As the largest electricity distributor on Australia’s east coast by population served, Ausgrid plays a critical role in delivering energy to 1.8 million customers across Sydney, the Central Coast, and the Hunter region. Managing a complex network spanning 22,275 km², the organisation is directly accountable for network performance, including outages, faults, and infrastructure upgrades.
To support these responsibilities, trusted data is essential. While Ausgrid has invested significantly in building its data capability, it identified a key opportunity to improve how it models the time it takes to restore energy after a failure.
When power outages occur, customers want answers quickly. One of the most important questions is always:
“When will my power be restored?”
Traditionally, estimated restoration times (ERT) rely heavily on operational procedures, available information, and expert judgement. While these processes are effective, they can be resource-intensive and difficult to scale during periods of high outage activity. Recently, our team partnered with Ausgrid to explore how artificial intelligence could improve outage restoration forecasting and customer communications.
The Challenge
Ausgrid, among other utilities face a complex balancing act during outage events: they must provide accurate restoration estimates to customers, support operational teams with timely information, and manage high call volumes to maintain trust.
Many restoration estimates are generated using the limited information available at the very start of an outage. As more operational data flows in, those estimates require constant revision. The challenge wasn't just to determine if machine learning could help provide more accurate forecasts, but to figure out how to structure, clean, and pipeline the massive amounts of disparate operational data required to make those AI models viable.
Our Approach: The Data Foundation for AI
Working alongside Ausgrid subject matter experts, operational staff, and our AI modeling partners, we focused our expertise on the foundational element of any successful predictive project: data engineering.
We led the Exploratory Data Analysis (EDA) to understand Ausgrid’s existing broad information landscape. By thoroughly evaluating and integrating historical outage data, operational network information, workforce dispatch signals, and weather inputs, we created a structured, reliable data environment. This heavy lifting on the data side is what enabled our modelling partners to successfully develop and train the explainable machine learning models.
What We delivered
Instead of treating AI as a standalone tool, we designed a solution rooted in strong data infrastructure. The engagement produced:
- Exploratory Data Analysis (EDA): We conducted deep-dive analyses into legacy systems, identifying historical patterns, mitigating data gaps, and engineering the features necessary for predictive modeling.
- Real-Time Data Pipelines: We built a production-style architecture capable of consuming operational data in real-time, transforming it seamlessly, and publishing the ML-ready datasets automatically to reporting systems.
- Operational Dashboards: We facilitated the data layer for interactive dashboards designed to monitor active incidents, providing users with total transparency into the underlying data drivers that produced a forecast.
- Future-State Architecture: We developed a comprehensive roadmap demonstrating how modern data platforms could integrate predictive forecasting into long-term operational workflows and customer-facing channels.
Key Lessons Learned
One of the most important findings from this proof of concept was that successful AI solutions in operational environments require much more than just model development.
The true success of the predictive forecasts depended entirely on data quality improvements, clear governance, and integration into existing operational workflows. In many cases, the strength and reliability of the underlying data engineering has a greater impact on business adoption than the specific modeling techniques themselves.
Opportunities Across Industries
While this project focused on outage restoration forecasting for Ausgrid, the underlying pattern applies to nearly all industries. Similar data-first approaches are required for asset maintenance planning, workforce scheduling, supply chain disruption management, and risk prioritisation.
Where organizations already collect operational data, there is a significant opportunity to create predictive, explainable decision-support tools, provided the data is properly structured and engineered for AI consumption.
Looking Ahead
The future of operational AI is not about replacing people; it is about enabling teams with better information, faster insights, and greater confidence in decision-making. The most successful outcomes occur when specialized data engineering, deep domain expertise, and advanced data science work together to solve real operational problems.
Interested in Exploring Similar Opportunities?
Our team works with many organizations across infrastructure, utilities, operations, and analytics to build the modern data platforms required to make AI a reality.
If your organisation is exploring data platform modernisation, real-time analytics, or preparing your data infrastructure for operational AI, we would love to have a conversation. Contact us to discuss how a strong data foundation can support your operational challenges and business objectives.
