Migrating from Avaya to Amazon Connect: What UK Contact Centres Get Wrong and How to Land AI Voice Agents from Day One

Arkadas Kilic 9 min read · 2026-07-25

Avaya's financial turbulence has accelerated a decision that many UK contact centre leaders had been deferring for years. The question is no longer whether to migrate, but how to do it without repeating the mistakes that turn a 12-week project into a 12-month ordeal, and without leaving AI capabilities as a vague phase two promise that never ships.

This guide is for operations and technology leaders who want to migrate from Avaya to Amazon Connect and have production AI voice agents handling real calls on day one of go-live, not six months later.

Why UK Contact Centres Are Moving Now

Avaya's Chapter 11 filing in 2023 removed any remaining doubt about platform longevity. UK enterprises running Avaya Aura, Avaya IX Contact Centre, or legacy AACC environments are now actively evaluating exits. The drivers are consistent across regulated industries:

• Support contracts becoming harder to renew or prohibitively expensive
• Integration debt with modern CRM and workforce management platforms
• Inability to deploy AI capabilities without third-party bolt-ons
• AWS-native architecture aligns with existing cloud strategy and data residency requirements under UK GDPR

Amazon Connect is the destination for the majority of these migrations. It is consumption-based, requires no upfront licence commitment, and its native integration with the AWS ecosystem means AI voice capabilities are not an add-on. They are part of the architecture from the start.

The Seven Mistakes UK Contact Centres Make During Avaya Migration

1. Treating the Migration as a Lift and Shift

The most expensive mistake is replicating your Avaya call flows inside Amazon Connect. Avaya IVR trees built over 10 to 15 years accumulate dead branches, redundant routing logic, and workarounds for platform limitations that no longer exist. Migrating them verbatim means you inherit the technical debt and forfeit the opportunity to redesign for AI-assisted routing from the outset.

The correct approach is a flow audit before any build begins. Categorise every existing flow as: retain and rebuild, redesign, or retire. In our experience, 30 to 40 percent of legacy IVR branches can be retired entirely.

2. Underestimating Number Porting Lead Times

UK number porting from Avaya-hosted or carrier-hosted DDI ranges to Amazon Connect via a UK-approved carrier is not a same-week activity. Porting 50 to 500 numbers through BT Wholesale, Gamma, or Vonage typically takes 4 to 8 weeks depending on the losing carrier's cooperation. Migrations that do not account for this in the project plan hit go-live delays that have nothing to do with the technology.

Start the porting process in parallel with your build. Do not wait for the Connect environment to be complete before submitting port orders.

3. Skipping the Agent Desktop Integration Design

Amazon Connect's native Contact Control Panel works. But most UK enterprise contact centres have Salesforce, Microsoft Dynamics, or a bespoke CRM that agents rely on for case context. The integration between Connect and your CRM determines whether agents see a screen pop with full customer history or a blank screen and a phone number.

Connect's Streams API and the pre-built Salesforce CTI adapter are production-grade. The mistake is leaving this integration to the final two weeks of the project. It should be scoped, built, and tested in the first half of the delivery timeline.

4. Deferring AI to Phase Two

This is the mistake we see most consistently, and it is the most costly in terms of missed business value. Teams spend 8 to 10 weeks migrating call flows, complete a successful cutover, and then start a separate AI project. That separate project then takes another 6 to 12 months to reach production because the foundational decisions around data, authentication, and contact flow architecture have already been made without AI in mind.

If you design the Amazon Connect architecture with AI voice agents as a first-class requirement from week one, the marginal cost of deploying those agents at go-live is a fraction of retrofitting them later. We cover exactly how to do this below.

5. Ignoring UK Compliance Requirements During Architecture Design

UK contact centres operating in financial services, insurance, utilities, or healthcare face specific obligations: FCA call recording rules, PCI DSS for payment handling, UK GDPR for data residency, and Ofcom regulations around automated calling. Amazon Connect supports all of these natively, but only if the architecture is designed with them in mind.

PCI DSS pause-and-resume recording, data residency in eu-west-2 (London), and consent capture flows need to be designed in, not added after go-live. A compliance review at the architecture stage costs hours. A compliance remediation after go-live costs weeks.

6. Underresourcing the Telephony Testing Phase

Avaya environments often have complex PSTN configurations, hunt groups, and overflow routing that behave differently under load. Amazon Connect's behaviour under concurrent call volume needs to be tested against your actual traffic patterns, not a theoretical model. Load testing with 80 to 120 percent of peak concurrent calls, testing queue overflow, testing agent failover, and testing AI agent fallback to human should all happen in a dedicated UAT phase of at least two weeks.

7. Not Training Supervisors on Real-Time Analytics

Amazon Connect's real-time and historical reporting is significantly more capable than most Avaya deployments. Contact Lens provides conversation analytics, sentiment scoring, and automated quality evaluation. The mistake is completing the migration and leaving supervisors using the same reporting habits they had on Avaya. The ROI from Connect's analytics is only realised if supervisors are trained to act on the data.

How to Land AI Voice Agents from Day One

Deploying production AI voice agents at go-live is achievable in a 4 to 6 week migration timeline if the architecture decisions are made correctly at the start. Here is the approach we use.

Define AI-Ready Contact Flows from Week One

Amazon Connect contact flows are the orchestration layer for both human and AI interactions. When you design flows with AI agents in mind, you build in the branching logic that routes calls to an AI agent first, with deterministic fallback to a human queue based on intent confidence, call type, or customer segment.

