Customer support · Production

DTDC builds DIVA 2.0 with Amazon Bedrock, reducing customer support queries by 51.4%

The problem

DTDC's existing logistics agent DIVA handled over 400,000 monthly queries through a rigid, guided workflow that forced users along a structured path, leading to longer resolution times, greater burden on human support agents, and poor customer experience.

First attempt

The original DIVA system used a rigid, guided workflow that lacked flexibility and could not handle dynamic conversations, creating poor customer experience and increased reliance on human agents.

Workflow diagram · grounded in source
1
User submits natural language query
trigger
“An end-user accesses the logistics agent through the DTDC website and submits queries like tracking shipments, checking service availability, calculating shipping rates, FAQs, and so on using natural language”
2
Bedrock Agent interprets intent
ai_action
“An Amazon Bedrock agent receives the request and interprets the user's intent using its natural language understanding capabilities”
3
Route to Lambda function
routing
“Based on the interpreted intent, the agent triggers an appropriate Lambda function, such as: Tracking consignments, Pricing information, Location serviceability check, Support ticket creation”
4
API data retrieval
integration
“The triggered Lambda function calls the following client APIs, retrieves the relevant data, and returns the response to the agent”
5
Knowledge base semantic search
ai_action
“Using semantic similarity search, relevant chunks of information are retrieved from the knowledge base based on the user's query”
6
LLM generates contextual response
ai_action
“the agent passes the response to the large language model (LLM), in this case Anthropic's Claude 3.0 on Amazon Bedrock, which understands the context of the retrieved data, processes it, and generates a meaningful response for the user”
7
Dashboard logs and improvement
feedback_loop
“Real-time data is logged and analyzed on the dashboard, enabling continuous improvement and quick issue resolution”
Reported outcome

DIVA 2.0 achieved 93% response accuracy and reduced the volume of queries handled by the customer support team by 51.4%, with 51.4% of consignment inquiries resolved without requiring a support ticket.

Reported metrics
Response accuracy93%
Customer support query reduction51.4%
Consignment inquiries not resulting in support ticket51.4%
Consignment inquiry share of total71%
Show all 7 reported metrics
response accuracy93%
customer support query reduction51.4%
consignment inquiries not resulting in support ticket51.4%
consignment inquiry share of total71%
general inquiry share of total29.5%
monthly customer queries volumeover 400,000
consignment inquiries resulting in ticket creation48.6%
Reported stack
Amazon BedrockAmazon Bedrock AgentsAmazon Bedrock Knowledge BasesAmazon Bedrock GuardrailsAWS App RunnerAWS LambdaAmazon CloudFrontAmazon S3Amazon OpenSearch ServiceAmazon RDSAmazon CloudWatch LogsAWS CloudTrailAmazon GuardDutyAmazon API GatewayAWS Identity and Access ManagementClaude 3.0
Source
https://aws.amazon.com/blogs/machine-learning/the-diva-logistics-agent-powered-by-amazon-bedrock?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

DIVA 2.0 achieved 93% response accuracy and reduced the volume of queries handled by the customer support team by 51.4%, with 51.4% of consignment inquiries resolved without requiring a support ticket.

What tools did this team use?

Amazon Bedrock, Amazon Bedrock Agents, Amazon Bedrock Knowledge Bases, Amazon Bedrock Guardrails, AWS App Runner, AWS Lambda, Amazon CloudFront, Amazon S3, Amazon OpenSearch Service, Amazon RDS.

What results were reported?

Response accuracy: 93%; Customer support query reduction: 51.4%; Consignment inquiries not resulting in support ticket: 51.4%; Consignment inquiry share of total: 71% (source-reported, not independently verified).

What failed first in this deployment?

The original DIVA system used a rigid, guided workflow that lacked flexibility and could not handle dynamic conversations, creating poor customer experience and increased reliance on human agents.

How is this customer support AI workflow structured?

User submits natural language query → Bedrock Agent interprets intent → Route to Lambda function → API data retrieval → Knowledge base semantic search → LLM generates contextual response → Dashboard logs and improvement.

Related customer support cases