NASTT Case Study Compilation: 4 Studies, 1 Industry-Transforming Solution

America's sewers receive a failing grade. The American Society of Civil Engineers has given the country's wastewater infrastructure a 'D' grade. Multiple studies find that 40% of CCTV image interpretations differ between reviewers, that sewer defect severity is underestimated by 15%, and overestimated by 20%. See how SewerAI turned speculation into science in these four case studies — while raising America's 'D' grade to an A+.

Case Study #1: Unmatched Speed. Unparalleled Accuracy. (Cross-Bore Detection)

The Problem

SewerAI conducted an analysis of a dataset of 1,040 lateral inspection videos to measure the efficacy of AutoCode in detecting cross-bore conflicts of gas distribution lines with side sewer laterals. The dataset included 40 videos with known cross-bores and 1,000 videos not containing cross-bores — approximately 64 hours of combined playback time covering roughly 25,000 linear feet of pipe data. With over 60 hours of footage to process, manual review workflows were severely backed up.

The Solution

AutoCode by SewerAI leverages AI and machine learning to rapidly and accurately analyze sewer system data to detect specialized defects and conditions. AutoCode was specifically calibrated to identify cross-bores and any other regions of interest in the pipes that resemble or indicate evidence of cross-bores.

The Results

AutoCode delivered a 70% reduction in review time and 100% overall accuracy after QA review. Key outcomes included:

• 39 out of 40 cross-bores correctly identified before human review
• 30% reduction in footage requiring review (64 hours narrowed to 19 hours)
• 70% reduction in total review time (108 hours reduced to 32 hours)
• 97.5% accuracy before human review; 100% overall accuracy after review
• AI model continues to improve with every use

Case Study #2: Critical Tool for an Essential Need. (City of Houston)

The Problem

A looming EPA Consent Decree pressured the City of Houston to inspect its 6,200-mile sewer system and 129,000+ manholes. The City employed SewerAI's AutoCode to assess its accuracy against contractor submittals — with 1,500+ surveys coded by AutoCode and 150 surveys coded by four different CCTV inspection contractors used as a benchmark. The results of the citywide inspection program will directly influence how $6 billion is spent over the next 15 years.

The Solution

The City of Houston integrated SewerAI's AutoCode AI into its assessment and capital planning workflow, using it to evaluate the accuracy of contractor-submitted CCTV inspection data at scale.

The Results

Houston found AI to be 8x more accurate than contractor submittals:

• 97% average accuracy at properly coding identified defects
• 8x greater accuracy compared to outputs from CCTV inspection contractors

Case Study #3: More Detections. Less Time. (West Coast Water Utility)

The Problem

A U.S. West Coast water utility assessed 200,000+ linear feet of CCTV inspections using SewerAI and compared its findings against what a contractor assessed from the same pipe assets. Manual surveys were reported as highly inconsistent. The dataset covered 252,302 linear feet of AI-inspected footage and 192,473 linear feet of manually-inspected footage from the same source — comprising 1,278 total inspections and 5,379 lateral (tap) connections.

The Solution

SewerAI deployed AutoCode to detect infiltration and structural defects across the utility's inspection footage. The Pipe Survey Comparison Index (PSCI) was used to quantify the consistency between observations recorded under both AI-assisted and manual methods.

The Results

Compared to manually-coded surveys, AI-assisted surveys:

• Identified 75.6% more groundwater infiltration defects
• Identified 42.4% more structural defects
• Identified lateral connections missed entirely by manual methods

Case Study #4: Saving Time, Money, and Infrastructure. (Pacific Northwest Utility)

The Problem

A Pacific Northwest wastewater utility was seeing resources go down the drain with manual sewer assessments requiring hours of CCTV review. The process was exacerbated by errors and misdiagnoses that resulted in costly reactive measures. The dataset covered 29,748 linear feet of CCTV inspections across 132 total surveys.

The Solution

AutoCode assessed 29,000+ linear feet of CCTV inspections. SewerAI QC staff cross-checked findings against manual assessments, and a PSCI score was produced to quantify inconsistencies between inspection methods.

The Results

• AI identified 32.99% more conditions (defects and features) than manual survey
• AI displayed less than 10% margin of error, missing 90.13% fewer conditions than the manual survey

Summary: The Numbers Speak for Themselves

Across all four case studies, SewerAI's AutoCode AI consistently outperformed manual inspection methods:

• 70% less review time
• Less than 10% margin of error
• Lower chance of over- or underestimating defect severity
• 97% average coding accuracy
• 32.99% more conditions found
• 42.4% more structural defects found

SewerAI meets NASSCO, PACP, and other industry standards and is used around the country to serve more than 130 million people.

Download the full NASTT Case Study compilation to explore the complete methodology, data, and findings from all four SewerAI deployments presented at the 2023 NASTT No-Dig Show.