2027 Event Agenda

Distribution grid operators worldwide are caught in a perfect storm: skyrocketing electrification demand colliding with volatile, extreme weather events and the operational bottleneck of manual asset inspections. While individual utilities struggle to deploy automation due to the "AI data desert"—the inability to generate enough clean data alone—New Zealand has emerged as a compelling global grid test lab. Facing aggressive electrification targets alongside a hyper-volatile climate that packs alpine snow, coastal corrosion, high winds, and dense urban corridors into a single footprint, New Zealand’s grid experiences a concentrated mix of global operational challenges. This is further enabled by a uniquely coordinated industry and regulatory environment that allows competing utilities to collaborate on shared data challenges, providing a model for global replication.

This session details a collaborative model built to overcome this data scarcity and drive practical, end-to-end field resolution. Five of New Zealand’s largest electrical distribution businesses (EDBs)—Vector, Orion, Northpower, Unison, and WEL Networks, which collectively serve 2.2 million consumers—formed a data-sharing alliance with Tapestry, Alphabet’s moonshot for the electric grid. By pooling data from tens of thousands of high-resolution images, this partnership demonstrates how a unified computer vision consortium can rapidly improve grid visibility and transform asset management from initial aerial capture through to field remediation across diverse operating environments. Initial deployments have been led within individual networks, with the consortium now working toward shared performance baselines across all participating utilities.

The panel will present verified field data and technical milestones across three core operational pillars:

1. Operational Realities & Quantified Field Performance The presentation will share empirical results demonstrating how a shared computer vision framework changes field operations across rural, urban, and remote networks, connecting data capture directly to maintenance execution. Initial deployment data includes:

• Up to 83% Reduction in Inspection Time: Optimizing manual roadside utility pole inspections from 30–45 minutes down to 5–7 minutes per asset, speeding up work-order creation and improving team safety in challenging terrain.
• Up to 221% Increase in Asset Visibility: Improving data accuracy for the location, condition, and status of overhead infrastructure to enable proactive grid maintenance and targeted remediation before major weather events occur.

The consortium is currently working toward harmonized, cross-utility performance metrics to demonstrate repeatable, end-to-end operational efficiency outcomes across all network environments.

2. Model Portability: Cross-Utility Scaling Attendees will learn how a baseline model trained on one network has adapted across four adjacent networks managing over 150,000 kilometers of lines, creating a continuous feedback loop where data from one utility benefits the entire consortium. This technical framework strictly isolates access entirely to authorized personnel and adheres to the energy industry's most stringent regulatory and data custody standards.

3. Maximizing Capacity via Human-in-the-Loop (HITL) Workflows Fully autonomous systems face regulatory, safety, and trust barriers in operational environments. This session highlights the mechanics of an optimized HITL workflow designed to eliminate manual data-entry bottlenecks, accelerate the transition from defect detection to field resolution, and maximize engineering capacity:

• Pre-Detection Automation: Computer vision models pre-render and locate poles, crossarms, and specific defects before an inspector opens the image file.
• UX Optimization: Moving utility inspectors into an accelerated "accept/reject" bounding-box verification interface, freeing up hundreds of person-hours and allowing limited utility resources to pivot from data triage to high-value engineering and field repairs.

Three Core Takeaways for Attendees

1. The Utility Consortium Playbook: Practical steps to build multi-utility data alliances, overcome machine learning data scarcity, and safely navigate data custody and regulatory boundaries.
2. Deployable HITL Architecture: Software design and UX guidelines to successfully integrate computer vision into regulated utility workflows to accelerate the timeline between asset data capture and field remediation.
3. Data-Driven Asset Management: How to integrate AI-generated insights into GIS, asset systems, and planning processes to support proactive maintenance and network decision-making.