analysis, real-time chat analysis, or email analysis[^50]. A conversational analytics rule can match contacts by 4 condition types: Words or Phrases - Exact Match, Words or Phrases Pattern Match, Natural Language - Semantic Match, and Words or Phrases - Semantic Match[^47].

Natural Language - Semantic Match conditions use generative AI, cannot run on real-time analysis events, and require administrators to have the Rules - Generative AI permission[^48]. When conditions match, rules can categorize the contact, generate an Amazon EventBridge event, assign a task, generate a case, transmit email notifications, or submit an automated evaluation[^46]. Rules apply solely to interactions occurring after rule creation and cannot retroactively evaluate historical conversation records[^49].

## Performance evaluations

Quality evaluations require designating or generating an Amazon S3 bucket within the AWS console under Data storage, Content evaluations, to hold and export evaluation records[^22]. Questions within an evaluation form can be automated across 3 methods: rules-based contact categories (single- and multiple-selection questions), generative AI (single-selection and text field questions), and contact metrics such as longest hold duration or sentiment scores (numeric questions)[^12].

Fully automated evaluation requires, in order: automation set up on every question of the form, the Enable fully automated submission of evaluations toggle turned on before activation, activation of the form, and then a rule that submits an automated evaluation and selects which contacts are evaluated[^14].

Once a new form version is activated, evaluators cannot use the previous form version for new evaluations, though earlier submitted evaluations retain their original form version[^13]. Automated evaluations fail if a manual evaluation has already been submitted for that contact, reporting the failure in Amazon CloudWatch[^21]. Each contact is capped at a maximum of 10 combined manual and automated evaluations, a quota that cannot be adjusted[^20].

AWS notes that generative AI evaluation scoring is not 100% accurate, advising manual reviews and retaining manual evaluation processes[^15]. When generative AI cannot resolve an appropriate option, it marks non-optional questions as Not Applicable, removing that question from score calculations[^15]. Generative AI evaluation answers rely entirely on the conversation transcript paired with the question instructions; AWS notes the system cannot assess agent screen recordings, inspect external systems, cross-reference multiple contacts, or detect vocal tone[^19].

## See also

- [Amazon Connect contact flows and routing profiles](https://warmtransfer.net/knowledge/amazon-connect-flows)
- [Amazon Connect pricing](https://warmtransfer.net/knowledge/amazon-connect-pricing)
- [Testing and evaluating AI virtual agents](https://warmtransfer.net/knowledge/ai-agent-evaluation)

## Applicability

Applies to: Amazon Web Services Amazon Connect Customer. Deployments: multi-tenant. Sources checked 2026-09-30. Standard performance evaluations are available across all 11 listed AWS Regions, but generative AI-powered performance evaluations are unavailable in Africa (Cape Town), Asia Pacific (Seoul), and AWS GovCloud (US-West)[^18]. AWS GovCloud (US-West) does not support generative AI-powered contact categorization, post-contact summaries, or information extraction[^23]. The Africa (Cape Town) Region supports post-contact summaries, but excludes generative categorization, information extraction, and theme detection[^23].

## What remains uncertain

Whether performance evaluations or screen recording have specific per-agent or per-minute pricing rates is not covered by the sources below. Whether existing external integrations and bookmarks referencing `contact-lens` continue to resolve normally is not covered by the sources below. Whether manual evaluations support voice, chat, email, and task contacts while automated evaluations are restricted to voice and chat interactions analyzed by conversational analytics is not covered by the sources below. The exact rate limits and burst thresholds for the ListRealtimeContactAnalysisSegments and ListRealtimeContactAnalysisSegmentsV2 polling APIs are not covered by the sources below.

