Amazon Connect has grown from a simple cloud telephony service into a comprehensive contact centre platform. With so many features now available, it's easy to miss capabilities or misuse them. This guide covers every major feature — what it does, how to use it well, and what to avoid.
Channels & Routing
Voice (Telephony)
The core channel. Inbound and outbound voice calls with PSTN connectivity, DID numbers, toll-free, and softphone support for agents.
✓ Best Practices
- Claim numbers in the region closest to your customers to minimise latency
- Use DTMF as a fallback for speech recognition failures in noisy environments
- Set up emergency call routing that bypasses IVR menus entirely
✗ Anti-Patterns
- Claiming numbers in a single region for a global deployment — causes latency and poor audio quality
- Not configuring a timeout/no-input handler — callers sit in silence
- Using the CCP softphone over unstable WiFi without a network quality check
Chat
Asynchronous and synchronous messaging channel. Supports persistent conversations, rich messaging (attachments, links), and integrates with web, mobile, and third-party messaging platforms.
✓ Best Practices
- Set customer-facing typing indicators and read receipts to signal responsiveness
- Use persistent chat for complex cases that span multiple sessions
- Configure auto-disconnect timers appropriate to your use case (not too aggressive)
✗ Anti-Patterns
- Treating chat like voice — forcing synchronous responses when async would be better
- Not setting concurrent chat limits per agent — leads to quality degradation
- Ignoring chat abandonment metrics — customers leave silently unlike voice
Native email channel allowing customers to contact via email, with routing, prioritisation, and agent handling within the same workspace as voice and chat.
✓ Best Practices
- Set SLA-based routing priorities — email doesn't need real-time response but does need predictability
- Use auto-acknowledgement to confirm receipt immediately
- Integrate with Cases so email threads are linked to a single case record
✗ Anti-Patterns
- Routing emails to the same queue as real-time channels without priority differentiation
- Not parsing email content for intent — treating all emails as equal priority
- Allowing unlimited email assignments per agent without capacity management
SMS
Two-way SMS messaging channel for customer communication, notifications, and conversational support.
✓ Best Practices
- Use SMS for proactive notifications (appointment reminders, delivery updates) to reduce inbound volume
- Keep messages concise — SMS has character limits and customers expect brevity
- Implement opt-in/opt-out compliance from day one
✗ Anti-Patterns
- Sending SMS without explicit customer consent — regulatory risk
- Using SMS for complex multi-turn conversations better suited to chat
- Not handling delivery failures gracefully — messages can silently fail
In-App, Web & Video Calling
Embed voice and video calling directly into your website or mobile app using the Connect communication widget or SDK. Customers connect without leaving your digital experience.
✓ Best Practices
- Use contextual data from the web session (page URL, cart contents) to route intelligently
- Offer video selectively for high-value or complex interactions (e.g. insurance claims)
- Implement graceful fallback to voice if video quality degrades
✗ Anti-Patterns
- Offering video calling without training agents on camera presence and environment
- Not testing across browsers and devices — WebRTC compatibility varies
- Ignoring bandwidth requirements — video on poor connections is worse than no video
Routing (Queues & Routing Profiles)
Skills-based routing engine that matches contacts to the best available agent based on queue priority, agent skills, and routing profiles. Supports priority-based, longest-idle, and custom routing logic.
✓ Best Practices
- Design routing profiles around customer intent, not organisational structure
- Use queue priority to ensure high-value or time-sensitive contacts are handled first
- Regularly review and prune unused queues — complexity grows silently
✗ Anti-Patterns
- Creating a queue per team/department without considering the customer journey
- Setting all queues to the same priority — defeats the purpose of prioritisation
- Not using agent availability metrics to inform routing profile design
AI & Automation
Contact Flows (IVR Builder)
Visual drag-and-drop flow designer for building IVR experiences, chatbot interactions, and routing logic. Supports branching, Lambda integrations, and dynamic prompts.
✓ Best Practices
- Use modular flows (transfer to flow) to keep individual flows manageable and reusable
- Set logging to enabled at the start of every flow for troubleshooting
- Version control your flows — export JSON and store in source control
✗ Anti-Patterns
- Building monolithic flows with hundreds of blocks — impossible to maintain or debug
- Hardcoding values (phone numbers, messages) instead of using contact attributes or Lambda
- Not handling error paths — every Lambda invoke and integration needs a failure branch
Amazon Lex Integration (Conversational AI)
Native integration with Amazon Lex for natural language understanding in voice and chat. Enables free-speech IVR, intent recognition, slot filling, and conversational routing.
