Amazon Connect Architecture Explained: Components, Flows & Reference Designs (2026)

Amazon Connect is AWS’s cloud contact centre service. There are no servers, PBXs or SBCs to run: you build the whole contact centre from managed AWS building blocks and pay only for what you use. This guide explains Amazon Connect architecture end to end — the core components, how a call or chat flows through them, how it connects to the rest of AWS, and the reference architectures used in real projects.

It is written for solution architects, developers and contact centre engineers, and for anyone preparing for an Amazon Connect interview. If you want to build these architectures hands-on with an instructor, see our Amazon Connect training course.

Amazon Connect architecture at a glance

Every Amazon Connect deployment is built from five layers:

1. Channels and entry points – phone numbers, web/mobile chat, tasks, email and outbound campaigns.
2. The Amazon Connect instance – the contact centre itself: flows, queues, routing profiles, users and security profiles.
3. Intelligence – Amazon Lex bots, Amazon Polly voices, Contact Lens analytics and Amazon Q in Connect for AI agents and agent assist.
4. Integration – AWS Lambda, Amazon DynamoDB, API Gateway and CRM connectors (Salesforce, ServiceNow, Zendesk, Dynamics 365).
5. Data and reporting – contact records, agent events, recordings and transcripts streamed to Amazon S3, Kinesis, the analytics data lake and dashboards.

The rest of this guide walks through each layer.

1. The Amazon Connect instance

An instance is a single, isolated contact centre in one AWS Region. Everything else hangs off it:

• Identity management – users are stored in Connect’s own directory, in an existing AWS Directory Service directory, or federated through SAML 2.0 (Okta, Entra ID, etc.). Choose this at creation time; it cannot be changed later.
• Users, security profiles and agent hierarchies – who can log in, what they can see and how they are grouped for reporting.
• Data storage – call recordings, chat transcripts, exported reports and attachments go to Amazon S3 buckets you own, encrypted with AWS KMS keys.
• Service quotas – concurrent calls, phone numbers, queues and flows per instance are limited by default and raised through Service Quotas.

Most organisations run at least two instances (non-production and production). Large or regulated organisations add more per business unit or Region.

2. Telephony and channels

Amazon Connect includes carrier-grade telephony. You claim DID or toll-free numbers directly in the console, or port existing numbers in. Each number is associated with a contact flow, which is the entry point for that call.

• Voice – inbound and outbound calls, with recordings and real-time media streaming available through Amazon Kinesis Video Streams.
• Chat – the hosted communications widget for websites, or the Chat APIs for custom and mobile apps. Persistent chat keeps the conversation history.
• Tasks – back-office work items (from CRM cases, emails or API calls) routed to agents just like calls.
• Email and outbound campaigns – email contacts and high-volume outbound dialling are now part of the same routing model.

Because every channel lands in the same queues and routing profiles, agents work all channels from a single desktop.

3. Contact flows: the brain of the contact centre

Contact flows are the visual workflows that decide what happens to each contact. They replace the IVR scripts and routing scripts of traditional platforms. There are several flow types:

• Inbound flow – the main entry point: greeting, menu, self-service, authentication and transfer to a queue.
• Customer queue flow – what customers hear while waiting (music, position in queue, callback offers).
• Customer whisper / agent whisper flows – messages played to each side as the call connects.
• Transfer to agent and transfer to queue flows – used for warm and cold transfers.
• Disconnect flows – run after the agent leaves, for example for post-call surveys.

Flows call out to Lambda functions for business logic, use contact attributes to carry data through the call, and use modules to reuse common steps. In mature projects flows are exported as JSON and deployed through CloudFormation, CDK or Terraform so every environment stays identical.

4. Routing: queues and routing profiles

Routing in Amazon Connect is based on three objects:

• Queues – hold contacts waiting for an agent. Each queue has hours of operation and an outbound caller ID.
• Routing profiles – assigned to each agent. A routing profile lists the queues the agent serves, the priority and delay for each queue, and how many contacts per channel the agent can handle at once.
• Proficiencies and routing criteria – skills-based routing that matches a contact to agents with the right attributes (language, product, level), relaxing the criteria step by step if nobody is available.

Contacts are always delivered to the longest-available agent whose routing profile includes that queue, respecting priorities and delays. This simple model replaces complex skill-group scripting found in platforms like Cisco UCCE.

5. AI and self-service layer

• Amazon Lex – conversational voice and chat bots for intent recognition, slot filling and self-service.
• Amazon Polly – neural text-to-speech for prompts, so you rarely need recorded audio files.
• Amazon Q in Connect – generative AI agents for customer self-service and real-time agent assist, grounded in your knowledge base.
• Contact Lens – real-time and post-contact transcription, sentiment, categories, redaction of sensitive data, conversational analytics and agent performance evaluations.