## Sources

[^1]: Analytics settings are overwritten by each subsequent Set recording and analytics behavior block in a flow; if a later block has post-call speech analytics unselected, no post-call analytics is produced for the call. Source: [Flow block in Connect Customer: Set recording and analytics behavior](https://docs.aws.amazon.com/connect/latest/adminguide/set-recording-behavior.html), section 'Configuration tips', first bullet Note and 'For calls' bullet. Checked 2026-09-30.
[^2]: AWS states that conversational analytics is built on Amazon Bedrock and that AWS implements automated abuse detection for it. Source: [Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/contact-lens.html), Note 'Powered by Amazon Bedrock' at top of page. Checked 2026-09-30.
[^3]: For new flows, the Set recording, analytics and processing behavior block replaces the Set recording and analytics behavior block; the earlier block remains supported in existing flows. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), Note following the first image, before the Topics list. Checked 2026-09-30.
[^4]: Conversational analytics analyzes conversations between customers and agents, or customers and conversational AI, across voice, chat and email. Source: [Analyze conversations using conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/analyze-conversations.html), opening paragraph. Checked 2026-09-30.
[^5]: When conversational analytics is enabled on a chat contact, each processed chat message is counted for billing in the same way, even if not every feature is applied to every message. Source: [Analyze conversations using conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/analyze-conversations.html), 'Chat analytics support', third bullet. Checked 2026-09-30.
[^6]: Enabling chat analytics delivers both real-time and post-chat analytics; there is no separate post-chat-only choice in the enablement step. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Enable chat analytics', step 1 Note. Checked 2026-09-30.
[^7]: For chat, real-time analysis starts as soon as any block in the flow enables it, and no block later in the flow disables the real-time chat settings. Source: [Flow block in Connect Customer: Set recording and analytics behavior](https://docs.aws.amazon.com/connect/latest/adminguide/set-recording-behavior.html), section 'Configuration tips', 'For chat' bullet. Checked 2026-09-30.
[^8]: Default conversational analytics quotas, set at the account level and adjustable, are 300 concurrent real-time calls with analytics, 200 concurrent post-call analytics jobs and 200 concurrent chat analytics jobs. Source: [Connect Customer service quotas](https://docs.aws.amazon.com/connect/latest/adminguide/amazon-connect-service-limits.html), section 'Conversational analytics service quotas' table, first three rows. Checked 2026-09-30.
[^9]: Email analytics categorizes email contacts, redacts sensitive data from email transcripts and generates contact summaries; because email is asynchronous there are no real-time analytics or sentiment scores for email. Source: [Analyze conversations using conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/analyze-conversations.html), 'Email analytics support' bullets and section 'Sample Contact details page for email analytics'. Checked 2026-09-30.
[^10]: Enabling conversational analytics takes two steps in order: first enable it on the instance, then add a Set recording and analytics behavior block to a flow and configure it for voice, chat, email or a combination. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), numbered list at top of page; section 'Enable conversational analytics for your Connect Customer instance'. Checked 2026-09-30.
[^11]: The service endpoints for conversational analytics still use the contact-lens hostname prefix, for example contact-lens.us-east-1.amazonaws.com. Source: [Availability of Connect Customer features by Region](https://docs.aws.amazon.com/connect/latest/adminguide/regions.html), section 'Conversational analytics availability by Region' table, Endpoint column. Checked 2026-09-30.
[^12]: Evaluation questions can be automated three ways: contact categories defined by rules (single and multiple selection questions), generative AI (single selection and text field questions), and metrics such as longest hold time or sentiment score (numeric questions). Source: [Create an evaluation form in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/create-evaluation-forms.html), section 'Step 6: Enable automated evaluations', list 'three ways of automating evaluations'. Checked 2026-09-30.
[^13]: After a new version of an evaluation form is activated, evaluators can no longer start new evaluations with the previous version, though completed evaluations remain viewable with the form version they were based on. Source: [Create an evaluation form in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/create-evaluation-forms.html), section 'Step 10: Activate an evaluation form'. Checked 2026-09-30.
