How APS dove head-first into AI [partner content]

Episode Summary

Executive Summary: The episode explores how utilities are combining vertical AI built for sector-specific tasks with horizontal AI platforms to turn energy data into actionable decisions. Using APS and Bidgely as a case study, the discussion shows how appliance-level disaggregation began with high-bill explanations and expanded into broader utility use cases like forecasting, resiliency, customer programs, and revenue protection, with data governance and keeping models close to the data emerging as critical.

Main Topics: Vertical AI meets horizontal AI in utilities (Priority: 5/5): The conversation frames utilities as a strong use case for sector-specific AI that is now converging with general-purpose AI to produce more powerful outcomes. High-bill explanation as the entry use case (Priority: 5/5): APS began with AI to explain emotionally charged high bills by disaggregating household energy use, improving customer understanding and call center responses. Bringing models to the data (Priority: 5/5): Bidgely describes containerizing models and deploying them inside the utility's cloud environment to address security, sovereignty, and latency concerns. From one use case to an enterprise data layer (Priority: 4/5): The collaboration evolved from a SaaS point solution into a broader data fabric that supports multiple business units and applications across customer and grid domains. Governance, operating model, and value realization (Priority: 5/5): Both speakers stress centralized governance, clean foundational data, and clear build-buy-partner decisions to scale AI responsibly in a regulated industry. Future utility AI use cases (Priority: 4/5): The utility sees disaggregation data supporting load forecasting, grid reliability, outage analysis, theft detection, customer experience, and marketing/efficiency programs.

Key Arguments: Vertical AI is most effective when trained on industry-specific data, workflows, and constraints, and utilities are a prime example. AI in utilities is shifting from digitizing processes to taking actions on data and directly influencing operational decisions. High bill explanations improve customer trust because they replace generic answers with appliance- and time-specific causes. Data sovereignty makes it impractical to ship AMI data to multiple vendors for every use case; models should run near the data. A shared utility data layer creates reusable intelligence that can serve many departments without rip-and-replace IT changes. The most successful utility AI programs start with a real business problem, not with technology for its own sake. Centralized governance is needed early to avoid duplicate tools, wasted spend, and inconsistent outcomes across departments. AI adoption should move from pilots to production by focusing on the highest-value use cases and clear value capture.

Data Points: Length of Bidgely investment in disaggregation: Over 12 years - Karthik says Bidgely spent more than a decade building disaggregation at scale with high accuracy. Scale of metadata processed: 30-50 million+ - Bidgely has invested in taking large amounts of metadata at scale. Utility cloud environment example: Oracle Cloud Infrastructure - APS uses OCI, and Bidgely deploys models into that environment. Series length: 4-part series - The episode is the first in a four-part series on utility AI applications. Timeframe of rapid AI change: Last 18 months - Venkat says the pace of AI change has accelerated significantly over the last 18 months. Historical comparison window: 12 months back - Karthik notes that a year earlier the question was whether and how to use AI; now that debate is largely settled. Industry tenure: Almost 25 years - Karthik describes his experience at the intersection of software, AI, and industrials. Utility tenure: More than 2 decades - Venkat says he has worked in the utility sector for over 20 years.

Pivotal Quotes: "We build the foundational data layer for a utility to disaggregate and identify for each data point what is happening behind a household." — Karthik Murthy: Describing Bidgely's core value proposition and disaggregation capability. "You should never start from an AI point of view. You should actually start from solving a business problem." — Venkat Nimula: Summarizing the utility-first approach to AI adoption and governance. "How do you containerize the model? How do you run it in a utility? How do you keep the data in a way which helps them innovate with the latest technology in hand?" — Karthik Murthy: Explaining the technical and governance challenges of scaling AI across utility use cases.

Implications: Utilities that treat data as core infrastructure, deploy models near the source, and govern AI centrally can scale from point solutions to enterprise transformation. The result is better customer experience, more resilient grids, and faster value realization.