✓ Best Practices
- Start with a focused set of intents and expand based on production data — don't over-engineer upfront
- Use the confidence score thresholds and confirmation strategies appropriate to each intent
- Build a continuous optimisation process from Day 1 (see my other blog on this)
✗ Anti-Patterns
- Training with synthetic utterances only — real customers speak differently
- Not monitoring missed utterances post-launch — the bot stagnates
- Using a single fallback intent as a catch-all without analysing what's falling through
Amazon Q in Connect (Agent Assist)
Generative AI-powered real-time agent assistance. Surfaces relevant knowledge articles, suggests responses, and recommends next-best-actions based on the live conversation.
✓ Best Practices
- Curate and maintain your knowledge base — AI assist is only as good as the content it draws from
- Train agents to validate AI suggestions before using them — trust but verify
- Measure adoption rates and correlation with handle time/resolution improvements
✗ Anti-Patterns
- Enabling it without updating stale knowledge articles — AI confidently surfaces wrong information
- Not giving agents the ability to dismiss or flag bad suggestions
- Expecting it to replace training — it augments skilled agents, it doesn't create them
Contact Lens (Conversational Analytics)
ML-powered analytics for voice and chat. Provides real-time and post-contact transcription, sentiment analysis, keyword detection, talk-time metrics, non-talk-time detection, theme identification, and automated contact categorisation.
✓ Best Practices
- Define categories aligned to business outcomes (complaints, churn signals, compliance risks) not just keywords
- Use real-time alerts for supervisor intervention on negative sentiment trends
- Feed Contact Lens data into your BI tools for trend analysis over time
✗ Anti-Patterns
- Enabling transcription without redacting PII — sensitive data ends up in logs
- Creating hundreds of categories without a governance process — signal becomes noise
- Using sentiment scores as a punitive agent performance metric — it measures customer emotion, not agent quality
Contact Lens Evaluation Forms
Structured quality management forms that supervisors use to score agent interactions. Supports weighted scoring, automated pre-population from Contact Lens analytics, and trend tracking.
✓ Best Practices
- Use automated scoring where possible (e.g. "did agent use greeting?") to reduce supervisor workload
- Keep forms focused — 10-15 questions max, aligned to what actually drives outcomes
- Calibrate regularly across supervisors to ensure consistent scoring
✗ Anti-Patterns
- Building 50+ question forms that nobody completes consistently
- Evaluating only negative interactions — creates a biased view of agent performance
- Not acting on evaluation insights — collecting data without driving coaching
Step-by-Step Guides (Agent Workspace)
Visual workflow guides presented to agents in the CCP/agent workspace. Walks agents through complex processes step by step, reducing errors and training time.
✓ Best Practices
- Trigger guides dynamically based on queue, contact attributes, or customer segment
- Keep steps concise — each screen should require one decision or action
- Use guides for compliance-critical processes where consistency is non-negotiable
✗ Anti-Patterns
- Building guides for simple tasks that experienced agents can handle without guidance — it slows them down
- Not updating guides when processes change — agents learn to ignore them
- Making guides mandatory for all contact types regardless of complexity
Customer Data & Case Management
Customer Profiles
Unified customer data layer that aggregates information from multiple sources (CRM, order systems, marketing platforms) into a single profile presented to agents at the start of each interaction.
✓ Best Practices
- Connect all relevant data sources — the more context agents have, the faster they resolve
- Use identity resolution to merge duplicate profiles from different systems
- Surface the most actionable data first — recent orders, open cases, account status
✗ Anti-Patterns
- Dumping raw data without curation — agents get information overload
- Not setting up identity resolution — same customer appears as multiple profiles
- Treating it as read-only — agents should be able to update profiles during interactions
Cases
Native case management within Connect. Create, track, and resolve multi-touch customer issues across channels. Links contacts, notes, and resolution history to a single case record.