[^14]: Fully automated evaluation requires, in order: automation set up on every question of the form, the Enable fully automated submission of evaluations toggle turned on before activation, activation of the form, and then a rule that submits an automated evaluation and selects which contacts are evaluated. Source: [Create an evaluation form in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/create-evaluation-forms.html), section 'Step 6: Enable automated evaluations', procedure 'To automatically fill and submit evaluations'. Checked 2026-09-30.
[^15]: AWS states that AI-generated evaluations are not 100% accurate, recommends reviewing a sample and keeping a manual evaluation process, and notes that if the AI cannot identify an appropriate answer option it selects Not Applicable even on a non-optional question, which excludes that question from scoring. Source: [Evaluate agent performance in Connect Customer using generative AI](https://docs.aws.amazon.com/connect/latest/adminguide/generative-ai-performance-evaluations.html), Note after 'Process to automate evaluations using generative AI' list; paragraph 'AI answer mapping and scoring'. Checked 2026-09-30.
[^16]: The generative AI evaluations guide page states that Ask AI recommendations are available for up to 10 questions per contact and that automatically submitted evaluations can use generative AI to answer up to 10 questions per contact (disputed). Source: [Evaluate agent performance in Connect Customer using generative AI](https://docs.aws.amazon.com/connect/latest/adminguide/generative-ai-performance-evaluations.html), sections 'Use Ask AI to get generative AI answer recommendations' step 2b and 'Set up automated evaluations using generative AI on the evaluation form'. Checked 2026-09-30.
[^17]: The service quotas page lists default quotas of 30 evaluation questions answered with AI assistance on a single contact when manually initiated and 15 evaluation questions on a single contact completed automatically by generative AI, both adjustable through a support ticket (disputed). Source: [Connect Customer service quotas](https://docs.aws.amazon.com/connect/latest/adminguide/amazon-connect-service-limits.html), section 'Conversational analytics service quotas' table, rows on evaluation questions, and asterisk footnote. Checked 2026-09-30.
[^18]: The Region feature table shows generative AI-powered performance evaluations as unavailable in Africa (Cape Town), Asia Pacific (Seoul) and AWS GovCloud (US-West), while standard performance evaluations are available in all eleven listed Regions. Source: [Availability of Connect Customer features by Region](https://docs.aws.amazon.com/connect/latest/adminguide/regions.html), section 'Conversational analytics features by Region' table, columns 'Performance evaluations' and 'Generative AI-powered performance evaluations'. Checked 2026-09-30.
[^19]: Generative AI evaluation answers are derived from the conversation transcript using the question title and instructions to evaluators; AWS says the AI cannot analyze screen recordings, access internal or third-party systems, evaluate across multiple contacts, or determine tone of voice. Source: [Evaluate agent performance in Connect Customer using generative AI](https://docs.aws.amazon.com/connect/latest/adminguide/generative-ai-performance-evaluations.html), intro paragraph 3; section 'Selecting questions to be answered by generative AI', Don'ts. Checked 2026-09-30.
[^20]: The service quotas page lists a maximum of 10 evaluations, manual and automated combined, per contact, and marks this quota as not adjustable. Source: [Connect Customer service quotas](https://docs.aws.amazon.com/connect/latest/adminguide/amazon-connect-service-limits.html), section 'Conversational analytics service quotas' table, last row. Checked 2026-09-30.
[^21]: An automated evaluation cannot override a manually submitted evaluation; if an evaluation already exists the automated evaluation fails for that contact and the failure is visible to administrators in CloudWatch. Source: [Create a rule in conversational analytics that submits an automated evaluation](https://docs.aws.amazon.com/connect/latest/adminguide/contact-lens-rules-submit-automated-evaluation.html), section 'Frequently Asked Questions (FAQ)', question 1. Checked 2026-09-30.
[^22]: Evaluation capabilities are enabled per instance in the console under Data storage, Content evaluations, where an S3 bucket is created or chosen for storing and exporting evaluations. Source: [Evaluate agent and self-service interaction performance in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/evaluations.html), Tip box, 'IT administrators' paragraph. Checked 2026-09-30.
[^23]: In AWS GovCloud (US-West) the Region feature table shows no generative AI-powered contact categorization, post-contact summaries or information extraction; in Africa (Cape Town) post-contact summaries are available but generative categorization, information extraction and theme detection are not. Source: [Availability of Connect Customer features by Region](https://docs.aws.amazon.com/connect/latest/adminguide/regions.html), section 'Conversational analytics features by Region' table, rows Africa (Cape Town) and AWS GovCloud (US-West). Checked 2026-09-30.
