Amazon Connect (Next-Gen, Mar 18, 2025): Architecture & Provisioning Deltas

What changed (official): Amazon Connect now delivers first-party AI across all channels with an “all-you-can-eat” AI model: pricing is tied to channel usage, not AI consumption. This bundles capabilities such as Contact Lens (incl. analytics, transcription, redaction, QA, screen recording), agent assistance (Amazon Q in Connect), summaries, forecasting & scheduling (WFM), etc., under the Connect service’s per-channel charges (telephony/comms remain separate).

Next-Gen Amazon Connect Architecture: Unified Channels, Native AI, and Simplified Integrations

• Channels: Voice (PSTN/SIP), Chat/Web/Mobile, Email/SMS, and first-party self-service flows/bots feed into Amazon Connect Core (routing, queues, contact flows, unified agent workspace).
• First-Party AI Layer (bundled):
  • Contact Lens: analytics, transcription, PII redaction, screen recording, automated QA.
  • Generative AI: summaries, guidance/knowledge answers (Amazon Q in Connect), real-time agent assist.
  • WFM: forecasting & scheduling.
• Data & Events: S3 (recordings/transcripts/screen captures), Kinesis/EventBridge (contact events), KMS (encryption), Athena/Glue/QuickSight (reporting/BI).
• Integrations: CRM/ITSM (Salesforce, ServiceNow, Zendesk), Lambda/Step Functions, DynamoDB/RDS/APIs.

The big shift: AI features are native and consistently available across channels without per-feature AI metering, you plan capacity by channels, not by “how much AI.”

Expand Image

Provisioning changes vs. “classic” Connect

1) Pricing & scoping (Plan phase)

• New reality: Budget and forecast per channel (voice, chat, email, SMS, etc.); do not create separate line items for AI interactions/summaries/assist. Telephony and other communication costs are still separate.
• Action: Update TCO calculators, FinOps dashboards, and chargeback models to attribute AI benefits under channel usage, not per-AI unit. (We’ve already updated our calculators/templates.)

2) AI features (Enable, don’t “bolt-on”)

• Contact Lens: Turn on analytics + screen recording (new in the default AI bundle), configure KMS keys, S3 buckets & lifecycle, and PII redaction policies.
• Agent Assist / Amazon Q in Connect: Enable generative features for guidance and auto-summaries in the Agent Workspace; map access to knowledge sources (FAQs/KBs/CRM).
• WFM (forecasting & scheduling): Activate built-in forecasting/scheduling instead of third-party modules where possible.

3) Architecture simplification

• Fewer third-party AI dependencies: Replace external QA, summarization, and assist engines with first-party features in Connect. This reduces API hops, secrets/keys, and failure domains. (Industry press and AWS positioning emphasize simplification and cost predictability.)
• Unified channel behavior: Provision omnichannel flows with consistent AI policy, the same AI stack serves voice & digital without per-channel add-ons.

4) Data, security, and compliance

• S3 & lifecycle: Size for screen recordings + transcripts growth; set lifecycle to Glacier as needed.
• KMS: Centralize keys for recordings, transcripts, and Contact Lens artifacts; rotate per policy.
• Contact events: If you stream contact events, verify Kinesis/EventBridge throughput for higher AI metadata volume (summaries, QA, analytics).

5) Operations & observability

• Agent workspace config: Expose AI insights, summaries, and next-best actions; define guardrails and redaction scopes.
• Reporting: Point Athena/QuickSight to the enriched artifacts for QA coverage, compliance, and coaching dashboards.

“Before vs After” (at a glance)

VirtueCloud rollout (what we did post-announcement)

• Ref-arch update: Published the above AI-first Connect blueprint with S3/KMS/Kinesis and CRM hooks (Salesforce/ServiceNow/Zendesk).
• Templates: Ready-made Contact Lens + Agent Assist templates with PII redaction, auto-summaries, and QA scoring out-of-the-box.
• FinOps: Upgraded cost models to channel-based AI accounting; benchmarks for storage growth (audio + screen).
• Delivery playbooks: Cut average go-live time by removing third-party AI glue, fewer integrations, faster UAT. (Press and AWS guidance align with this simplification theme.)

Quick migration checklist

• Enable Contact Lens (incl. screen recording) and configure KMS + S3 lifecycle.
• Turn on Amazon Q in Connect (agent assist + summaries) and wire to knowledge sources.
• Activate WFM; validate forecasts vs historical volume.
• Update Event streaming (Kinesis/EventBridge) for higher metadata throughput.
• Refresh FinOps dashboards for channel-based AI accounting; keep telephony cost lines separate.
• De-scope third-party AI where redundant; reduce keys, secrets, and failure points.