A score without the recording is an opinion. A recording without a form is an archive.
Evaluation forms were previewed 29 November 2022 and became generally available 26 April 2023. They sit with conversational analytics. You can score a contact, calibrate reviewers, and coach. Generative scoring, where you turn it on, is still a score someone must be willing to defend.
The quota table limits how many evaluation questions AI may answer on one contact. Read that limit before you design a 40-question form and assume a model fills it.
Calibrate reviewers on the same five contacts.
The transcript can miss a word. The audio decides.
Redaction must run before a reviewer who should not see the secret sees the form.
Coaching is a workflow. The form is not the coaching.
Number on this page: Form GA 2023-04-26. Evaluation capabilities
Confirm before relying. Prices, quotas, and exam names change. Use the source linked on this page.
Which term matches this description? Scoring contacts against a form, coaching from the score, and keeping the recording that proves it.
Show the cited answer
Scoring contacts against a form, coaching from the score, and keeping the recording that proves it. The term is Quality management.
confirm Evaluation forms. AWS docs. last checked 2026-10. Opened 2026-10. The link left this article and opened the Amazon Connect Customer guide home.
Question 2
Which term matches this description? The WAV Connect writes when recording is on. Third-party QM tools often import this object plus the CTR.
Show the cited answer
The WAV Connect writes when recording is on. Third-party QM tools often import this object plus the CTR. The term is S3 recording object.
Start media streaming Sends live customer audio to Kinesis Video Streams for a third-party recorder or bot.
Screen recording Records the agent desktop. On the Basic price list it is 0.006 USD per minute.
Forecasting and agent scheduling The 4 Aug 2026 name for native WFM: forecast, capacity, schedules, and adherence.
Schedule adherence How closely an agent's actual status matched the published schedule. Native metrics appear after schedules are published. Third-party WFM reads agent events.
Connect answers the contact. Workforce, quality, emergency location, and the system of record sit beside it, and they only stay honest if they share ContactId.
Config and troubleshooting
Auto-failing an agent on a summary that paraphrased badly.
Adherence needs agent events, not a person refreshing a report. ResourceType AGENT_EVENTS and StorageType KINESIS_STREAM is the hook. A second consumer should read the stream, not a second delivery on the instance.
Join records on ContactId. The integration map room lists the seams.
Cross-platform bridges
A CUCM calling search space is not a routing profile, but the profile is the closest idea of 'who can receive what'. A partition is not a Connect object. A UCCX CSQ and a UCCE skill group map to a queue. Precision steps map to routing criteria. A script, an Architect flow, and an Avaya vector map to a contact flow.
Hooks you can adopt later: EventBridge contact events, the CTR stream, agent events, Kinesis Video, Customer Profiles events, Cases, Show view, data tables, External Tool, and the flow language JSON in your pipeline. Live Sync and Agentic CX are the 2026 hooks for guided and agentic experiences.