How we achieved 93% reduction in manual dialogue management by modernizing an elderly care chatbot with Amazon Bedrock

The customer turned to zeb, an AWS Premier Tier Consulting Partner, to implement a modern, cloud-native solution. Our team of experts utilized AWS's scalable, secure, and cost-effective architecture to rebuild SimpleC's legacy rule-based chatbot into a generative AI-powered platform using Amazon Bedrock, enabling real-time personalization, improved user satisfaction, and streamlined operational efficiency.

Elderly users often face barriers in engaging with modern digital platforms, complex navigation, fragmented apps, and unintuitive interfaces. SimpleC sought to overcome this by building a natural, conversational interface that felt intuitive and inclusive for older adults. Their legacy rule-based system required frequent manual intervention and couldn't accommodate open-ended or personalized interactions.

Key Challenges:

• High effort and cost in managing scripted dialogues manually
• Limited flexibility to handle unstructured or unexpected queries
• Absence of real-time personalization or context-based suggestions
• Fragmented user experience requiring multiple interfaces

Our customer aimed to modernize their platform with the following goals:

• Conversational Intelligence: Enable open-ended dialogue using generative AI for more natural interactions.
• Operational Efficiency: Reduce manual content management and simplify system maintenance.
• Personalized Care: Support proactive recommendations like video calls or health reminders tailored to user behavior.
• Scalability: Ensure a secure, resilient, and auto-scaling backend to accommodate growing user adoption.

The solution was built using a suite of AWS services, architected in line with the AWS Well-Architected Framework to ensure security, performance, and long-term cost optimization.

Compute & Orchestration

• Amazon ECS with Fargate – Containerized backend processes (e.g., web scraping, context enrichment) with auto-scaling and no server management

API Management & Integration

• Amazon Application Load balancer – Secure, scalable endpoints to expose chatbot functions to web and mobile clients
• Amazon Bedrock – Foundation model integration (e.g., Anthropic, Amazon Titan) for real-time, generative responses without managing custom ML infrastructure

Identity & Access Management

• Amazon Cognito – Elder-friendly authentication with passwordless login and secure session handling
• IAM Roles & Policies – Fine-grained access controls to enforce least privilege across Lambda, ECS, and API layers

Data Storage & User Context

• Amazon S3 – Storage for fallback content, templates, and program assets
• Amazon RDS – Real-time database for user sessions, preferences, and interaction histories

Observability & Monitoring

• Amazon CloudWatch – Monitoring across ECS and Lambda workloads, with alarms for latency, health checks, and content delivery

Timeline

The project was completed within 2 months, including research and development tailored specifically for elderly user needs. Infrastructure was deployed using Terraform for consistency across dev, staging, and production environments. The chatbot underwent phased testing with real users, incorporating accessibility features like large font modes and slower pacing options.

Operational KPIs

• Dialogue Automation: Reduced manual content handling by 93%
• Maintenance Efficiency: Saved over 120 hours/month on updates and manual configurations
• Faster Rollouts: Automated CI/CD pipeline enabled weekly feature releases with zero downtime

Engagement KPIs

• User Retention: Achieved a 68% increase in daily usage among senior users
• Trust & Accuracy: Real-time responses improved user trust in the system
• Cognitive Simplicity: Reduced interface fatigue by unifying weather, news, wellness, and reminders into one conversational flow

Security KPIs

• User Protection: Secure, federated identity with Cognito and IAM minimized risk of unauthorized access
• Blast Radius Control: Isolated permissions prevented misconfigurations from affecting other services

Cost & Scalability KPIs

• Auto-Scaling: ECS and Lambda auto-scaled based on load, keeping compute costs aligned with usage
  Cost Control: Budgets and alarms for high-volume components (e.g., Bedrock usage, scraping engines) kept cloud spend within thresholds

Qualitative KPIs

• Technology Adoption: Elderly users reported reduced tech hesitancy and higher satisfaction
• Proactive Suggestions: Family video calls and wellness reminders increased emotional well-being
• Human-Centric Design: User-tested interface created a comforting, non-intrusive chat experience

Scalability and Automation

• Serverless Architecture: Enabled dynamic scaling and pay-as-you-go cost optimization
• CI/CD Pipelines: Automated deployments reduced manual intervention and accelerated delivery
• Infrastructure as Code: Used Terraform to maintain repeatable, version-controlled infrastructure
• Resilience: Designed with auto-recovery for ECS tasks and graceful failure handling in Lambda

Lessons learned

• Simplicity in both conversation and UI is critical for effective engagement with elderly users.
• AI interactions must feel natural and intuitive, prioritizing clarity over complexity.
• Real-time and accurate responses are essential to establish and maintain user trust.
• Early testing with elderly users is necessary to design accessible features like larger text, relevant prompts, and appropriate response pacing.

By reimagining their elderly care platform with AWS generative AI services, SimpleC transformed their chatbot into a reliable digital companion that understands, assists, and empowers senior users. Powered by Amazon Bedrock, the new solution enabled 93% automation, increased engagement by 68%, and eased operational overhead.

zeb's cloud-native, user-first implementation addresses immediate challenges while laying a strong foundation for SimpleC's long-term vision of delivering intuitive, accessible technology for the elderly.