A typical modern design puts a Lex or Amazon Q self-service step at the front of the inbound flow, passes the collected intent and customer data as contact attributes, and only then routes to a human agent with full context.

6. Integration layer: Lambda, data stores and CRMs

AWS Lambda is how Amazon Connect talks to everything else. A flow invokes a Lambda function, waits for the response (up to 8 seconds per invocation) and branches on the returned values. Common patterns:

• Customer lookup by phone number (ANI) in Amazon DynamoDB, a CRM or a core banking API.
• Authentication with a PIN, OTP or voice ID, then personalised routing.
• Dynamic prompts and holiday or emergency messages stored in DynamoDB so business users can change them without editing flows.
• Creating or updating tickets in ServiceNow, Zendesk or Salesforce.

For the agent desktop, Amazon Connect offers the Contact Control Panel (CCP), the customisable agent workspace, and the Amazon Connect Streams JavaScript library for embedding the softphone inside a CRM. Ready-made CTI adapters exist for Salesforce, ServiceNow, Zendesk and Microsoft Dynamics 365. Customer Profiles and Cases add a built-in customer 360 view and case management.

7. Data, reporting and monitoring architecture

Every contact produces data you can stream out of Amazon Connect:

• Contact records (CTRs) – one record per contact, streamed to Amazon Kinesis Data Streams or Kinesis Data Firehose.
• Agent events – real-time agent state changes, streamed to Kinesis Data Streams.
• Recordings and transcripts – stored in Amazon S3.
• Analytics data lake – a managed data lake you can query with Amazon Athena and visualise in Amazon QuickSight.
• CloudWatch – metrics and flow logs for monitoring, alarms and troubleshooting.

A common reporting pipeline is: Connect → Kinesis → Lambda or Firehose → S3 → Glue / Athena → QuickSight, giving supervisors historical dashboards beyond the built-in real-time and historical reports.

8. Resilience and multi-region design

An Amazon Connect instance is already spread across multiple Availability Zones inside its Region. For Region-level resilience you can use Amazon Connect Global Resiliency: a replica instance in a second Region and traffic distribution groups that move telephony and agents between Regions in minutes. Supporting services (Lambda, DynamoDB global tables, Lex bots) must be deployed in both Regions too.

9. Reference architectures used in real projects

a) Self-service IVR with CRM lookup

Phone number → inbound flow → Lambda customer lookup (DynamoDB or CRM) → Lex bot for intent → self-service or queue → agent with screen-pop in the CRM through Streams.

b) Omnichannel contact centre

Voice, chat, email and tasks all enter their own flows but share queues, routing profiles and the agent workspace. Customer Profiles unifies the history so agents see every previous interaction.

c) AI-first contact centre

Amazon Q in Connect handles routine questions end to end, Contact Lens provides real-time sentiment and summaries, and agents receive AI suggestions during the conversation. Only complex cases reach a human.

d) Analytics and quality management

CTRs and agent events stream to S3 through Kinesis; Contact Lens adds transcripts and categories; QuickSight dashboards and automated evaluations give supervisors full visibility.

10. Security best practices

• Use SAML federation and least-privilege security profiles.
• Encrypt S3 buckets, Kinesis streams and recordings with customer-managed KMS keys.
• Use Contact Lens redaction to remove card numbers and personal data from transcripts and recordings.
• Restrict Lambda permissions to the specific resources each function needs.
• Manage everything as code (CloudFormation, CDK or Terraform) and log changes with AWS CloudTrail.

Amazon Connect architecture interview tips

Interviewers usually ask you to draw one of the reference architectures above and explain the trade-offs: Lambda timeouts, quotas, data retention, multi-region failover and CRM integration. Practise with our 50 Amazon Connect architecture interview questions, the Top 50 Amazon Connect interview questions and the free Amazon Connect practice test.

Frequently asked questions

Is Amazon Connect serverless?

Yes. There is no infrastructure to manage. AWS runs the telephony, media and routing; you configure the instance and extend it with other serverless services such as Lambda and DynamoDB.

What is the difference between a contact flow and a routing profile?

A contact flow decides what happens to a contact (prompts, self-service, which queue). A routing profile decides which queues an agent takes contacts from, and with what priority.

How does Amazon Connect integrate with Salesforce or ServiceNow?

Through the official CTI adapters or the Amazon Connect Streams library for the agent softphone, and through Lambda functions or AppFlow for data lookups and updates.

How do I learn to build Amazon Connect architectures?

Start with the AWS documentation and a free-tier instance, then work through real projects. Our live Amazon Connect training covers all of the architectures in this guide across 17 modules with hands-on labs.