[^24]: The earlier Set recording and analytics behavior block supports voice and chat contacts; task and email contacts take its Error branch. Source: [Flow block in Connect Customer: Set recording and analytics behavior](https://docs.aws.amazon.com/connect/latest/adminguide/set-recording-behavior.html), section 'Contact types' table. Checked 2026-09-30.
[^25]: With the post-call analytics option, the call recording is analyzed after the conversation and After Contact Work are complete, and AWS says this option provides the best transcription accuracy. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Enable call recording and speech analytics', step 4a. Checked 2026-09-30.
[^26]: The pricing appendix says some customers might still be on Customer Basic, which offers a subset of AI capabilities that excludes, among other things, conversational analytics for email and performance evaluations of self-service interactions. Source: [Amazon Connect Customer Pricing Appendix](https://aws.amazon.com/products/connect/customer/pricing/appendix/), Customer Basic paragraph beginning 'Some customers might still be on Customer Basic'. Checked 2026-09-30.
[^27]: Read together, the pricing page and appendix suggest that conversational analytics is bundled into the standard per-channel price for Amazon Connect Customer and billed as a separate line item only for customers on Customer Basic (inferred). Source: [Amazon Connect Customer Pricing](https://aws.amazon.com/connect/pricing/), communication channel price table and capability lists. Checked 2026-09-30.
[^28]: The Amazon Connect Customer pricing page lists channel prices of $0.038 per voice minute (standard telephony rates apply), $0.010 per chat message, $0.080 per email and $0.014 per SMS or third-party message, and lists conversational analytics, performance evaluations and screen recording among capabilities without a separate price on that page. Source: [Amazon Connect Customer Pricing](https://aws.amazon.com/connect/pricing/), communication channel price table near top of page; capability lists 'Unlock intelligence & automation' and 'Empower your workforce'. Checked 2026-09-30.
[^29]: The Amazon Connect Customer pricing appendix lists conversational analytics for voice at $0.015 per minute for the first 5 million minutes per month and $0.0125 per minute beyond that, and conversational analytics for chat and messaging at $0.0015 per message. Source: [Amazon Connect Customer Pricing Appendix](https://aws.amazon.com/products/connect/customer/pricing/appendix/), section 'Analytics, insights, and optimization', subsection 'Conversational analytics'. Checked 2026-09-30.
[^30]: AWS documentation states that Amazon Connect Customer is the current name for the product previously called Amazon Connect. Source: [What is Connect Customer?](https://docs.aws.amazon.com/connect/latest/adminguide/what-is-amazon-connect.html), Note box at top of page 'What is Connect Customer?'. Checked 2026-09-30.
[^31]: For queue-to-queue transfers, the conversational analytics configuration is copied to the transferred contact. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Important things to know', Note under first bullet. Checked 2026-09-30.
[^32]: Two polling APIs return real-time analysis: ListRealtimeContactAnalysisSegments for voice contacts and ListRealtimeContactAnalysisSegmentsV2 for chat contacts; a streaming alternative through Amazon Kinesis Data Streams avoids their rate limits. Source: [Use conversational analytics APIs for chat analytics](https://docs.aws.amazon.com/connect/latest/adminguide/contact-lens-api.html), opening bulleted list and following paragraph. Checked 2026-09-30.
[^33]: Choosing real-time analytics for calls provides both real-time insights during the call and post-call analytics after the conversation has ended and After Contact Work is complete. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Enable call recording and speech analytics', step 4b. Checked 2026-09-30.
[^34]: Both agent and customer call recordings are required to use conversational analytics for voice contacts. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Enable call recording and speech analytics', step 2. Checked 2026-09-30.
[^35]: Redaction can be set dynamically per contact with the contact attribute redaction_option, whose case-sensitive values are None, RedactedOnly and RedactedAndOriginal, together with a language attribute, set either by a Set contact attributes block or by a Lambda function. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Dynamically enable redaction based on the customer's language'. Checked 2026-09-30.
[^36]: With default redaction settings, conversational analytics redacts all personally identifiable information it identifies and replaces it with [PII] in the transcript; alternatively an administrator can choose specific PII entities to redact. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Enable redaction of sensitive data' and subsection 'Select PII entities to redact'. Checked 2026-09-30.
