Skip to content

Classifying Emails Like Never Before using Amazon Bedrock

Discover how we used Generative AI and AWS to classify thousands of emails in real time, improve accuracy, and eliminate manual triage through serverless automation.

aws-bedrock-3

Client Overview

The client is an Event Services company that understands the power of digital platforms when it comes to building and maintaining community engagement. They combine deep knowledge of events, technology, marketing, and operations to deliver strong, well-rounded projects. They are using a tech stack built specifically for this kind of work, backed by proven industry processes.

customer (1)

Challenge

Manually sorting through thousands of daily emails was slowing the business down. Critical messages risked being missed or delayed, costing valuable opportunities and affecting customer satisfaction.

The client needed a secure, automated, and intelligent way to classify incoming emails, prioritise urgent issues, and reduce operational overhead.

Additional key challenges included:

Data Protection & Privacy

Given that emails contain sensitive personal information, safeguarding customer data is paramount. This requires robust encryption and strict access controls to ensure data security at all times.

Integration with Existing Systems

The classification solution needed to fit seamlessly into the client’s existing email and workflow infrastructure without disrupting ongoing operations.

Classification & Data Extraction

A GenAI-powered application should scan and classify emails while simultaneously extracting valuable additional information.

Scalability & Cost Efficiency

Receiving tens of thousands of emails daily necessitates a scalable and cost-effective storage solution.

Solution

Bion Consulting delivered a Generative AI-enabled serverless data pipeline on AWS, processing tens of thousands of emails in near-real time.

Generative AI Enablement
  • LLM Integration via AWS Bedrock: A tailored Large Language Model (LLM) is deployed using Amazon Bedrock.
  • These models process incoming emails, extract valuable information, and classify them into relevant categories.
Scalable Serverless Infrastructure
  • Incoming emails are received through Amazon SES and processed using AWS Lambda functions.
  • The classification results are then stored in Amazon DynamoDB for fast and scalable access.
Near Real-time Data Processing and Workflows
  • Emails are classified immediately upon arrival, triggering the appropriate workflows without delay.
  • If an email is marked as urgent, an AWS Step Functions workflow is initiated to notify the relevant team and ensure prompt action is taken.
Reporting and Visualisation
  • Data extracted by the LLMs, along with processing metrics—including LLM usage and cost—are stored in AWS Athena tables within the data lake.
  • These tables power Amazon QuickSight dashboards, enabling users to explore the data and generate insights using natural language queries through the QuickSight Q feature.

Results

Implementing the GenAI-powered email classification pipeline delivered immediate and measurable value to the client.
024-project

Increased Operational Efficiency

Manual triage of emails that previously consumed hours of staff time each day has been reduced to near zero. The serverless pipeline processes and classifies incoming emails within seconds of arrival, ensuring timely routing and faster issue resolution.


024-project

Improved Accuracy and Prioritisation

The custom LLM model achieved over 95% accuracy in email classification, significantly reducing the risk of human error. Critical emails are now flagged and routed to the correct teams without delay, preventing potential revenue loss and boosting customer satisfaction.

026-data network

Scalability Without Overhead

Thanks to the serverless architecture, the solution scales effortlessly to meet demand, even on peak days when email volume surges. This eliminates the need for constant infrastructure management, reducing both operational complexity and cost.

010-monitor

Actionable Insights Through Visualisation

The integration with Amazon QuickSight empowers business users with real-time dashboards and natural language querying, making it easy to track response times, monitor LLM usage, and identify trends. This data-driven approach helps the client continuously optimise their workflows.

Technology Stack

To build the end-to-end pipelines, the following technologies were used:
  • AI/ML: AWS Bedrock, Amazon QuickSight Q
  • Email Management: AWS SES
  • Workflow Management: AWS Step Functions
  • Data Storage: AWS S3, AWS DynamoDB
  • Data Processing: AWS Lambda (Python), AWS Athena, AWS Lake Formation
  • Visualisation: Amazon QuickSight
  • Monitoring: CloudWatch
  • Infrastructure-as-Code: Terraform
01