← All News
How PowerMarket Uses AI and Machine Learning to Win Business and Acquire Customers Efficiently
By Leland Gohl
Customer acquisition in community solar has long been a major challenge—costly, unpredictable, and difficult to scale. As PowerMarket began enrolling tens of thousands of customers annually, it became clear that relying solely on traditional tactics wouldn’t support the rapid growth of community solar nationwide. We needed data—real, actionable insights that could turn customer acquisition from a guessing game into a repeatable science.
Through experimentation and investment in AI and machine learning, we’ve transformed how we identify and convert prospects. Here’s how we did it—and what we’ve learned along the way.
Where Acquisition Channels Fall Short
With over a decade of experience in community solar, we’ve tested nearly every major acquisition channel in the industry. Each has strengths, but also clear limitations.
Door-to-Door Sales (D2D)
Still widely used across the industry, D2D has serious constraints: reputational risks, unpredictable customer volume, limited scalability, and potentially high compensation. While effective in specific cases, D2D alone can’t support PowerMarket’s required scale.
Facebook
Facebook offers strong targeting capabilities—up to a point. Its algorithms don’t always provide sufficient customer targeting capabilities, and even when cost-effective, identifying who’s converting can feel like a black box.
Partnership Marketing
Municipal partnerships can be highly effective. A key challenge, however, is identifying which households are located within that community. Unfortunately, a lot of residential data (voter records, tax rolls, USPS change of address requests, etc.) is limited, inconsistent, or out of date—and municipalities are, for good reason, reluctant to share it. To make these partnerships viable, we had to prove we could generate accurate, actionable lists of local residents ourselves.
Direct Mail
Direct mail offers clear targeting advantages, but it’s expensive, slow to yield results, and extremely sensitive to list quality. We’ve seen performance range from $5 to $2,000 per acquired customer, largely depending on how the prospect list was built.
Traditionally, there are three ways to build these lists:
The Problem: Guesswork & Unpredictability
We were treating every household as if it were equally likely to convert—which we now know is far from true.
Renters behave differently than homeowners. Younger adults have different motivations than retirees. Income, location, and lifestyle all influence how people engage with community solar. A one-size-fits-all approach was wasting budget, jeopardizing project timelines, and capping PowerMarket’s potential for growth.
That’s when we found a better way.
The Breakthrough: Predictive Modeling
That’s when we turned to a new option—using data science for predictive acquisition—and it transformed our approach.
Predictive modeling uses historical customer data, machine learning, and statistical algorithms to forecast, in our case, which households are most likely to enroll in community solar—before we ever reach them.
Instead of blanketing entire zip codes, we could pinpoint the top-performing prospects—sometimes just 2% of the population—and focus our efforts there. The result? Faster enrollment, lower acquisition costs, and more reliable project timelines. At the top end of our model, prospects are 5x more likely to convert than a random sample.
How does this work?
Working with a team of data scientists, we analyzed over 1,000 traits for each lead in our database—customers and non-customers alike—and compared those traits to the general U.S. population. These traits include:
After analyzing these traits, each individual is then assigned a predictive score from 1 to 100, representing their likelihood to convert. These scores are mapped to a lift curve—a tool that shows how much more likely someone in a given score bracket is to enroll compared to a random prospect.
Think of this tool like a treasure map. Without it, you’re wandering blind. With it, you know where to prospect. And with that in mind, we headed west—putting our model to the test on a major project in California.
Case Study: California & Dimension Renewable Energy
We applied this data-driven approach to win and fill a landmark community solar project in Southern California with Dimension Renewable Energy, which was launching the largest third-party-owned community solar project in the U.S.
The challenge: Enter a new market, target customers in a complex utility territory (Southern California Edison), and acquire thousands of customers—profitably.
Here’s how we did it:
Beyond California: A Repeatable Playbook
Predictive modeling is now central to how we operate across all markets. We've used this approach to:
By combining data-driven insights from AI and machine learning, we’ve moved from guesswork to science. And we’re just getting started.
Through experimentation and investment in AI and machine learning, we’ve transformed how we identify and convert prospects. Here’s how we did it—and what we’ve learned along the way.
Where Acquisition Channels Fall Short
With over a decade of experience in community solar, we’ve tested nearly every major acquisition channel in the industry. Each has strengths, but also clear limitations.
Door-to-Door Sales (D2D)
Still widely used across the industry, D2D has serious constraints: reputational risks, unpredictable customer volume, limited scalability, and potentially high compensation. While effective in specific cases, D2D alone can’t support PowerMarket’s required scale.
Facebook offers strong targeting capabilities—up to a point. Its algorithms don’t always provide sufficient customer targeting capabilities, and even when cost-effective, identifying who’s converting can feel like a black box.
Partnership Marketing
Municipal partnerships can be highly effective. A key challenge, however, is identifying which households are located within that community. Unfortunately, a lot of residential data (voter records, tax rolls, USPS change of address requests, etc.) is limited, inconsistent, or out of date—and municipalities are, for good reason, reluctant to share it. To make these partnerships viable, we had to prove we could generate accurate, actionable lists of local residents ourselves.