[^37]: A user holding both redacted and unredacted access permissions sees redacted content when redaction is enabled on the flow and unredacted content when it is not; the admin website does not expose both versions of a conversation at once. Source: [Assign permissions to use conversational analytics in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/permissions-for-contact-lens.html), section 'Conversational analytics permissions', Important box. Checked 2026-09-30.
[^38]: When redaction is enabled, a redacted analysis file is generated by default, an original analyzed file is generated only when the option to get redacted and original transcripts is chosen, and for voice contacts a redacted audio file is produced in which sensitive data is replaced by silence. Source: [Use sensitive data redaction to protect customer privacy using conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/sensitive-data-redaction.html), section 'About redacted files', list beginning 'When redaction is enabled'. Checked 2026-09-30.
[^39]: In the AI features language table, the Redaction column is checked for the English locales (en-AU, en-IN, en-IE, en-NZ, en-AB, en-ZA, en-GB, en-US, en-WL), fr-CA, fr-FR, de-DE, it-IT, pt-BR, pt-PT, es-ES and es-US, and is not checked for ja-JP, ko-KR, zh-CN, de-CH or hi-IN although those locales have post-call analytics. Source: [Languages supported by Connect Customer features](https://docs.aws.amazon.com/connect/latest/adminguide/supported-languages.html), section 'AI features', table column 'Redaction'. Checked 2026-09-30.
[^40]: AWS states that the redaction feature may not identify and remove all sensitive data, recommends reviewing redacted output, and states it does not meet the requirements for de-identification under medical privacy laws such as HIPAA. Source: [Use sensitive data redaction to protect customer privacy using conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/sensitive-data-redaction.html), Important box in the introduction. Checked 2026-09-30.
[^41]: A footnote to the AI features language table states that redaction is supported only for post-call analytics and chat analytics and is not supported for real-time call analytics. Source: [Languages supported by Connect Customer features](https://docs.aws.amazon.com/connect/latest/adminguide/supported-languages.html), section 'AI features', double-asterisk footnote beneath the table. Checked 2026-09-30.
[^42]: For voice contacts, the original analyzed file is the only place where the complete conversation is stored; if it is deleted there is no record of the sensitive data that was redacted. Source: [Use sensitive data redaction to protect customer privacy using conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/sensitive-data-redaction.html), section 'About redacted files', Important box. Checked 2026-09-30.
[^43]: Redacted voice, chat and email files are stored in the instance's Amazon S3 buckets under Analysis/Voice, Analysis/Chat and Analysis/Email paths respectively, and all redacted, unredacted and raw files are reachable through the Amazon S3 console. Source: [Use sensitive data redaction to protect customer privacy using conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/sensitive-data-redaction.html), section 'About redacted files', first four paragraphs. Checked 2026-09-30.
[^44]: For voice contacts, sensitive data redaction is applied after the call disconnects; for email contacts it is applied after the email contact ends. Source: [Use sensitive data redaction to protect customer privacy using conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/sensitive-data-redaction.html), third paragraph of the introduction. Checked 2026-09-30.
[^45]: The Amazon Connect admin guide's document history records, in an entry dated August 4, 2026, that Contact Lens is now called conversational analytics. Source: [Connect Customer document history](https://docs.aws.amazon.com/connect/latest/adminguide/doc-history.html), entry 'Feature name changes for Amazon Connect Customer', dated August 4, 2026. Checked 2026-09-30.
[^46]: Besides categorizing a contact, a rule can generate an EventBridge event, create a task, create a case, send email notifications, or submit an automated evaluation. Source: [Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases](https://docs.aws.amazon.com/connect/latest/adminguide/rules.html), section 'Step 2: Define actions'. Checked 2026-09-30.
[^47]: A conversational analytics rule can match contacts by four condition types: Words or Phrases - Exact Match, Words or Phrases Pattern Match, Natural Language - Semantic Match, and Words or Phrases - Semantic Match. Source: [Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases](https://docs.aws.amazon.com/connect/latest/adminguide/rules.html), section 'Step 1: Define conditions', step 6. Checked 2026-09-30.
[^48]: Natural Language - Semantic Match conditions, which use generative AI, cannot be used for real-time analysis, and creating rules that use generative AI requires the additional permission Rules - Generative AI. Source: [Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases](https://docs.aws.amazon.com/connect/latest/adminguide/rules.html), section 'Step 1: Define conditions', step 6, Note under Natural Language - Semantic Match. Checked 2026-09-30.
