United StatesFinance and insurance2022 – ongoingUSAA
Vehicle repair decisions and labor estimates from damage photographs
USAA's vehicle-repair models achieved a peak 28% performance improvement, Google reported
Problem
Identifying damaged vehicle parts did not establish whether they needed repair or replacement. USAA also needed estimates of repair labor hours.
Approach
USAA and Google combined vehicle-damage information with structured claims data. They selected Vertex AI AutoML models and built prediction APIs with retraining pipelines. Model managers retained manual control over promotion into production.
Results
Takeaway
The reported improvement concerned model performance against baselines. It did not measure claims-processing savings.
Categories
- Size
- Large enterprise
- Industry
- Finance and insurance
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Computer visionPredictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
Google described models for repair-or-replace decisions and labor-hour estimation, together with production-serving and retraining infrastructure. It reported a peak 28% improvement against baseline models.
Full details are in the original sources.