AI-Powered Cross Bore Safety Program Reduces Review Time by 70%

Overview

PG&E, one of the largest combined natural gas and electric utilities in the United States, partnered with SewerAI to address a critical public safety challenge: detecting utility cross bores in sewer laterals at scale. Using SewerAI's PIONEER platform and AutoCode™ defect detection engine, PG&E reduced manual review time by 70% while achieving 97.5% AI accuracy and 100% final detection accuracy through a combined AI and human review workflow.

The Challenge

Horizontal directional drilling (HDD) is widely used to install utilities with minimal excavation — but it carries the risk of accidentally drilling through existing pipelines. These utility "cross bores" often go undetected and pose serious public safety threats, particularly when gas lines intersect sewer laterals.

PG&E needed to scale cross bore detection across a large inventory of sewer lateral CCTV inspections — without relying solely on time-intensive manual review. With over 1,040 lateral inspection videos and 64 hours of footage to process, a manual-only approach was neither efficient nor sustainable.

The Solution

SewerAI deployed its AI-powered PIONEER platform and AutoCode™ defect detection engine to assist with cross bore identification across PG&E's lateral inspection program. AutoCode analyzed over 1,040 lateral inspection videos — approximately 25,000 linear feet of pipe — detecting key regions of interest and flagging likely cross bores for human validation.

The workflow used a two-step review process:

• Step 1: AutoCode AI model analyzes footage and flags likely cross bore locations
• Step 2: PACP-certified NASSCO technicians validate AI-flagged segments and confirm true cross bores

This human-in-the-loop approach preserved technician oversight and confidence in results while dramatically reducing the volume of footage requiring manual attention — from 64 hours down to 19 hours.

Results

70% Reduction in Review Time

Manual review time dropped from 108 hours to just 32 hours — a 70% reduction in person-hours. AutoCode narrowed the focus from 64 hours of raw footage to 19 hours of flagged segments, allowing certified technicians to concentrate their time where it mattered most.

97.5% AI Accuracy — 100% Final Accuracy

AutoCode correctly identified 97.5% of known cross bores. The single missed cross bore was attributable to heavy deposits obscuring the defect in the footage. With the addition of human review, final detection accuracy reached 100% — demonstrating that the AI-plus-human workflow delivers both speed and completeness.

25,000 Linear Feet Processed

Across 1,040+ lateral inspection videos and approximately 25,000 linear feet of pipe, AutoCode delivered consistent, reliable cross bore detection at a scale that would have been impractical with manual review alone.

Key Outcomes

By deploying SewerAI's AutoCode™ platform, PG&E achieved measurable gains across safety, efficiency, and operations:

• Save time: AutoCode reduced 108 hours of review to 32 — saving 70% in person-hours
• Improve accuracy: 97.5% of known cross bores flagged by AI; human review brought total accuracy to 100%
• Lower costs: Faster inspections translate to lower costs and more proactive safety measures
• Keep control: Technicians retained oversight of every AI prediction to ensure confidence in results
• Centralize operations: All data, AI analysis, and reviews in one integrated, cloud-based platform
• Increase efficiency: Fewer hours, faster decisions, and safer outcomes — with less manual overhead