Conversational AI for Customer Service: 22% Higher Call Handling Success
See how Omilia improved successful conversational AI customer service with smarter data collection and annotation.
This case study shows how Omilia improved conversational AI customer service performance by 22% using annotated multilingual voice data to optimize call handling, IVR success rates and real-time customer interactions.
Our customer: A global leader in conversational AI technology
Omilia is a Gartner-recognized leader in conversational AI solutions, providing voice and chatbot solutions that help enterprises deliver human-like customer service at scale.
The context: Customer service challenges in multilingual conversational AI
As Omilia expanded its reach into global markets, it needed to scale its AI-powered support platform without sacrificing the quality or accuracy of customer interactions. The team faced multiple challenges:
- Supporting over 30 languages (including hard-to-source regional dialects)
- Maintaining conversational naturalness
- Reducing average call handling time by up to 50%
The company also needed to improve the performance of its natural language understanding systems to better capture context, sentiment and intent in real time—all while adhering to strict compliance and data privacy regulations. Scaling this level of AI performance across such a wide range of languages and environments required not just more data but data designed for real-world performance.
The solution: Data collection and annotation for conversational AI services
To meet these demands, Omilia partnered with Defined.ai to build a multilingual, scalable AI infrastructure grounded in high-quality, ethically sourced data. We began by collecting custom data for speech carefully selected to match Omilia’s global customer base. These datasets were designed to reflect phonetic diversity and different accents, ensuring robust speech recognition in both high- and low-resource languages. They also included a wide variety of speakers using industry-relevant terms, which enabled Omilia’s AI to recognize and handle nuanced customer intents.
Key Deliverables:
- Custom speech datasets in four languages, covering 95% of Omilia’s global customer base
- Phonetic diversity and regional accents, like Mandarin tones and global Spanish variants
- A wide range of temporal patterns, speaker demographics and domain-specific jargon for contextual accuracy
- Emotionally annotated data to enhance real-time sentiment and intent detection
Our team worked closely with Omilia to develop language models tuned for conversational AI in healthcare and banking. By integrating terminology and contextual patterns unique to these fields, the models enabled Omilia’s platform to carry out more natural, precise conversations with users, even in highly regulated or complex scenarios.
To further enhance the intelligence of the platform, Defined.ai supplied data annotation that improved real-time sentiment detection and helped fine-tune models to identify customer intent and churn risk. These capabilities proved crucial for delivering responsive, adaptive service during live interactions.
The results: Improved IVR performance and customer experience
Defined.ai delivered measurable improvements across key performance indicators within a tight timeline.
Key Results:
- 15% fewer recognition errors
- 22% higher IVR success rates
- 35,000 utterances per locale
- Eight-week delivery time
- Faster handling and better customer ratings
By addressing the real-world acoustic challenges we faced in multilingual customer service environments, Defined.ai minimized our call times and improved customer satisfaction. — Petr Mizera, Head of Speech Recognition, Omilia
Last but definitely not least, we ensured privacy and regulatory compliance at every step. Our workflows included GDPR-aligned data redaction and support for biometric authentication, enabling Omilia to meet the highest standards of data protection across markets.
Interested to see how Defined.ai could support your company’s AI solutions? Check out our Data Marketplace for data ideas and complete the short form below to discuss your project with one of our experts.