DiDi Hits 86% QA Accuracy Using Amazon Bedrock AI

DiDi has partnered with Amazon Web Services (AWS) to launch a generative AI-powered quality assurance (QA) system for its International Business Group’s Customer Experience (CX) department. The move shifts the company away from rigid third-party tools toward a transparent, self-owned architecture designed to scale across diverse global markets.

According to the AWS Machine Learning Blog, the new system utilizes Amazon Bedrock to power three primary operational pipelines: intent verification, compliance evaluation, and Voice of Customer (VOC) analysis. The implementation covers three of DiDi's core business lines—ride-hailing, food delivery, and financial services—and provides native support for Spanish and Portuguese. The impact on performance was immediate; during production validation, the accuracy of intent verification surged from 38% to 86%.

The Shift to Transparency

DiDi's International Business Group previously relied on a third-party QA solution that the company described as opaque. This lack of transparency hindered the company's ability to fully control its customer experience metrics and refine its quality audits. By migrating to a generative AI-driven approach via Amazon Bedrock, DiDi was able to build a customized technology stack tailored specifically to its regional languages and the unique requirements of its various service lines.

Industry Implications

This transition highlights a growing trend among large-scale mobility and delivery platforms to replace generic auditing software with flexible, internal systems powered by Large Language Models (LLMs). The dramatic increase in intent verification accuracy suggests that foundation models can significantly outperform traditional QA tools in understanding complex customer interactions. For the broader industry, this demonstrates a path toward reducing manual audit overhead while simultaneously increasing the precision of customer service quality checks.

Future Outlook

As DiDi integrates this AI architecture further into its CX operations, the company will likely focus on refining its VOC analysis to derive deeper consumer insights. While the initial production validation shows a massive leap in accuracy, the long-term effect on customer satisfaction scores and operational costs remains the key metric to watch. The success of this deployment serves as a blueprint for other multinational firms seeking to repatriate their data auditing processes from third-party vendors to internal AI stacks.