ChinaTechnology and telecom2026 – ongoingDiDi International Business Group
Contact-center quality assurance on Amazon Bedrock
AWS reported DiDi's contact-reason verification accuracy rose from 38% to 86%
Problem
DiDi's international customer-service team used an opaque third-party quality-assurance system. Changing service rules made consistent multilingual review difficult.
Approach
DiDi and AWS built a replacement on Amazon Bedrock for Spanish and Portuguese conversations. The system checked contact reasons and evaluated compliance, with a separate workflow for recurring customer issues. It verified the assigned contact reason before exposing the model to alternative classifications. Programmatic checks reviewed deterministic criteria after model scoring.
Results
Takeaway
Separating label verification from reclassification improved accuracy in DiDi's production validation. The disclosed results did not establish performance on an independent benchmark.
Categories
- Size
- Large enterprise
- Industry
- Technology and telecom
- Business function
- Customer service
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
AWS described DiDi's production QA architecture and its context-management changes. It reported intent-verification accuracy rising from 38% to 86% during production validation.
Full details are in the original sources.