The key architectural decision is where the AI agent sits in the flow. We position AI agents as the first point of contact for defined intent categories, typically: appointment scheduling, account balance enquiries, payment processing, and status checks. These categories typically represent 35 to 55 percent of inbound call volume in UK financial services and utilities contact centres.

Use Amazon Lex for Intent Classification, AWS Lambda for Orchestration

Amazon Lex handles the speech recognition and intent classification layer. AWS Lambda functions handle the business logic: CRM lookups, authentication, data writes, and handoff payloads to human agents. This architecture keeps the AI agent stateless and auditable, which matters for FCA and ICO compliance.

A production AI voice agent for a single intent category, for example payment arrangement scheduling, can be built and tested in 2 to 3 weeks alongside the core migration work. It does not require the migration to be complete first.

Build Authentication Once, Use It Everywhere

Customer authentication is the most common bottleneck in AI voice agent deployments. UK contact centres typically authenticate via date of birth, postcode, account number, or a combination. This logic needs to be built once as a reusable Lambda function and invoked consistently across both AI-handled and human-handled calls.

Building authentication as a shared service rather than duplicating it in each contact flow reduces build time by 30 to 40 percent and ensures consistent compliance behaviour across all interaction types.

Set Measurable Targets Before Go-Live

AI voice agents without defined success metrics drift. Before go-live, define:

• Target containment rate per intent category (a realistic starting target for a well-scoped intent is 60 to 75 percent)
• Maximum acceptable handle time for AI-handled calls
• Escalation rate threshold that triggers a flow review
• Customer satisfaction score baseline for AI-handled versus human-handled calls

These targets should be agreed with operations leadership before the first call is handled, not after the first month of data.

Contact Lens from Day One

Enable Amazon Connect Contact Lens at go-live, not as a phase two activity. Contact Lens provides post-call transcription, sentiment analysis, and automated evaluation against custom rubrics. For UK regulated industries, it also supports compliance category detection, flagging calls that contain specific phrases relevant to FCA Consumer Duty obligations or vulnerability indicators.

The data Contact Lens generates in the first 30 days of operation is the foundation for optimising AI agent performance. Starting it at go-live means you have 30 days of baseline data before you begin your first optimisation sprint.

A Realistic 4 to 6 Week Migration Timeline

• Audit existing Avaya call flows and categorise for retain, redesign, or retire
• Define AI agent scope and target intent categories
• Complete compliance architecture review (PCI DSS, UK GDPR, FCA recording)
• Submit number port orders
• Design agent desktop integration with CRM

• Build Amazon Connect contact flows for all retained and redesigned categories
• Build authentication Lambda and CRM integration
• Build AI voice agent for priority intent category in parallel
• Configure Contact Lens and reporting dashboards

• UAT with real agents across all call types
• Load testing at 120 percent of peak concurrent volume
• AI agent testing with edge cases and escalation scenarios
• Supervisor training on real-time analytics
• Phased cutover: start with one queue or one site, full cutover by end of week 6

This timeline is achievable for contact centres with up to 300 seats and up to 40 distinct call flow types. Larger environments require additional discovery time but the build and test phases scale proportionally.

What Compliance-Ready Looks Like on Amazon Connect

For UK regulated industries, a compliant Amazon Connect deployment includes:

• All data processed and stored in eu-west-2 (London) with no cross-region transfer without explicit data processing agreements
• PCI DSS pause-and-resume recording implemented in all payment-related flows
• Consent capture and recording notification played before call recording commences, logged with timestamp in the contact record
• Contact Lens compliance categories configured for FCA Consumer Duty vulnerability indicators
• IAM roles following least-privilege principles with CloudTrail audit logging enabled
• Encryption at rest and in transit for all contact records and recordings using AWS KMS with customer-managed keys

These are not optional enhancements. They are the baseline for operating a contact centre in UK financial services or utilities on any cloud platform.

The Build vs. Buy Decision for AI Voice Agents

Some UK contact centres evaluate third-party AI voice platforms alongside Amazon Connect's native capabilities. The calculus is straightforward for AWS-native environments:

Third-party AI voice platforms introduce an additional integration layer between the telephony platform and the AI engine. That integration layer adds latency (typically 200 to 400ms per turn), adds a data transfer path that requires its own compliance review, and adds a commercial dependency.

Amazon Connect with native Lex and Lambda has sub-100ms response latency for intent classification, all data stays within the AWS account boundary, and there is a single commercial relationship. For UK regulated industries where data residency and audit trails are non-negotiable, the native architecture is the correct choice.

What to Expect After Go-Live

A well-executed Avaya to Amazon Connect migration with AI voice agents deployed at go-live typically delivers the following outcomes within 90 days:

• 35 to 55 percent of inbound calls handled end-to-end by AI agents without human intervention
• Average handle time for AI-handled calls 40 to 60 percent lower than equivalent human-handled calls
• Cost per contact reduction of 25 to 45 percent for AI-handled intent categories
• First call resolution improvement of 8 to 15 percentage points driven by consistent AI handling and better agent screen pops
• Supervisor time on manual quality evaluation reduced by 60 to 70 percent through Contact Lens automation

These are outcomes from production deployments in UK regulated industries, not projections.

Ready to Migrate?

We build Amazon Connect environments and production AI voice agents for UK contact centres in 4 to 6 weeks. We do not advise on migrations. We execute them.

If you are running Avaya today and want a clear picture of what your migration timeline, compliance architecture, and AI agent scope look like, start with a discovery call.