A state benefits contact center, moved to Amazon Connect
A U.S. state health and human services agency runs high-volume contact centers for residents applying for and managing public assistance, including income maintenance and child care assistance. Its legacy, on-premises platform could not scale with demand, and every IVR, routing or language change was manual. Working with AWS, we moved it to Amazon Connect, with voice bots in English and Spanish, routing by ZIP code, direct agent extensions and emergency closures. All of it uses standard Amazon Connect features, with as little custom code as possible.
- 9–11%
- abandonment, down from 18–22%
- −18–20%
- average handle time
- 2
- languages, English and Spanish, in parity
- 3
- waves: foundation, core, enhanced
- A state benefits contact center moved from a legacy on-premises platform to Amazon Connect, delivered with AWS.
- Abandonment fell from 18–22% to 9–11%; handle time fell 18–20%.
- English and Spanish voice bots on Amazon Lex, ZIP-code routing and emergency closures.
- About half as many residents hang up before reaching someone.
- Calls are 18–20% shorter, so the same staff serve more residents.
- Built on out-of-the-box features, with as little custom code as possible.
- Child care callers reach the right team by ZIP code; others reach a worker directly by extension.
- Real-time and historical reports and conversational analytics views for supervisors.
- Emergency closures switched on at will.
- Amazon Lex main-menu bots in English and Spanish, with no back-end lookups; Spanish flows mirror the English structure.
- Direct extension lookup with redirect to the existing voicemail, and ZIP-code routing in contact flows.
- IAM and single sign-on, CloudWatch, and promotion from development to test to production.
One call, from both sides
On the left, what the resident hears. On the right, the contact flow: language, the Lex voice bot’s intent, and where the call goes. Pick a call to see ZIP-code routing in Spanish, a worker’s extension or an emergency closure.
- Call comes in
- Pick language
- Understand need
- Choose team
- Connect
- Language
- —
- What they need
- —
- ZIP code
- —
- Goes to
- —
- Waiting for a call
- Bot questions
- 0
- Transfers
- 0
- Wait
- —
- Result
- …
Illustrative calls. Residents, ZIP codes and extensions are fictional; the flow follows the production design: Lex voice bots in English and Spanish, ZIP-code routing, direct extensions with voicemail, and emergency closures.
The challenge
Residents call about benefits they depend on, and demand is not steady. It jumps with policy changes, enrollment cycles and emergencies. The legacy platform could not scale with it, and its single-site, infrastructure-bound design meant an outage took the service down.
Change was slow. IVR updates, routing changes and language support were all manual, with long change cycles. Keeping the English and Spanish experiences the same meant doing every change twice by hand, and reporting was fragmented, so leaders had little real-time view of performance.
The agency wanted a platform that scales on its own, a consistent experience in both languages, faster changes, and a foundation for self-service and digital channels later, while meeting government security and governance standards.
What made this migration different
The best public-sector migration is the one residents do not notice. That meant keeping it simple.
- 9–11%
of callers hang up before reaching someone, down from 18–22%.
Automatic scaling, clearer menus, ZIP-code routing and emergency closure messages all cut the number of callers who give up.
One call design for both English and Spanish.
Spanish flows reuse the English structure; only prompts and bot language change. A parity check is part of every change review.
Built on standard Amazon Connect features wherever possible.
Standard reporting, conversational analytics views and the standard agent workspace, instead of custom pipelines and screens. Faster to deliver, easier to support.
Delivered with AWS, under public-sector governance.
Approval cycles were built into the sprint plan, so governance shaped the schedule instead of stalling it.
What changed
- Main menuBefore: A fixed menu of key pressesAfter: An Amazon Lex voice bot that understands what callers say
- SpanishBefore: Every change made twice, by hand, and drifting apartAfter: Spanish flows mirror the English ones; only prompts and bot language differ
- Child care callsBefore: Calls routed without locationAfter: The caller’s 5-digit ZIP code routes them to the right queue or agent
- Reaching a personBefore: Transfers to find the right workerAfter: Direct agent virtual extensions, with voicemail when they are away
- ClosuresBefore: Hard to react to an emergency closureAfter: Emergency closure messages switched on at will, without infrastructure changes
- ScaleBefore: Fixed capacity; peaks meant long waitsAfter: Scales automatically with demand
- ReportingBefore: Fragmented, little real-time viewAfter: Real-time and historical reports, plus conversational analytics views
- AccessBefore: Fragmented access modelsAfter: Single sign-on, role-based access in IAM, CloudWatch logging
Hover or tap a tile to see what it replaced.