[^49]: Rules apply only to new contacts that occur after the rule was added; they cannot be applied to past, stored conversations. Source: [Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases](https://docs.aws.amazon.com/connect/latest/adminguide/rules.html), section 'Step 3: Review and save', step 2. Checked 2026-09-30.
[^50]: A conversational analytics rule is created for one of five analysis events: post-call analysis, real-time analysis, post-chat analysis, real-time chat analysis, or email analysis. Source: [Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases](https://docs.aws.amazon.com/connect/latest/adminguide/rules.html), section 'Step 1: Define conditions', step 5. Checked 2026-09-30.
[^51]: AWS describes conversational analytics as providing contact center analytics and quality management capabilities, and lists under it conversation analysis, performance evaluations, agent screen recording, contact search, and monitoring of live and recorded conversations. Source: [Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/contact-lens.html), intro paragraph and bulleted capability list. Checked 2026-09-30.
[^52]: Sentiment analysis is enabled by default when speech or chat analytics is enabled with a language that sentiment analysis supports, and it can be turned off for all agents and customers by deselecting Enable Sentiment Analysis in the block. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Disable sentiment analysis'. Checked 2026-09-30.
[^53]: Each speaker turn is classified as positive, negative or neutral; each period of a call is then given a score from -5 to +5 based on sentiment frequency and sentiment streaks, and the overall sentiment score is the average of the period scores. Source: [Investigate sentiment scores during contact conversations using conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/sentiment-scores.html), section 'How sentiment scores are determined'. Checked 2026-09-30.
[^54]: To show post-contact summaries on the agent's Contact Control Panel for voice contacts, the block must have call recording on for agent and customer, speech analytics enabled with the Real-time and post-call analytics option, and Post-contact summary selected under Generative AI capabilities. Source: [View generative AI-powered post-contact summaries in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/view-generative-ai-contact-summaries.html), section 'Enable post-contact summaries', procedure 'To enable post-contact summaries on the agent's CCP for voice contacts', step 2. Checked 2026-09-30.
[^55]: No more than one generative AI post-contact summary is generated per contact, and not every contact receives one. Source: [View generative AI-powered post-contact summaries in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/view-generative-ai-contact-summaries.html), section 'View post-contact summaries on the Connect Customer admin website', final paragraph. Checked 2026-09-30.
[^56]: The default quota for concurrent post-contact agent conversation summary jobs is 10, shared across voice and chat, and it is adjustable. Source: [Connect Customer service quotas](https://docs.aws.amazon.com/connect/latest/adminguide/amazon-connect-service-limits.html), section 'Conversational analytics service quotas' table, row 'Concurrent post-contact agent conversation summary jobs'. Checked 2026-09-30.
[^57]: Granular redaction is not supported for post-contact summaries; when granular redaction is selected, the summary redacts all PII identified in text and replaces it with a [PII] tag. Source: [View generative AI-powered post-contact summaries in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/view-generative-ai-contact-summaries.html), section 'Enable post-contact summaries', procedure for the admin website, step 1c. Checked 2026-09-30.
[^58]: When a summary is not generated, the analytics output file carries one of the reason codes QUOTA_EXCEEDED, INSUFFICIENT_CONVERSATION_CONTENT, INVALID_ANALYSIS_CONFIGURATION, FAILED_SAFETY_GUIDELINES or INTERNAL_ERROR; insufficient content means fewer than one utterance (voice) or one text/plain or text/markdown message (chat) from each participant. Source: [View generative AI-powered post-contact summaries in Connect Customer](https://docs.aws.amazon.com/connect/latest/adminguide/view-generative-ai-contact-summaries.html), section 'Why a summary is not generated'. Checked 2026-09-30.
[^59]: To keep collecting conversational analytics data after a contact is transferred to another agent or queue, another Set recording and analytics behavior block with analytics enabled is needed, because a transfer generates a second contact ID and contact record. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Important things to know', first bullet. Checked 2026-09-30.
[^60]: Conversational analytics supports calls with up to two participants; AWS says transcription and analytics quality can degrade with more parties and recommends disabling conversational analytics for multi-party or third-party calls. Source: [Enable conversational analytics in Connect Customer conversational analytics](https://docs.aws.amazon.com/connect/latest/adminguide/enable-analytics.html), section 'Multi-party calls and conversational analytics'. Checked 2026-09-30.