Enhancing Voice AI Performance Through Human-Led Multilingual QA 

Discover how DATAmundi delivered human-led multilingual data QA to enhance and increase performance for AI and voice systems for a Fortune 100 ML operations team. 

The Client 

Our client is the centralized machine learning (ML) operations team of a global Fortune 100 technology company, a function that consolidates data annotation, labelling, and quality control requirements from across the client’s AI and voice technology divisions. With a large pool of internal annotation teams and external production vendors, the client relies on specialist data solution partners to maintain the quality and integrity of their datasets before they feed into live machine learning systems. 

DATAmundi is retained as a key data solutions and quality control partner for this program, continuing to solidify a 5+ year relationship with the client. 

The Brief 

The client required quality control services for a series of machine learning data projects, each involving AI-generated utterances produced, annotated, or recorded by internal teams or third-party vendors. Work spanned: 

• Reviewing AI-generated utterances against detailed project guidelines to assess accuracy and compliance. 

• Evaluating recordings and transcriptions for quality, consistency, and fidelity to the client’s specifications. 

• Flagging low-quality or non-compliant outputs for rework and providing structured QC reports in the client’s required format. 

• Managing 40+ languages per project, spanning European, South Asian, Nordic, Middle Eastern, and other language families. 

Work took place within the client’s proprietary annotation platform. As the client’s ML projects were global, this required sourcing and managing a multilingual workforce of linguists and AI contributors. Due to the complexity of the ML projects, each had complex guidelines – approx. 50 pages – with requirements shifting. This meant continually simplifying and updating the workforce. 

Quality underpinned everything with quality benchmarks of 95-97% acceptance rates. DATAmundi held full accountability for quality across 40+ languages (per project), sourcing and verifying from the DATAtalent network to ensure the right talent for each engagement. 

How DATAmundi Met the Client’s Requirements 

DATAmundi addressed each data project through a combination of operational discipline, talent management, and adaptive process design. 

• Every new project began with a structured onboarding session. DATAmundi held a kickoff call with the client to walk through the platform configuration for each specific project. This session was recorded and shared with all contributors as their primary orientation resource, ensuring consistent understanding from day one. 

• DATAmundi’s QA Leads distilled 50-page guidelines into concise working briefs. Contributors received a prioritized set of key checks and decision rules, with full guidelines available for edge cases.  

• DATAmundi built a pre-submission review step into its workflow. Contributors were instructed not to submit completed jobs directly. They notified the QA lead upon completion so the team could review before delivery to the client’s platform.  

• When early quality checks indicated the production vendor’s output was falling short, DATAmundi halted rather than continued. At 20-30% completion, the DATAmundi team raised the issue with the client and requested remediation before proceeding. This proactive quality flagging protected the integrity of the final dataset. 

• DATAmundi built and maintained a contributor talent pool of linguists capable of covering 40+ languages simultaneously. This included less common languages, such as Hebrew, Arabic, Nordic dialects, and South Asian languages. Test jobs were used to assess capability on new tool configurations before full deployment, minimizing rework and ensuring only qualified contributor linguists were assigned to live tasks. 

The Results 

DATAmundi consistently met quality thresholds and delivery expectations, measured by operational metrics: 

The Business Impact 

By combining rigorous process discipline with multilingual expertise, DATAmundi enables the client’s ML operations team to meet its internal quality commitments at scale, on an ongoing basis, without compromising on accuracy or turnaround. 

The engagement also demonstrates DATAmundi’s ability to adapt for custom needs and scale quickly. For ML teams managing large, complex annotation pipelines across multiple vendors, this kind of trusted QC partnership – one that delivers increased ML model performance while absorbing operational complexity – is difficult to replicate. 

Learn more about DATAmundi’s AI Data Services or get in touch to discuss your project.