More residents get through, and calls end sooner.
The agency and Digiclarity agreed two measures before go-live and tracked them before and after.
- Callers who hang up before an agentBefore18–22%After9–11%
- Average handle timeBefore8.5–9 minAfter6.8–7.2 min
- English and Spanish journeysBeforeMaintained twiceAfterOne design, two languages
From Amazon Connect historical and real-time reports; abandonment tracked daily, averaged monthly, at peak periods for the baseline.
How a resident’s call is handled now
The resident says what they need. The voice bot works out where to send them, in English or Spanish, and the call reaches the right team.
- 01 · CallA resident callsIf the office is closed in an emergency, they hear that straight away.
- 02 · LanguageEnglish or SpanishThe same journey in both languages.
- 03 · UnderstandThe voice bot listensAn Amazon Lex bot recognizes what the caller needs from a short set of intents.
- 04 · RouteTo the right placeChild care callers enter a ZIP code; callers with a worker’s extension reach that person or their voicemail.
- Income maintenance
- Child care assistance
- Direct extension
- Voicemail
- 05 · ResolveAn agent picks upIn the Amazon Connect Agent Workspace, with real-time reporting for supervisors.
The steps in every wave
The same 5 steps for each wave.
- 1PlanEntry criteriaAgree what must be true before the wave starts.
- 2BuildBuildFoundation: access, single sign-on, monitoring. Core: bots, flows, queues. Enhanced: ZIP routing, extensions, voicemail, dashboards.
- 3ProveTestTesting checkpoints, then customer-led UAT.
- 4TrainPrepare peopleAgent and supervisor readiness confirmed before cutover.
- 5Go liveGo/No-Go and cut overApproved maintenance window, then post-cutover validation and sign-off.
The outcome
- Call abandonment down from 18–22% to 9–11%
- Average handle time down 18–20%, from 8.5–9 to 6.8–7.2 minutes
- English and Spanish voice bots on Amazon Lex, with parity between them
- ZIP-code routing for child care assistance, and direct agent extensions with voicemail
- Emergency closure messages the agency can switch on at will
- Real-time and historical reporting and conversational analytics views, with no custom data pipelines
- Runbook, knowledge transfer and promotion steps from development to test to production
- Contact center
- Amazon Connect
- Agent Workspace
- Contact flows
- AI
- Amazon Lex voice bots, English and Spanish
- Reporting
- Real-time and historical reports
- Conversational analytics views
- Security and operations
- AWS IAM
- Single sign-on
- Amazon CloudWatch
Five lessons came out of this one. Multilingual parity has to be engineered with shared templates, not copied by hand. Reusing a legacy voicemail system was quick but created a dependency we now map during discovery. Reporting expectations need agreeing early, out of the box first and custom later. Mixing infrastructure as code with console changes causes drift, so promotion checklists are now standard. And public-sector approvals belong in the plan from day one.
A partner AWS has validated, on exactly this kind of work.
- AWS Advanced Tier Services PartnerRecognized by AWS for proven delivery, certified people and customer outcomes.
- AWS Migration Services CompetencyValidated by AWS for moving and modernizing workloads on AWS at scale.
- Amazon Connect Delivery PartnerValidated by AWS to deliver Amazon Connect, including migrations from Avaya, Genesys, Cisco and other platforms.
- AWS AI Services CompetencyValidated by AWS for building AI into real products, including the contact center.
- AWS Experience-Based Acceleration (EBA) partnerWe run EBA engagements with AWS to speed up migrations with hands-on, time-boxed sprints.
- ISO/IEC 27001 certifiedOur information security management system is independently certified to the international standard.
- SOC 2 Type IIIndependent auditors have tested our security controls over time, not just at a single point.
More work
All case studies →- Amazon Connect
- Amazon Lex
- AWS Lambda
- Amazon Connect
- Contact flows
- AWS Lambda
- Amazon EKS
- Amazon Kinesis
- Amazon DynamoDB








