Financial services · Databricks
Experian: Delivering faster, more helpful customer service
Public case study · not a Metric Oak engagement
- Faster in fine-tuning time for Llama 8B model
- 11x
- Of email responses to customers are now automated
- 35%
- Increase in NPS, signaling stronger customer satisfaction and loyalty
- 8%
What happened
The story in brief
A global credit bureau wanted to answer high volumes of customer emails faster without losing quality. It fine-tuned an open large language model on Databricks to draft responses, automating a large share of replies and lifting customer satisfaction scores.
How it compares
- 11x faster in fine-tuning time for Llama 8B model, against a median speed-up of 5x across 8 financial services results.
Planning something similar?
AI-ready data
Governed, retrievable data for GenAI, RAG and agents. Senior engineers, fixed scope, measured results.
Related case studies
Financial services · Databricks
ABN AMRO10x
Faster time to market: use cases deployed in 2 months
Financial services · Databricks
Addepar5x
Faster pipeline development, accelerating client onboarding
Financial services · Databricks
AME Digital90%
Accuracy in fraud prevention models
Related insights
Frequently asked questions
What results did Experian report?
Experian reported 11x faster in fine-tuning time for Llama 8B model; 35% of email responses to customers are now automated; 8% increase in NPS, signaling stronger customer satisfaction and loyalty, according to its customer story published by Databricks.
Which platform and technologies did Experian use?
Experian used Databricks, including Unity Catalog, MLflow, Model Serving, Large language models.
How do these results compare with other financial services programs?
11x faster in fine-tuning time for Llama 8B model, against a median speed-up of 5x across 8 financial services results.
Source
Customer story published by Databricks: Delivering faster, more helpful customer service (opens databricks.com in a new tab) on databricks.com. Figures as published, retrieved 27 October 2026. The summary above is Metric Oak's own. This is not a Metric Oak engagement.
Have 30 minutes? Get a senior architect’s view of your data platform.
No sales deck. You talk to the engineer who would lead the work.
Book an architecture review