✓ Best Practices
- Auto-create cases from specific contact types to ensure nothing falls through the cracks
- Define clear case statuses and SLAs that align with customer expectations
- Use case templates for common issue types to standardise data capture
✗ Anti-Patterns
- Creating a case for every single contact — not everything needs case tracking
- Not linking related contacts to the same case — losing the full picture
- Allowing cases to remain open indefinitely without escalation rules
Workforce Management
Forecasting
ML-powered contact volume and handle time forecasting. Predicts future demand across channels using historical patterns, enabling accurate staffing plans.
✓ Best Practices
- Feed at least 6 months of historical data for reliable forecasts — more is better
- Create separate forecasts for different contact types/queues — don't aggregate everything
- Review forecast accuracy weekly and adjust for known events (marketing campaigns, outages)
✗ Anti-Patterns
- Trusting forecasts blindly without validating against actuals
- Not accounting for seasonality or one-off events that skew historical data
- Forecasting at too granular an interval without sufficient data to support it
Capacity Planning
Long-term staffing planning based on forecasted demand. Helps determine how many agents to hire, train, and schedule weeks or months in advance.
✓ Best Practices
- Factor in shrinkage (breaks, training, meetings) — raw headcount ≠ available capacity
- Model multiple scenarios (best case, worst case, expected) for budget planning
- Align capacity plans with recruitment lead times — hiring takes weeks, not days
✗ Anti-Patterns
- Planning capacity based on averages without accounting for peak variance
- Not revisiting plans when business conditions change (new product launch, market shift)
- Treating all channels as having the same concurrency — chat agents handle multiple conversations
Scheduling
Automated agent schedule generation based on forecasted demand, agent availability, skills, and business rules. Supports shift patterns, breaks, and schedule adherence tracking.
✓ Best Practices
- Publish schedules with enough lead time for agents to plan their lives
- Use schedule adherence monitoring to identify systemic issues, not punish individuals
- Build flexibility into schedules — allow shift swaps and voluntary overtime
✗ Anti-Patterns
- Generating schedules without agent input on preferences — kills morale
- Over-optimising for coverage without considering agent wellbeing
- Not configuring schedule notification rules — agents miss changes
Outbound & Proactive
Outbound Campaigns
High-volume outbound dialling with predictive, progressive, and agentless modes. Supports voice, SMS, and email campaigns with list management, retry logic, and compliance controls.
✓ Best Practices
- Start with progressive dialling and move to predictive only when you have enough agents to absorb the pace
- Implement time-zone aware dialling and respect do-not-call lists from day one
- Use answering machine detection to avoid wasting agent time on voicemails
✗ Anti-Patterns
- Using predictive dialling with too few agents — causes abandoned calls and regulatory violations
- Not scrubbing contact lists against DNC registers — legal exposure
- Running campaigns without clear success metrics — dialling for the sake of dialling
Tasks
Route, prioritise, and track non-real-time work items alongside voice and chat. Tasks can be created manually, from flows, or via API integrations with external systems (CRM, ticketing).
✓ Best Practices
- Use tasks for follow-up work that needs tracking — callbacks, form processing, approvals
- Set due dates and priority levels to ensure tasks don't languish in queues
- Integrate with external systems so tasks auto-create from triggers (e.g. failed payment)
✗ Anti-Patterns
- Using tasks as a general to-do list without routing or SLA management
- Not connecting tasks to the originating contact — losing context
- Creating tasks that duplicate work already tracked in another system
Analytics & Reporting
Real-Time Metrics
Live dashboards showing queue depth, agent states, service levels, and contact volumes in real time. Enables supervisors to make immediate staffing and routing decisions.
✓ Best Practices
- Configure wallboards for supervisors showing the 3-5 metrics that drive action
- Set threshold alerts for queue depth and wait time — don't rely on humans watching screens
- Use real-time data to trigger dynamic routing changes (overflow, skill relaxation)
✗ Anti-Patterns
- Displaying 30+ metrics on a single dashboard — information overload, no action taken
- Not acting on real-time data — having visibility without authority to change anything
- Using real-time metrics for historical analysis — they're snapshots, not trends
Historical Metrics & Reporting
Detailed historical reporting on contacts, queues, agents, and routing. Supports custom date ranges, grouping, filtering, and scheduled report delivery.
