Fig. 1 — RoofMetrics AI Platform featured view (click to expand)
01
The Problem
Business Impact
Sales reps spent over 45 minutes manually tracing satellite lines for every single bid request, creating quotes with a 23% estimation error rate and slowing bid times.
Constraints
Estimations had to fall below a strict margin of error of 5% to prevent underquoting losses.
02
Technical Execution
Architectural Strategy
Engineered a pipeline that consumes address coordinates from Google Sheets, pulls high-resolution satellite imagery from Google Maps API, and routes them to a computer vision segmentation microservice.
Depth of Execution
Custom-trained a YOLOv8 instance on 2,500 annotated roof photos. Used Google Maps scale metadata to dynamically calculate physical roof square footage from the bounding mask.
03
Results
✓Estimation processing time dropped from 45 minutes to 2 minutes (95.5% reduction).
✓Estimation error rate reduced from 23% down to 6%.
✓Bid proposal win-rate increased from 31% to 48% due to faster response times.
Future Proofing
Built confidence-score routing: any estimate scoring below 85% is flagged for 30-second manual verification, eliminating false automation.