United StatesTechnology and telecom2025 – ongoingAT&T
Ask AT&T customer-care agents with fine-tuned models
NVIDIA reported up to 40% higher answer accuracy after AT&T trained its AI agents further
Problem
Computing costs and response times were rising. Nearly 10,000 documents changed several times a week, making it difficult to keep model answers accurate.
Approach
AT&T and Quantiphi used NVIDIA NeMo and NIM to prepare data, fine-tune models and evaluate the results repeatedly. Mistral 7B performed best among the models tested. The agents used retrieval to access current documents.
Results
Takeaway
AT&T combined model fine-tuning with repeated evaluation and retrieval of current documents.
Categories
- Size
- Large enterprise
- Industry
- Technology and telecom
- Business function
- Customer serviceIT and R&D
- Innovation type
- Process innovation
- AI technology
- AI agentsGenerative AI
- Adoption
- Co-developed with a partner
- Evidence
- Vendor's case study
Sources
Source summaries
NVIDIA reported that AT&T fine-tuned models for its customer-care agents with NeMo and NIM. It reported up to 40% higher response accuracy after further training and an 84% decrease in call center analytics cost.
Full details are in the original sources.