✓ Best Practices
- Export CTR data to S3 and query with Athena for complex analysis beyond the built-in reports
- Build a reporting cadence — daily operational, weekly tactical, monthly strategic
- Track trends over time, not just point-in-time snapshots
✗ Anti-Patterns
- Relying solely on the Connect UI for reporting — it has limitations at scale
- Not retaining historical data beyond the default retention period
- Comparing metrics across time periods without accounting for volume changes
Security & Compliance
Data Redaction (PII)
Automatic detection and redaction of sensitive data (credit card numbers, SSNs, etc.) from transcripts, recordings, and logs. Configurable per data type.
✓ Best Practices
- Enable redaction before go-live, not after — retroactive redaction is painful
- Test redaction rules with real-world data patterns — customers express PII in unexpected ways
- Redact both audio and transcript — one without the other leaves gaps
✗ Anti-Patterns
- Assuming default redaction catches everything — custom patterns need custom rules
- Enabling redaction without testing — over-aggressive rules redact non-sensitive content
- Storing unredacted data in S3 "just in case" — defeats the purpose entirely
Recording & Storage
Call recording, screen recording, and chat transcript storage. Configurable retention policies, encryption at rest, and access controls.
✓ Best Practices
- Define retention policies aligned to regulatory requirements from the start
- Use S3 lifecycle policies to move old recordings to cheaper storage tiers
- Implement access controls — not everyone needs access to recordings
✗ Anti-Patterns
- Keeping recordings forever without a retention policy — storage costs grow silently
- Not encrypting recordings at rest — compliance failure
- Recording all calls including payment segments without pause/resume controls
Integration & Extensibility
Lambda Integration
Invoke AWS Lambda functions from contact flows to perform data lookups, business logic, API calls, and dynamic routing decisions in real time.
✓ Best Practices
- Keep Lambda functions fast — aim for under 2 seconds; every millisecond is silence for the caller
- Implement error handling and timeouts — a failed Lambda shouldn't crash the entire flow
- Use environment variables for configuration — don't hardcode endpoints or credentials
✗ Anti-Patterns
- Putting all business logic in a single monolithic Lambda — impossible to debug or maintain
- Not warming Lambdas for latency-sensitive paths — cold starts cause noticeable delays
- Logging sensitive customer data in CloudWatch without redaction
EventBridge Integration
Real-time event streaming from Connect to EventBridge. Contact events (created, connected, disconnected), agent events, and Contact Lens events can trigger downstream workflows.
✓ Best Practices
- Use events to trigger real-time actions — supervisor alerts, CRM updates, analytics pipelines
- Filter events at the rule level to avoid processing noise
- Build idempotent consumers — events can be delivered more than once
✗ Anti-Patterns
- Processing every event without filtering — creates unnecessary compute cost
- Building critical workflows that assume exactly-once delivery
- Not monitoring dead-letter queues for failed event processing
CTR (Contact Trace Records) & Data Streaming
Detailed records of every contact including timestamps, queue times, agent info, attributes, and outcomes. Streamable to Kinesis for real-time processing or S3 for batch analytics.
✓ Best Practices
- Stream CTRs to S3 via Kinesis Firehose for long-term analytics with Athena/QuickSight
- Use contact attributes to enrich CTRs with business context (customer segment, intent, outcome)
- Build a data lake early — retrofitting analytics is much harder than building it from the start
✗ Anti-Patterns
- Not streaming CTRs anywhere — losing valuable operational data
- Relying on the Connect UI as your only reporting tool at scale
- Not setting contact attributes in flows — CTRs end up with no business context
Agent Workspace (CCP & Custom)
The agent-facing application. Includes the Contact Control Panel (CCP) for call handling, plus customisable workspace with embedded applications, customer profiles, cases, and guides.
✓ Best Practices
- Customise the workspace to surface only what agents need for their role — less is more
- Embed third-party applications (CRM, knowledge base) to eliminate tab-switching
- Test the workspace on the actual hardware and network agents will use
✗ Anti-Patterns
- Forcing agents to use multiple disconnected applications alongside the CCP
- Not testing CCP performance on agent machines — browser requirements matter
- Customising the workspace without agent input on what they actually need
Summary
Amazon Connect is no longer just a phone system. It's a platform with 20+ distinct capabilities spanning channels, AI, workforce management, analytics, security, and integration. The organisations that get the most value are those that adopt features deliberately — understanding not just what each one does, but how to use it well and what pitfalls to avoid.
This guide will be updated as new features launch. Check the Release Radar for the latest changes.