Amazon Connect Customer
Also known as: Amazon Connect, AWS Connect, Connect Customer, Amazon Q in Connect, Connect AI agents, Amazon Connect AI agents
Amazon Connect Customer, called Amazon Connect until April 2026, is the AWS contact center whose AI agents resolve returns, account changes and rebookings across voice, chat, email, SMS and WhatsApp, calling tools through MCP and handing off to people with full context. Teams pick Bedrock models per prompt, add guardrails, trace every reasoning step and test agents in simulation. It handles 20 million interactions a day for over 10,000 customers, priced per use with AI included.
Amazon Connect Customer is the AWS contact center, called Amazon Connect until April 2026, when AWS made Connect the name of a family that also includes Talent for hiring, Decisions for supply chain and Health for clinical administration. It handles more than 20 million interactions a day for over 10,000 customers, among them TUI, Toyota, American Airlines and Cochlear, and ran 12 billion minutes of AI powered conversations in 2025. Zepz resolves 67% of customer inquiries instantly through self service on it, and Cochlear scaled its quality evaluations 22 times over.
Its AI agents resolve requests end to end, processing returns, updating accounts or rebooking flights, over voice, chat, email, SMS, web, WhatsApp and Apple Messages, and a customer can start in chat and move to voice while Live Sync walks them through forms on screen.
Orchestration agents reason over each turn and call tools through MCP: ready tools that update contact attributes or read cases, flow modules turned into tools so existing business logic is reused, a built in server for Connect, Customer Profiles, Cases and the assistant, and any API or Lambda function exposed through Amazon Bedrock AgentCore Gateway. Multi agent setups hand work between agents over the open A2A standard.
Steps that must be exact, such as identity checks and compliance disclosures, run as fixed rule based steps inside a governed flow, built on a no code agentic CX designer that reached general availability in September 2026. The same AI assists human agents in the workspace with answers, next steps and after call work, and on escalation the full context passes to the person.
Each AI prompt runs on a model chosen from the Amazon Bedrock models available in the instance's Region, including Amazon Nova and Anthropic Claude, AI guardrails filter content and denied topics, and security profiles bound what an AI agent can do just as they do for staff.
Dashboards and trace views show prompts, tool calls and reasoning across multi agent handoffs, and native testing and simulation, with APIs for voice since February 2026 and chat since March, validate self service conversations before release. Connect is HIPAA eligible and supports PCI DSS and FedRAMP workloads. It also carries SOC and ISO 27001 attestations and supports GDPR, with data kept in the Region the customer picks. Pricing is pay as you go with no seats or token charges and AI included: $0.038 a voice minute plus telephony, $0.010 a chat message, $0.014 an SMS and $0.080 an email.
It is built for large contact centers already on AWS, or ready to move there, that want AI agents and human staff on one platform with usage pricing. Buyers wanting a packaged agent with no design work, or a platform outside AWS, face more assembly here than with a ready made service agent.
Vendor details
Canonical URL
https://aws.amazon.com/connect/
Category
Customer support agent
Subcategory
AWS contact center with AI agents for self service and agent assistance across voice and digital channels, MCP tools, Bedrock model choice, guardrails, tracing and simulation testing
Funding status
Product of Amazon Web Services (Amazon.com, Inc., NASDAQ: AMZN).
Company status
first party product
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
MCP tools: out of the box Connect tools, flow modules as tools, built in MCP server for Connect, Customer Profiles, Cases and the assistant, and third party APIs and Lambda functions through Amazon Bedrock AgentCore Gateway; A2A for outside agents; Amazon Bedrock models; knowledge bases such as Amazon S3; voice, chat, email, SMS, web, WhatsApp and Apple Messages; Amazon Lex; Contact Lens; AWS SDKs and APIs.
In practice
An airline builds an orchestration agent that verifies the caller in a fixed flow step, then uses a rebooking tool exposed through AgentCore Gateway to move a passenger after a cancellation, escalating to a person with the full transcript when no seat fits.
A retailer starts a return in web chat; the customer switches to voice mid way, Live Sync shows the return form on their phone, and the agent books the courier pickup and updates the order case.
