Normalized claim
Alert volume reduction: 25% decrease
alert volume dropped 25%
Swann provides Generative AI to millions of IoT Devices using Amazon Bedrock
Use case typeAI platformUpdated Jun 13, 2026
Swann Communications uses Amazon Bedrock to power a generative-AI notification filtering system for its global DIY home security platform. The solution helps distinguish meaningful security events from false positives such as pets, cars, and delivery people, so customers receive fewer but more relevant alerts across millions of connected IoT devices. Swann combines edge pre-filtering, tiered model routing, and AWS serverless and storage services to scale alert processing with low latency and lower cost.
- Organization
- Swann Communications
- Industry
- Consumer & Food
- Location
- Australia
- Published
- February 2026
Reported outcomes
−99.7%
Cost reductionCost savings
−25%Alert volume reduction+89%Notification relevance increase+3%Customer satisfaction increase2-17%Bedrock API call reduction18-150%Token usage reduction
Strategic outcomes
Customer experience & trustReduced alert fatigue with more relevant notificationsNew product / capabilityBuilt AI-powered notification filtering systemScale & capacityEnabled alert processing across millions of devicesCost efficiencyLowered cost through model routing
Catalog median for cost savings deployments: −40% across 177 reported metrics. Compare benchmarks →
Why do we believe this?Outcome claims, sources, and evidence checks
Normalized claim
Notification relevance increase: 89% increase
notification relevance increased 89%
Normalized claim
Customer satisfaction increase: 3% increase
customer satisfaction increased by 3%
Normalized claim
Bedrock API call reduction: 2-17% decrease
reduced API calls by 88% (17,000 to 2,000 RPM)
Normalized claim
Token usage reduction: 18-150% decrease
88% token reduction (from 150 to 18 tokens per request)
Normalized claim
Cost reduction: 99.7% decrease
reduced costs by 99.7% compared to using Claude Sonnet for all requests
Normalized claim
Scale of deployment: 11.7 million devices increase
global network of more than 11.74 million connected devices
Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks
- Customer
- Swann Communications
- Provider
- AWS
- Maturity
- Scaled Production
- Linked source
- AWS Machine Learning Blog
1 million to about $6,000, with sub-300 ms p95 latency at scale
Customer identity supportedSource describes one deploymentMaturity supported
Primary read
Use case focus
Showing 3 of 3
- 1Generative AI notification filtering
- 2IoT alert prioritization
- 3Event classification and routing
- Alert fatigue and false positives overwhelmed customers.
- Approximately 20 daily notifications per camera caused users to ignore or disable alerts, risking missed security events.
- Swann needed intelligent filtering that could scale across more than 11.74 million connected devices.
- Swann built a multi-model generative AI notification system on Amazon Bedrock.
- The architecture routes requests across Amazon Nova Lite, Amazon Nova Pro, Claude Haiku, and Claude Sonnet based on task complexity and latency needs.
- Swann added edge pre-filtering, including motion, zone-based analysis, duplicate frame elimination, and a custom GPU-based pre-filtering model on Amazon EC2.
- AWS IoT Core, Amazon S3, Amazon SQS, and AWS Lambda support device connectivity, storage, queuing, and event-driven processing.
- Alert volume dropped 25%.
- Notification relevance increased 89%.
- Customer satisfaction improved 3%.
- Bedrock API calls dropped 88% and token usage dropped 88%.
- Projected monthly cost fell from $2.1 million to about $6,000, with sub-300 ms p95 latency at scale.
Architecture
Smart cameras and doorbells send video through AWS IoT Core into an AWS-based pipeline. Amazon S3 stores video feeds and Amazon SQS buffers requests. AWS Lambda functions invoke Amazon Bedrock and apply model-selection logic. A custom pre-filtering model runs on Amazon EC2 G3/G4 GPU instances to remove obvious false detections before Bedrock inference. The solution uses a tiered model strategy across Amazon Nova Lite, Amazon Nova Pro, Claude Haiku, and Claude Sonnet and is monitored with latency, token, cost, accuracy, and throttling metrics.
Sources & evidence1
Evidence: Medium50/100Evidence strength
- Customer explicitly identified
- Deployment status explicitly supported
- Quantified outcome available
- Technical implementation details available
Type: Blog PostPublished: Feb 11, 2026Publisher: AWSEvidence: VendorConfidence: Medium
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Customer evidence
Provider evidence
Market context
Browse the catalog
Was this useful?
Community
Comments
No published comments yet.