An ongoing billing snafu in ComEd territory in northern Illinois has some solar companies bracing for turbulence.
When the problem is remedied, community solar subscribers will see a backlog of credits on their utility bills, but also accumulated charges for participating in the project. It’s the kind of surprise that can sour customers on community solar and cause them to unsubscribe from a project.
A new tool being marketed to community solar developers promises to use artificial intelligence to help intervene before such customers drop out.
Subscriber turnover is known as “churn,” and it can be a major headache for owners and managers of community solar projects. Frustration with billing is among the main reasons people withdraw from projects. Others might leave if they are moving, or if a project takes longer than expected to come online.
Churn is “a huge issue in the community solar space,” said Sam Van Dam, director of asset management for the solar developer 38 Degrees North. “It’s something we spend a lot of time thinking about.”
The stakes include lost revenue and added uncertainty. In areas where community solar is already popular, it can be challenging to find customers still looking for subscriptions. In places where it hasn’t been widely embraced yet, it can mean hours of additional work educating potential subscribers before convincing them to sign up.
Solstice, a solar company that works with developers to enlist and serve community solar subscribers, is hopeful that artificial intelligence can help companies identify and intervene with customers at risk for churn, and also make community solar more accessible and inclusive in the process.
Solstice has developed an AI tool — now in testing — that predicts when certain subscribers may be vulnerable to churn, based on data from 15,000 accounts that the AI machine learning model has been trained on. The insight allows Solstice to proactively reach out to subscribers who may be distressed, making sure they are comfortable with their subscription and allaying concerns or confusion.
If a subscriber is likely to leave, advance warning also helps Solstice more efficiently manage a waitlist of aspiring subscribers, getting them more quickly enrolled.
Promising numbers
Solstice spokesperson Mary Jackson said the AI tool was especially useful when 38 Degrees North transferred management of several thousand subscribers from another company to Solstice, a transitional period when customers might have dropped away.
“Using our churn intervention strategies and high-touch customer service, we kept a staggeringly high number of those subscribers onboard” — 96% of them, Jackson said.
Van Dam called the AI tool “a fantastic idea.”
“It becomes particularly challenging when you have hundreds of residential customers on a single project,” he said. “There’s a lot to getting everyone signed up and making sure they keep paying their bills if they are replaced. When they have to be replaced, that potentially costs money and potentially results in lost revenue. Avoiding that is the preferred approach.”
During a pilot program, Solstice saw churn reduced from 48% to 8.3% among a targeted segment of at-risk customers who had a greater than 89% chance of churn, according to their predictions, the company said. A customer’s length of time as a subscriber, and whether a project has consolidated utility billing, are among important predictors of churn.
Solstice data engineer Jake Ford explained that the machine learning tools analyze the training set of data using “advanced algorithms, including deep learning neural networks to detect patterns between variables, gradient boosting classification algorithms and other tree based models, along with other traditional regression techniques.”
“All of these models are designed to learn patterns and relationships from large datasets,” he continued. “This dynamic nature of machine learning is what differentiates these approaches from traditional computer modeling, in particular static algorithms that were often hard-coded, meaning little to no flexibility to adapt to new inputs and data. This is critical, as often the inputs or customers we wish to analyze in our machine learning applications are different – locationally, demographically, behaviorally – than those we trained the models with.”
Redefining risk
Solstice is also hopeful that AI can provide a more accurate and fair way of vetting potential solar subscribers. Typically credit scores are used to decide whether someone is likely to pay their bills, but that means people with poor credit from past financial struggles, or lower-income people in general, may be left out.
Ford said their AI-based model known as EnergyScore appears to show that customers who might otherwise be sidelined by a poor credit score are actually good fits for community solar, since data shows people are likely to pay their energy bills, even when finances are tight. This might help the households who most need energy savings access community solar.
“When we’re talking about low-income participation in community solar projects from the perspective of a developer or financier, their concern comes down to risk — revenue risk, churn risk,” said Ford. “The perception is: risk is too high, so let’s not include any low-income customers. Our data is showing the perception of risk is greater than the real risk. There haven’t been that many efforts in the energy industry to measure what the real risk is.”
Solstice developed EnergyScore “in partnership with The Department of Energy and data scientists at MIT and Stanford using more than 800,000 individuals’ data across 5,000 variables,” the company says. Testing of the patent-pending product has shown that it is more accurate than FICO scores in predicting default rates on solar payments.
While Solstice is using AI tools to combat marginalization, Ford said they are vigilant regarding the well-known risks of discrimination, racism and other unintended consequences being generated by AI.
“It’s about being aware and continually reviewing what your model is doing, being cognizant of how it’s impacting real individuals on the ground, not just rows on a spreadsheet,” he said. “It’s important to be nimble and flexible in your methodology.”
Increasing equity
Solstice CEO Steph Speirs said AI tools could be especially useful as community solar blossoms in popularity and companies strive to manage larger subscriber bases in an equitable way.
“We’re at an incredible inflection point in the energy transition,” she said. “There’s been focus on the supply side, but there’s a lot more technology that could be applied to the demand side to improve both the customer experience and the perception of projects in the community. We wanted to apply machine learning lessons around customer behavior and start to improve the metrics that developers care most about for community solar projects.”
Metrics around churn, subscription levels and collection rates affect a solar developer’s ability to get financing for community solar projects.
“If those metrics get diminished, the project’s viability is threatened,” said Speirs, who co-founded the company in 2016 with the goal of increasing low-income participation in community solar. “We need to really make sure these projects have low churn rates, and high subscription rates and collection rates.”
Speirs said that currently, only about 10% of community solar subscribers are low-income. While that is changing thanks in part to equity incentives in the Inflation Reduction Act and state solar programs, “we have work to do as an industry.”
“The beauty of these machine learning and AI models is we can use data to rewrite the historical exclusion that has existed in this industry, and improve financial viability of these projects so we can build more of them faster,” Speirs continued. “That helps both sides of the marketplace. It helps developers and financiers building projects at a cost of millions and billions of dollars, and it helps low-income customers access these projects.”
Solstice isn’t the only company in the space. Erik Molinaro, senior vice president of customer experience & operations for solar developer Nexamp, said developers and brokers across the board are using advanced technology and artificial intelligence to facilitate community solar recruitment and retention. The company uses AI to create personalized videos that walk customers through the line items on their bills, he noted.
“Any time you have something a little out of the ordinary, it triggers a customer, it’s a pain point,” Molinaro said. “We’re looking at that data, leveraging things like ChatGPT to understand why our customer is calling us, things we can do to create a better environment.”
Community solar developers look to artificial intelligence to help manage subscribers and advance equity is an article from Energy News Network, a nonprofit news service covering the clean energy transition. If you would like to support us please make a donation.
Leave a comment
