Electric vehicles are rolling onto driveways. Heat pumps are replacing old furnaces. Solar panels are popping up on more rooftops. But for Nova Scotia Power, these changes created a puzzle.
“Once we adopted AMI, we could see when our customers were using electricity, but we didn’t know what they were using it for,” explained Riley Cook, Nova Scotia Power’s lead data scientist.
That missing piece of the puzzle was behind-the-meter intelligence. The utility needed to understand electrification patterns to design effective time-varying pricing rates, improve load forecasting accuracy, and make progress toward the province’s ambitious renewable energy goals of 80% clean power by 2030. They turned to Bidgely and our Analytics Workbench tool to help transform the millions of Advanced Metering Infrastructure (AMI) data points they were collecting into actionable intelligence that could inform these critical business decisions and more.
“We needed to understand who is electrifying and how they’re doing it. We also needed to understand how our ecosystem is changing, especially with renewable energy targets of 80% by 2030,” added Cook. “And we had to understand how that change is happening from a bottom-up approach rather than just a top-down approach.”
The AI Detective Work
First, Nova Scotia Power had to answer a crucial question: Could they trust what the AI was telling them? Cook’s team knew they couldn’t use the technology blindly, especially for insights that would inform key decisions and the utility’s strategic direction.
“It’s really a journey. At the beginning of the journey, you really don’t know what to expect,” said Cook. “You’re given these numbers and classifications, and you have to first try to validate whether or not they make sense.”
That’s why Cook and his team started by putting the technology through rigorous validation tests across three key areas:
- Electric Vehicle Detection: They compared Bidgely’s AI-powered targeting against 199 known EV owners from their pilot program. The results? Near-zero false positives and detection rates that improved after identifying edge cases of customers who rarely charged at home.
- Heat Pump Classification: Using customer data from their historical on-bill financing program as a baseline, Nova Scotia Power worked with Bidgely to develop a probabilistic model that assigned each customer a likelihood percentage of having a heat pump. The utility ran this model multiple times, improving classification rates with each iteration. By August 2024, only 14% of customers remained unclassified.
- Heating Type Analysis: When Cook and his team compared Bidgely’s AI insights to their internal three-category heating classification model, they found 82% agreement. The disagreements weren’t about accuracy. They were about scope. Their internal model looked for primary electric heating, while Bidgely’s AI detected any electric heating signature, even occasional use.
“We found that a lot of times, we might want to tune our model. So we actually refined our model a little bit. It was a small tweak, but it added a few more percent to that 82. So we might agree about 85% of the time now. But what it did was help us refine our methods. So that was really good,” Cook says.
Understanding EV Growth Patterns
“My group’s vision is to provide data-driven insights that can inform actionable business decisions. In order to be actionable, it has to have a sizable market associated with it. So of course, we’re looking into things that are growing and changing,” explained Cook. “EV adoption is a big one, because we see a lot of EVs coming onto the system.”
The bottom-up approach revealed insights about EV behavior that were valuable to the utility:
- Actual consumption: EVs use approximately 3,820 kWh per year—a precise figure that could be directly integrated into load forecasting models.
- Seasonal patterns: EV consumption follows predictable seasonal trends, with higher usage in winter and summer months.
- Peak demand reality: Despite theoretical charger capacities of 7-11kW, actual peak demand impact was under 1kW per EV.
- Grid impact: Real data showed EVs aren’t the grid-killer many feared. In fact, there is actually a potential benefit to the grid with engaged customers.
“People thought, ‘Oh, how can the grid handle EVs? Grids are already very constrained. These EVs are very high power. Everybody plugs-in at once. Everything’s going to fail.'” reflected Cook. “I think when you use real data-driven insights, which is what we’re getting from this, it helps support the idea that, ‘no, the grid doesn’t have to fail.’ In fact, it can actually help the grid in many different ways.”
Designing Smarter Time-Varied Pricing
“We do a lot of work around rate design as well, because we’re trying our very best to put as much downward pressure on rates as possible. The whole point of knowing how customers are consuming electricity is so that we can design better electricity rates,” Cook emphasized.
The ability to create detailed load profiles for specific customer segments proved instrumental in rate planning.
“One of the ways we can really use detection and classification is to create more sophisticated load profiles,” Cook continued. “For example, as customers electrify, what does their load profile look like?”
Analyzing behind-the-meter data, Nova Scotia Power discovered that different pricing structures create dramatically different behaviors:
- Standard rate customers with EVs and heat pumps increased consumption during both peak and off-peak periods.
- Time-of-use customers with EVs shifted consumption to off-peak hours, keeping peak usage similar to non-EV customers.
- Critical Peak Pricing customers with EVs and heat pumps dramatically reduced consumption during peak events, falling below average standard offer customers on non-event days.
“It’s incredible to see what the combination of these shiftable electric loads and critical peak pricing and time varying pricing in general can really do,” said Cook.
This granular understanding enabled Nova Scotia Power to identify pathways to beneficial electrification — scenarios in which technology adoption benefits both customers and the grid.
According to Cook, “if we can create this classification model that we’ve talked about and we combine it with time varying pricing, customers who adopt these rates can save on their electricity bill while we’re also saving money as a utility.”
The Goldilocks Zone of Energy Management
Better rate design is only one way that Nova Scotia Power is eager to put data to work to ensure electrification benefits everyone involved. A granular understanding of how customers actually use electricity provides the foundation for smart grid management across the board.
“These insights do help us get closer to the Goldilocks zone where customers are saving, we’re saving, everybody’s saving, and we’re doing things for the grid that are sustainable in the future,” Cook said.
To hear more from Cook and learn more about how Nova Scotia Power is unlocking grid insights with AI meter data analytics, download the full case study about the utility’s data-driven approach to the energy transition, including details about their validation methodologies, comprehensive results, and specific insights that demonstrate how utilities can turn smart meter data into strategic advantage. Or, watch the on-demand webinar, “Nova Scotia Power: Using AI to Optimize TOU Load Shaping.”