Direct Mail
Direct mail offers clear targeting advantages, but it’s expensive, slow to yield results, and extremely sensitive to list quality. We’ve seen performance range from $5 to $2,000 per acquired customer, largely depending on how the prospect list was built.
Traditionally, there are three ways to build these lists:
- USPS Every Door Direct Mail (EDDM): Great for broad, local targeting—but not precise enough for community solar, where deep customer segmentation and alignment along utility geographic areas are key.
- Data Brokers: Prepackaged lists from vendors like Experian, Acxiom, InfoUSA, or Epsilon are often outdated, subject to usage limitations, and imprecise.
- Manual Data Collection: Accurate, but tedious and limited in scale.
The Problem: Guesswork & Unpredictability
We were treating every household as if it were equally likely to convert—which we now know is far from true.
Renters behave differently than homeowners. Younger adults have different motivations than retirees. Income, location, and lifestyle all influence how people engage with community solar. A one-size-fits-all approach was wasting budget, jeopardizing project timelines, and capping PowerMarket’s potential for growth.
That’s when we found a better way.
The Breakthrough: Predictive Modeling
That’s when we turned to a new option—using data science for predictive acquisition—and it transformed our approach.
Predictive modeling uses historical customer data, machine learning, and statistical algorithms to forecast, in our case, which households are most likely to enroll in community solar—before we ever reach them.
Instead of blanketing entire zip codes, we could pinpoint the top-performing prospects—sometimes just 2% of the population—and focus our efforts there. The result? Faster enrollment, lower acquisition costs, and more reliable project timelines. At the top end of our model, prospects are 5x more likely to convert than a random sample.
How does this work?
Working with a team of data scientists, we analyzed over 1,000 traits for each lead in our database—customers and non-customers alike—and compared those traits to the general U.S. population. These traits include:
- Demographics (age, income, gender, homeownership status)
- Psychographics (values, interests)
- Behavioral data (purchase habits, payment histories)
- Geographic and physical attributes (roof type, zip code, utility territory)
After analyzing these traits, each individual is then assigned a predictive score from 1 to 100, representing their likelihood to convert. These scores are mapped to a lift curve—a tool that shows how much more likely someone in a given score bracket is to enroll compared to a random prospect.
Think of this tool like a treasure map. Without it, you’re wandering blind. With it, you know where to prospect. And with that in mind, we headed west—putting our model to the test on a major project in California.
Case Study: California & Dimension Renewable Energy
We applied this data-driven approach to win and fill a landmark community solar project in Southern California with Dimension Renewable Energy, which was launching the largest third-party-owned community solar project in the U.S.
The challenge: Enter a new market, target customers in a complex utility territory (Southern California Edison), and acquire thousands of customers—profitably.
Here’s how we did it:
- Define the Addressable Population
SCE’s service area spans millions, but utilities don’t follow clean zip code boundaries. So our engineering and data science teams used shapefiles to isolate exact addresses within SCE’s footprint.
We then removed households in Community Choice Aggregator (CCA) territories—because these customers must opt out of their CCA service to join community solar, which on average, only ~3–5% actually do. This cut millions from our addressable candidate pool.
Finally, we filtered for the head of household—the person most likely to make energy decisions—leaving us with just 27% of SCE’s total population to target. - Utility-wide direct mail
Using our model, we scored the remaining population and focused only on the top decile—just 2–3% of the total population, but with dramatically higher conversion potential. This meant fewer mailers, more conversions, and lower acquisition costs. - Municipal Engagement Strategy
With scores aggregated at the community level, we ranked all 430+ municipalities in SCE by likelihood to engage. We prioritized those with:- High aggregate scores
- Publicly stated interest in sustainability or energy savings
The results: The City of Long Beach and City of Corona both voted unanimously to support our program—enrolling municipal meters and actively promoting enrollment via press releases, social media, and direct mail. These efforts generated over 1,000 local customers. - Precision Targeting with Facebook
We also applied our model to Facebook, bypassing the platform’s default targeting and instead using our proprietary scores. This let us test niche segments, such as active-duty military, which acquired at just $35 per customer—our best-performing digital segment to date. Additionally, our model brought in customers at a lower customer acquisition cost than Facebook’s own model, demonstrating further scale.
Beyond California: A Repeatable Playbook
Predictive modeling is now central to how we operate across all markets. We've used this approach to:
- Forecast acquisition costs in new states or utilities
- Design channel mixes tailored to each project
- Build municipality-specific prospect lists that secure partnerships and trust
- Equip D2D reps with prioritized household lists for more efficient canvassing
By combining data-driven insights from AI and machine learning, we’ve moved from guesswork to science. And we’re just getting started.
Subscribe to get more insights on industry news, company updates, product solutions, and clean energy expertise.
.jpg)
.png)
