How DiDi built intelligent contact center QA with Amazon Bedrock

{“title”: “DiDi Builds Intelligent Contact Center QA System Using Amazon Bedrock”, “slug”: “didi-contact-center-qa-bedrock”, “content”: “

DiDi International Business Group turns to AWS for AI-powered quality assurance

DiDi’s International Business Group has partnered with Amazon Web Services (AWS) to build an intelligent contact center quality assurance (QA) system, the company has revealed in a new case study. The system, running on Amazon Bedrock, marks a significant step in the adoption of generative AI to automate and improve the critical task of evaluating customer service interactions at scale.

The move addresses a long-standing challenge for global enterprises: manual QA reviews of thousands of agent conversations across multiple languages are slow, expensive, and often inconsistent. By infusing AI into the process, DiDi aims to make quality monitoring faster, fairer, and more insightful.

Why manual QA no longer scales

DiDi’s international operations span ride‑hailing, food delivery, and financial services in markets including Latin America, Australia, and Japan. Each day, agents handle a high volume of text chats, voice calls, and emails in numerous languages. Traditional QA teams typically review only a tiny fraction of these interactions, leaving gaps in oversight and coaching.

“The need for consistent, multilanguage evaluation pushed us to look for an AI‑driven approach,” a DiDi spokesperson said, summarizing the motivation behind the project. The team wanted a system that could not only score interactions automatically but also understand context, tone, and compliance nuances across different cultures.

Inside the Amazon Bedrock-powered QA engine

The new QA system, detailed on the AWS Machine Learning Blog, uses foundation models accessed through Amazon Bedrock to analyze agent‑customer conversations. Bedrock provides a serverless environment where DiDi can choose from a range of large language models – including those from Anthropic, Meta, and Amazon – and customize them with its own data without managing infrastructure.

The system evaluates interactions against a set of predefined quality criteria, such as:

• Empathy and tone: Did the agent show understanding and maintain a professional demeanor?
• Problem resolution: Was the customer’s issue resolved effectively and within policy?
• Script adherence: Did the agent follow required disclosures and workflows?
• Compliance: Were all regulatory and data privacy rules observed?

Generative AI models score each interaction and generate a natural language summary explaining the rationale behind the score. This feedback is then routed to team leaders, who can review the AI’s findings and take corrective action when necessary. The result is a human‑in‑the‑loop workflow that combines AI speed with human judgment.

Multilingual capabilities and model flexibility

One of the key advantages of using Amazon Bedrock is the ability to tap into models that already support dozens of languages. DiDi fine‑tuned selected models on its own historical QA data, teaching the system to recognize company‑specific terminology and service scenarios. The platform can switch between models depending on the language or task, ensuring optimal performance across the diverse markets DiDi serves.

Because Bedrock is fully managed, DiDi’s engineering team avoided the heavy lifting of setting up GPU clusters and maintaining model endpoints. This allowed them to focus on data preparation, prompt engineering, and integration with existing contact center software.

Measurable benefits and operational gains

According to the AWS case study, DiDi has already observed several operational improvements:

• Faster QA cycles: Automated scoring cuts the time needed to review each interaction by as much as 70%, allowing the QA team to cover a far larger sample of conversations.
• Consistent evaluations: The AI applies the same standards across all agents and languages, eliminating the variability that comes with different human reviewers.
• Scalability: As DiDi expands into new countries, the system can be scaled immediately without hiring and training additional QA staff.
• Actionable insights: Aggregated data from the AI highlights common agent mistakes and emerging customer issues, enabling targeted coaching and process improvements.

These outcomes align with broader industry trends, where contact centers are rapidly adopting generative AI for tasks ranging from real‑time agent assistance to post‑interaction analytics. DiDi’s deployment demonstrates that the technology is moving beyond pilots into production‑grade, enterprise‑wide implementations.

A deepening partnership with AWS

DiDi has been a significant AWS customer for years, relying on the cloud giant for core infrastructure. The intelligent QA project deepens that relationship, positioning AWS as a strategic partner for DiDi’s AI ambitions. While the current focus is on the International Business Group, the blueprint could eventually be extended to DiDi’s massive domestic operations in China, where the company processes millions of rides daily.

The case study does not disclose financial terms, but it underscores how cloud providers are competing to embed AI services into the operational fabric of global enterprises. For DiDi, the project is a clear signal that it intends to lead, not follow, in the use of AI to elevate customer experience quality.

“, “category_name”: