Case Study
APS Optimizes Its Virtual Power Plant with FLEX
Recurve's FLEX identifies high-performing customers and improves program results
Company Overview
Arizona Public Service (APS) is the state's largest and longest-serving electric utility, serving approximately 1.4 million customers and committed to 100% clean, carbon-free power by 2050. The utility operates in one of the country's hottest service territories: Phoenix hit 118 degrees in summer 2025, when APS set an all-time peak demand record of 8,631 MW. APS projects that figure will reach 13,000 MW by 2038, a roughly 60% increase driven by data centers, industrial load, and EV adoption.
To manage this growth, APS has built one of the largest virtual power plant (VPP) programs in North America, now nearing 200 MW of dispatchable capacity. It runs through two core demand response programs: Cool Rewards, a residential smart thermostat program with roughly 98,500 enrolled devices delivering 160 MW of load reduction, and Peak Solutions, which engages commercial and industrial customers in voluntary curtailment during peak events. Together, these programs routinely deliver hundreds of megawatts of load reduction during peak events, deferring the need for equivalent conventional generation capacity.
Quick Facts
Program Type(s)
Demand Response (Commercial & Residential)
Location
Arizona
Projects Analyzed
Cool Rewards (Residential), Peak Solutions (C&I)
Challenge
Limited Visibility Into Customer Performance
While its VPP programs achieved strong overall results, APS lacked detailed visibility into which specific customers reliably delivered savings. Without customer-level performance data, the utility couldn't identify high-value participants or understand why some customers consistently underperformed. This made it harder to focus outreach and adjust incentives, and limited APS’ ability to address underperforming customers.
Persistence Questions for Long-Duration Events
Peak Solutions events could last up to five hours, but APS didn't know whether load reductions persisted throughout longer events or degraded over time. Different business types could respond differently to extended curtailments, but without visibility into performance by sector and event duration, APS wasn’t able to optimize dispatch strategies or set realistic capacity expectations.
Measurement Accuracy Concerns
The existing "N of 10" baseline methodology showed significant sector-level bias in measurements. This created uncertainty around actual program performance and risked overpaying poor performers while potentially underpaying strong contributors. Inaccurate measurements also made it difficult to share reliable results quickly with stakeholders or justify program investments. Without a trusted, independent measurement framework, APS had no reliable way to demonstrate the true value of its demand-side capacity to its stakeholders.
Solution
APS implemented Recurve’s FLEX platform to measure and optimize performance across its VPP portfolio. FLEX connects meter data with program activity and customer context, giving APS a clear view of how its programs perform and insights into how best to improve them. The methods behind FLEX were developed in partnership with CAISO and are grounded in open, independently verifiable measurement standards, giving APS a credible foundation for reporting, settlement, and procurement decisions.
As part of its initial engagement, APS used FLEX to track hourly changes in energy use during demand response events and identify which factors (and which aggregators) drove the most reliable performance. APS continues to rely on FLEX today to target high-value customers, forecast event performance, and dispatch its VPP with confidence.
Segment-Level Performance Visibility
FLEX enables APS to track performance across more than 1,200 commercial sites and 24,600 residential participants. Results are broken down by business type, climate zone, thermostat, and account manager assignment. APS can focus on customers that deliver reliable savings and address those that degrade or snap back.
Customer-Level Performance Tracking
FLEX tracks both average performance and consistent response patterns for every participant. APS can now identify customers that deliver reliable results and those that do not respond, enabling more targeted outreach and informed decisions on incentives or participation.
Targeting & Forecasting for Program Growth
Building on that foundation, APS has since expanded its use of FLEX to support active program management and growth. FLEX uses measured event impacts and load shape characteristics to identify high-value prospective customers, helping APS focus recruitment on participants with the highest savings potential. APS recently deployed FLEX's forecasting capability, which uses historical event performance, AMI interval data, and weather forecasts to generate continuously updated predictions of expected event impacts, giving APS forward-looking confidence in dispatch decisions and a direct feedback loop between predicted and measured performance.
Results
Standardized Measurements
FLEX provides more consistent, segment-level measurements than traditional N of 10 methods. This supports faster, more accurate customer payments and reduces the risk of overpaying low performers. For Cool Rewards, FLEX measured 7.4% more savings per device than previous methods had captured, performance that was there all along.
Measured Program Impacts
Over the course of a two-year period between 2020 and 2022, FLEX analysis showed that Peak Solutions delivered 19% load reduction during events, with an average of 25.5 kW per site. The program achieved a maximum single-hour reduction of 27.9 MW during a record-breaking heatwave in early September with 456 participants. Cool Rewards delivered a maximum event reduction of 51 MW on the hottest day of a July heatwave, across 24,600 participants. These results give APS confidence in the programs’ contribution to grid reliability.
High-Value Customer Segments
Performance varies significantly across both programs. In the analysis for Cool Rewards, FLEX revealed wide variation in residential participant performance: the top 20% of participants save an average of 1.43 kW per event hour compared to 0.37 kW for the bottom 20%, a nearly fourfold difference that points to clear opportunities for targeted outreach and smarter incentive design.
Fig. 1: Distributions of average kW savings, by customer, during the Cool Rewards event hours in which each customer was dispatched.
In Peak Solutions, FLEX analysis revealed that manufacturers achieve 35% savings that persist for events as long as five hours, while grocery stores deliver 25% savings in the first hour but drop to 7% by hour three. In another example, customers with assigned account managers deliver 66% higher savings than those without.
Fig. 2: Fractional savings by hour type for the Grocery/Convenience and Manufacturing sectors.
Insights into Program Design and Dispatch
FLEX provides APS a detailed view of the factors that shape program performance. FLEX analysis showed that Cool Rewards delivered strong impacts in the first one to two hours, with pre- and post-cooling patterns that inform how events are structured and timed.
Fig. 3: Left (🔵) and counterfactual (🟥) load shapes and hourly impacts (⬛) for an average Cool Rewards customer on an average event day. The top and bottom right-hand panels show the average hourly kW and fractional savings by event-day hour type.
Performance varied by thermostat manufacturer and rate design, with specific device brands achieving the highest savings and Time of Use customers showing lower event-driven savings due to existing load shifting. These insights help APS fine-tune dispatch strategies and continue raising the bar on an already high-performing portfolio.
With FLEX, APS has a clear picture of how its VPP is performing in near real-time and where to focus next. Independent, meter-based measurement turns flexible capacity from a planning assumption into a proven, dispatchable resource, giving APS the confidence to act on what it finds and stand behind the portfolio it's built.