DTDC
DTDC upgrades its DIVA courier chatbot with generative AI and cuts support-team queries by 51.4%
DTDC, an Indian courier network, replaced the rigid script of its DIVA chatbot with a generative AI agent on Amazon Bedrock that handles tracking, rates and serviceability questions. The AWS post, written with DTDC staff, reports 93% response accuracy and 51.4% fewer queries handled by the support team.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
over 400,000
Customer queries DTDC Express receives each month, across all channels1
93%
Response accuracy of DIVA 2.0, as reported in the post1
51.4%
Reduction in queries handled by the customer support team, as reported in the post1
51.4%
(131,530)
Consignment inquiries that did not result in a support ticket, last 3 months of dashboard data1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
DTDC Express receives over 400,000 customer queries a month, covering tracking, serviceability checks and shipping rates. Its existing assistant, DIVA, ran on a rigid guided workflow, so users had to follow a set path instead of asking in their own words. That put more load on customer support, lengthened resolution times and hurt the customer experience.
What they built
DTDC and its AWS partner ShellKode built DIVA 2.0, a conversational agent on the DTDC website. Customers ask in natural language about tracking, rates, service availability or general questions. The agent works out what the customer wants, calls DTDC's own systems for tracking, pricing and delivery coverage, and can create a support ticket. A large language model, Anthropic's Claude 3.0 on Amazon Bedrock, turns the data into a reply.
How they did it
For general questions the agent searches a knowledge base built from the DTDC website, internal support documents, FAQs and operational data. If it finds nothing relevant, it returns a preset message saying it cannot help. Queries and answers are logged to a dashboard open to DTDC's admin team, showing accuracy, unresolved queries and peak times. The post lists the main challenges as connecting real-time data from legacy systems, handling logistics terms and multi-step queries, and moving off the old DIVA without a service gap.
Results
The post reports response accuracy of 93% and says the support team now handles 51.4% fewer queries. Over the last three months of dashboard data, 51.4% of 256,048 consignment inquiries did not need a support ticket. DTDC says support staff can now focus on more critical issues.
Lessons for your project
Replace scripted menus with an agent that can call your live systems. DIVA 2.0 checks tracking, pricing and service-area systems and can open a support ticket.
Give the bot a safe fallback. When the knowledge base has no answer, DIVA 2.0 says it cannot help instead of guessing.
Build a dashboard from day one. DTDC tracks accuracy, unresolved queries and peak times, which shows where to improve.
Expect to keep human agents. The post says the agent has not removed the need for customer support, and many conversations still involve both.
Caveats
- The source is an AWS blog post co-written by DTDC staff and the integrator ShellKode, so the figures are vendor-reported. No independent source confirms them.
- The post does not say how the 93% accuracy was measured, over which period, or the baseline for the 51.4% fall in support-team queries. The same 51.4% also appears as the share of consignment inquiries with no ticket, so the two figures may be linked.
- The post gives no go-live date. The deployed date shown is the post date.
- The post names Anthropic's Claude 3.0 on Amazon Bedrock. The 400,000 figure is all monthly queries to DTDC, not only those the agent handles.
Sources
- The DIVA logistics agent, powered by Amazon Bedrock
AWS Machine Learning Blog · 2025-08-07 · Vendor case study · aws.amazon.com
Summarized in our own words from the sources above. Spotted an error? See corrections.