JapanTechnology and telecom2025 – ongoingNTT DOCOMO
LLM-based analysis of usage on its internal generative AI platform
NTT DOCOMO cut its monthly usage analysis from 66 hours to 6 hours with GPT-4o and natural-language queries
Problem
NTT DOCOMO runs an internal LLM platform for employees. Analyzing how staff used it took a great deal of manual data processing each month.
Approach
The team used Azure OpenAI's GPT-4o, deployed through Databricks Model Serving, to extract entities from user prompts and responses and sort them into categories that show usage patterns. It also adopted Databricks Genie so analysts could query the usage data in natural language.
Results
Takeaway
Using an LLM to categorize free-text prompts turned a monthly reporting chore into a short task.
Categories
- Size
- Large enterprise
- Industry
- Technology and telecom
- Business function
- IT and R&D
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Databricks describes how NTT DOCOMO analyzes usage of its internal LLM platform with GPT-4o and Genie. It reports a 90% drop in manual data processing time and monthly analysis falling from 66 hours to 6 hours. The platform had about 10,000 monthly active users as of February 2025.
Full details are in the original sources.