A contact center team writes test cases in the simulation builder that play a caller asking to change a billing address, runs them against each new prompt version, and checks the AI agent dashboard and traces after release.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
12.0 / 14 capabilities · 86%
| Integrations & Tool Calling | Full |
|---|---|
|
AI agents call tools through MCP: ready tools that update contact attributes or read case details, flow modules turned into tools so existing business logic is reused, a built in server for Connect, Customer Profiles, Cases and the assistant, and any API or Lambda function exposed through Amazon Bedrock AgentCore Gateway. Security profiles decide which of those tools each AI agent may use. Sourcedocs.aws.amazon.com/connect/latest/adminguide/ai-agent-mcp-tools.htmlread 2026-11-19 |
|
| Workflow Orchestration | Full |
|
Orchestration agents reason over each turn and choose tools, while steps that must be exact, such as identity verification and compliance disclosures, run as fixed rule based steps inside a governed flow built on the no code agentic CX designer. Multi agent setups connect existing agents over the open A2A standard, with Connect coordinating handoffs and passing context between them. Sourceaws.amazon.com/products/connect/customer/ai-agentsread 2026-11-19 |
|
| Knowledge Grounding & RAG | Full |
|
Agents retrieve answers from knowledge bases connected to the assistant, such as policy documents in Amazon S3, and read a customer's full history from Customer Profiles and Cases during a contact. Answers are auditable, and the same knowledge serves human agents in the workspace through recommended answers and manual search, so self service and staff work from one source. Sourceaws.amazon.com/products/connect/customer/ai-agentsread 2026-11-19 |
|
| Human Oversight & Guardrails | Full |
|
Each orchestration agent has Complete and Escalate tools that return control to the flow or pass the contact to a person with full context. Security profiles bound what an AI agent can do the same way they bound staff, AI guardrails filter content, denied topics and profanity, and identity checks and disclosures run as fixed steps the model cannot skip. Sourcedocs.aws.amazon.com/connect/latest/adminguide/agentic-self-service.mdread 2026-11-19 |
|
| Security, Identity & Governance | Full |
|
The service is HIPAA eligible and supports PCI DSS and FedRAMP workloads. It also carries SOC and ISO 27001 attestations and supports GDPR. Role based permissions govern who builds AI agents and which models are approved, security profiles bound each agent's abilities, and Amazon Bedrock's automated abuse detection runs underneath, with data kept in the customer's chosen Region. Sourceaws.amazon.com/connectread 2026-11-19 |
|
| Observability & Auditability | Full |
|
Dashboards track AI agent performance, compliance and optimization, and trace views show the prompts, tool calls and reasoning behind each answer, including across multi agent handoffs. Every reasoning step is governed per turn with auditable answers, and Contact Lens analytics and quality evaluations cover AI and human contacts alike in the same views. Sourceaws.amazon.com/products/connect/customer/ai-agentsread 2026-11-19 |
|
| Memory & State Persistence | Partial |
|
Context follows the customer across channels and into a human handoff, so nobody repeats themselves, and Customer Profiles give agents a unified record of past transactions and cases. That history is customer data the agent reads, not memory it keeps: an AI agent carries no store of its own from one contact to the next, with a scope and lifetime the contact center sets and can clear. Sourceaws.amazon.com/connectread 2026-11-19 |
|
| Deployment & Data Residency | Full |
|
Each instance runs in the AWS Region the customer selects and customer data stays in that Region, and the models offered for AI prompts follow the Region of the instance, with some marked as cross region options. The service is cloud only, with no on premises edition, but the choice of Region covers residency needs across AWS's global footprint of Regions. Sourcedocs.aws.amazon.com/connect/latest/adminguide/create-ai-prompts.htmlread 2026-11-19 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
A system AI agent exists for each use case, answer recommendation, manual search and self service, each preset with default prompts and models, and ready MCP tools cover common contact tasks. Customers still design their own self service journeys in the agentic CX designer, and there is no library of finished industry agents to switch on and run unchanged. Sourcedocs.aws.amazon.com/connect/latest/adminguide/create-ai-agents.htmlread 2026-11-19 |
|
| Triggers & Channel Coverage | Full |
|
