Utilities don’t need another reminder that load growth is accelerating. What’s changed are the options they have for responding. And vital to the success of many of these options is customer engagement.

One option is Virtual Power Plants (VPPs). The DOE’s latest VPP update highlights demand-side flexibility as one of the near-term paths that could help to meet rising demand without overreliance on slower and typically more expensive supply-side investments. The DOE notes that VPPs have the potential to scale to 80-160 gigawatts (GW), or 20% of peak load, by 2030. But that hinges on customers enrolling in utility-sponsored programs where garnering participation can prove to be challenging. DOE assumes 30-50% participation from new dispatchable distributed energy resources (DER), despite FERC reporting that only about 14% of eligible U.S. customers are enrolled in comparable programs today.

Closing the gap comes down to better customer engagement. Other consumer markets have shown that personalization drives 3.5x better results. Utilities must adopt a more modern playbook for engaging customers around energy, one that includes targeted, personalized outreach.

The key to unlocking that personalization? Predictive customer intelligence on every home served. And the ability to do so is now at our fingertips thanks to artificial intelligence (AI).

North American Electric Reliability Corporation (NERC) warns of multiple regions facing long-term reliability risks. Demand-side flexibility is now critical, but requires customers to participate at dramatically higher rates than they have historically.

Personalization requires rich, accurate, real-time customer data. But this data has historically been fragmented across numerous sources, from property records to satellite images to the utility’s own meter data. Complicating matters, some facts and figures, such as information on heating, ventilation, and air conditioning (HVAC) systems, are not well-documented in public records despite being crucial to a home’s energy profile.

Additionally, utility meter data and other property attributes aren’t the only characteristics that count. For example: Owners’ and occupants’ demographic data are critical to understanding household behavior.

Overcoming these data hurdles is too costly and time-consuming for even the most well-resourced utilities and talented data science teams, making it challenging for utility programs to develop a unified view of each home in their territory. And that’s made it difficult to personalize messaging that resonates with customers. The nation’s most successful residential energy programs still enroll just tens of thousands of homes, while millions are available and in need.

Importantly, raw household data alone still doesn’t answer the core questions utilities care about:

• Which customers are most likely to adopt new technologies or participate in programs in the future?
• How do we engage them with relevant messaging at the right time?

Efficient program planning demands sharp inferences, and sharp inferences demand an accurate present-day view. Trained on billions of residential energy-related data points, AI helps deliver both: ‘digital energy twins,’ or household profiles, and predictive models for utilities.

Just some of the hundreds of descriptive and predictive attributes that can be analyzed for every home in the U.S.

AI excels at making sense of massive amounts of information, especially when the signal isn’t obvious. Instead of looking at each field or attribute in isolation, AI learns how different variables tend to show up together. That means it can flag inconsistencies, standardize inputs that don’t match, and fill in missing fields based on patterns it sees elsewhere in a dataset.

And while each utility possesses valuable meter, billing, and program participation data for a portion of U.S. households, its visibility is naturally limited to just its own customer base. AI can analyze trends across all homes and energy providers in the country, extrapolating insights that no single utility could.

But even more magic can now happen when utilities use AI to predict future behaviors and outcomes—and act accordingly. The same underlying approach that lets AI augment individual attributes in a dataset can further be used to identify homes that already qualify for energy efficiency (EE), demand response (DR), and VPP programs, forecast which customers are likely to adopt efficiency technologies next, and prioritize outreach by expected grid value. Before and after DER assets show up.

This works by leveraging machine learning (ML) to analyze hundreds of attributes across an ideal customer segment and tailor a proprietary propensity model – i.e., an analytics tool that calculates the likelihood of someone performing a specific action – to find other households most closely resembling that profile. This ‘lookalike audience’ can then be targeted through direct mail, email, or advertisements.

For example, a battery and VPP operator in Electric Reliability Council of Texas (ERCOT) that partnered with 257 proved just how powerful predictive AI and personalized marketing can be, accelerating VPP enrollment by 30% over their prior efforts.

Meanwhile, a southwestern investor-owned utility (IOU) tapped us to tailor a propensity model to help boost enrollment in their smart thermostat program, and a global HVAC manufacturer working with a southeastern utility identified homes with resistance electric or older heat pumps that could save hundreds of megawatts (MW) of winter peak demand if retrofitted with a modern cold-weather heat pump.

Perhaps counterintuitively, even as AI enables us to analyze complex data, it also simplifies access for non-data scientists.

Generative AI has changed how people interact with data: what once required scoping, coding, and interpreting raw outputs—and the involvement of engineers and analysts—can now happen on demand through everyday conversation. By prompting in natural language and iterating on responses, users can get immediate answers to pressing questions and explore new angles in real time.

This gives utility program managers, planners, and marketers the ability to generate clear, useful insights themselves, significantly compressing the time from question to decision to action. That speed is critical: personalized communications and experiences only work when they are timely, and insights that arrive too late quickly lose their value.

At 257, Pink is the conversational interface that anyone can use for research, planning, or forecasting, regardless of technical background. Powered by generative AI, it’s the accessible “front door” to billions of residential energy data points.

More than 200 utility, solar, storage, HVAC, and retail electric companies have joined 257 to rewrite the playbook for consumer engagement in the residential energy space. Here’s how we help:

• 130 million ‘digital energy twins,’ or household profiles, for all U.S. homes, with hundreds of up-to-date attributes on every residence
• Predictive propensity models for thermostats, EVs, batteries, and beyond, helping utilities acutely understand and mobilize their customers
• Custom audience targeting across the media landscape so utilities can confidently reach the right households on channels like Meta, email, and direct mail

Propensity models let any U.S. utility predict which homes in their territory are most likely to benefit from an EE program, participate in a DR program, or purchase a flexible asset.

As grid growth explodes, utilities are under pressure to deliver both reliability and affordability, fast. Residential programs can help, but only if we dramatically scale participation. That means moving beyond more traditional, broadcast-style outreach to precise targeting and personalized messaging. Achieving that requires turning fragmented household data into predictive customer intelligence, which is now feasible, affordable, and immediately available.