Agents answer on voice, chat, email, SMS, web, WhatsApp and Apple Messages with the same context on every channel, and a customer can start in chat and switch to voice while Live Sync guides them through forms on screen. Inside the agent workspace the assistant works alongside human agents during live contacts and after them, handling after call work. Sourceaws.amazon.com/connectread 2026-11-19 |
|
| Model Flexibility & Routing | Full |
|
Every AI prompt runs on a model the customer picks from the Amazon Bedrock models offered in the instance's Region, such as Amazon Nova and Anthropic Claude, with defaults set by AWS and a ListModels call that shows what is available. Different prompts within the same agent can each be set to a different model, and role based permissions control which models are approved. Sourcedocs.aws.amazon.com/connect/latest/adminguide/create-ai-prompts.htmlread 2026-11-19 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
AI agents, prompts and guardrails can be configured through the AI agents API as well as the console, alongside the wider Connect APIs and AWS SDKs. A built in MCP server exposes Connect, Customer Profiles, Cases and assistant functions, the A2A standard links outside agents, and testing and simulation run through APIs as well as the console's visual test builder. Sourcedocs.aws.amazon.com/connect/latest/adminguide/connect-ai-agent.mdread 2026-11-19 |
|
| Testing, Debugging & Optimization | Full |
|
Native testing and simulation lets teams build test cases that call or chat into a flow, play the customer's side, observe responses and check outcomes before release, with APIs for voice since February 2026 and chat since March 2026. After launch, quality evaluations in Contact Lens score AI and human contacts, which Cochlear scaled 22 times over. Sourceaws.amazon.com/about-aws/whats-new/2026/02/amazon-connect-provides-apis-test-simulate-voice-interactionsread 2026-11-19 |
|
| Browser / Computer-use | Not documented |
|
AI agents act through MCP tools, flows, Lambda functions and APIs. Live Sync shows the customer forms on their own screen, but no agent drives a browser or a desktop to work in an application without an API, so systems reachable only through a screen, such as an older billing console, stay with human agents or need an API wrapper built for them first. Sourcedocs.aws.amazon.com/connect/latest/adminguide/ai-agent-mcp-tools.htmlread 2026-11-19 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Pay as you go with AI included: voice $0.038/min plus telephony, chat $0.010/message, SMS $0.014/message, email $0.080/email
Per voice minute, chat message, SMS or third party message and email, with AI included in the channel price
What is public
Amazon Connect Customer charges $0.038 a voice minute, $0.010 a chat message, $0.014 an SMS or third party message and $0.080 an email, with AI agents included in those rates.
Billing mechanics
Billing is pay as you go per interaction, with no seats, minimums, long term contracts or token charges.
Cost watchouts
Telephony is billed on top of the voice minute rate, and rates vary by Region.
Variable cost rationale
Every minute, message and email is billed as used, so cost follows contact volume directly, with telephony charged on top of voice.
Additional watchouts
Telephony costs come on top of the voice minute, and rates differ by Region and provider.
Sales call required
No, self serve available
Free / trial
No free plan or trial terms stated; usage billing with no seats or minimum commitment
Lowest paid plan
Pay as you go: chat at $0.010 a message, voice at $0.038 a minute
Commercial notes
A Customer Basic option serves contacts that do not use AI.
Key ambiguities
What Customer Basic costs for contacts that use no AI is not public in the headline rates.
Cancellation / refund
No long term contract required.
Missing data
Rates for Customer Basic and for each Region sit outside the headline price list.
Related vendors
- Decagon — AI agent platform purpose-built for customer support, with deep…
- 5.Y — 5
- Ada — Agentic customer experience platform whose AI agents autonomously…
- Aisera — Enterprise agentic AI platform that orchestrates specialized agents…
- ASAPP — Generative AI for enterprise contact centers that plugs into…
- Assembled — AI customer support orchestration platform that unifies autonomous…
Alternatives to Amazon Connect Customer
Other products that do the work Amazon Connect Customer does, and what each one is.
- Kapture CX
- Blits.aiEnterprise platform for chat, voice and digital human agents
- ElevenAgentsVoice first agent platform
- KhorosEnterprise brand community and social care platform
- QuiqEnterprise agentic AI platform for customer experience
- CognigyEnterprise conversational and agentic AI platform now owned by NiCE