Artificial Intelligence & Aurora Solar Report (Second Edition)

Introduction

AI has the potential to be the great equalizer. We have opportunities ahead of us to address pain points (in healthcare and climate) and to address the sustainable development goals.

Over the past year, significant advancements have been made in AI, enhancing its ability to generate more natural and nuanced language, create realistic photos and videos from text prompts, and apply its capabilities in diverse fields such as medicine and science. In climate tech, AI holds the promise of expediting material science and technological breakthroughs, securing regulatory and financial support by clearly demonstrating utility efficiency gains, and enabling smarter grid scaling as global energy demand grows exponentially.

The solar industry has faced significant headwinds over the past 12–18 months, underscoring the growing need for new tools to address these challenges. NEM 3.0 continues to hinder the California residential solar market, which experienced a 40% decline in 2024 (Wood Mackenize, US Solar Forecast). California’s pro-utility position has also led other states — including Illinois, Hawaii, and Idaho — to implement similar cuts to consumer incentives, and there is no indication the industry will return to consumer-friendly net-metering policies.

Interest rates remain higher than expected due to extended inflation compounding the impact of decreased incentives and tariff-related pressures, placing significant operational stress on solar companies to shift from a primarily financing-based model to a leasing model. As a result, the Solar Energy Industries Association (SEIA) has adjusted its forecasts downward. And although hiring sentiment has improved year-over-year, 70% of solar companies still report that it is difficult to find qualified talent, putting pressure on companies to train faster or augment their workforce with other tools. Given these economic pressures, companies are looking for reliable tools and processes to maintain or increase their ability to serve their customers and grow their business. 

In response to these challenges and the increased demand for cost-effective tools, Aurora continues to significantly boost its investment in AI and automation products. In the last year, the training dataset for Aurora AI has quadrupled in size. Thanks to this expanded dataset and several improvements to key infrastructure and training techniques, Aurora AI can generate a 3D model in under 10 seconds, determine obstructions with 25% more accuracy, and has brought AI SmartRoof to parity with AI Roof faces. 

• 4x

  more training data

• 25%

  more accurate obstruction detection

• 1:1

  performance parity between AI Smart roof and AI Roof Faces

Aurora AI has become an essential tool for solar companies, which have used it over 3 million times. Plus, there has been a shift in how customers leverage Aurora AI today, with more customers using it throughout their sales and design process rather than mostly as a pre-sales tool.

Update — Aurora AI’s accuracy and speed

Aurora continues to invest in the effectiveness of its AI and automation products. Similar to a human designer, Aurora AI produces higher-quality designs when higher-quality data is available, and so it leverages the best satellite and LIDAR data for each project location.

Most recently, Aurora has increased the training data for Aurora AI by a factor of four, from approximately 500,000 to 2 million designs. The increase in training data and multiple updates and improvements to the model infrastructure have yielded a significant increase in accuracy and decrease in run times.

How we measure accuracy and speed

There are several ways to validate the accuracy of a product like Aurora AI. This paper outlines two approaches, both of which compare how well Aurora AI performs relative to a human designer.

The first approach, called “Relative Accuracy,” measures the total number of panels that can fit on a roof as generated by Aurora AI and as drawn by a human designer. Then the two sets of panel counts are compared to the original design. This method was introduced in response to customer feedback, asking for real-world measurements. The closer the AI-generated design and the human-drawn design panel counts were, the better Aurora AI was considered to perform.

The second approach is called “Percent Match” and is new to this year’s report. It helps explain the incremental improvements we made with obstruction detection as a result of the increased training data and improved model infrastructure. Percent Match measures how accurate Aurora AI’s measurements are relative to a validated dataset (3D models drawn by EDS). Specifically, for this year’s assessment of obstruction detection, the distances between the obstructions were measured. The closer Aurora AI’s distances between obstructions matched the human-drawn EDS models, the better the Percent Match.

Both metrics measure the difference in accuracy in Aurora AI and Human Model 2 (EDS 2) relative to the Human Model 1 (EDS 1) all for the same address.
Relative Accuracy measures the difference in panels that can be placed.
Percent Match measures the difference in area measured.

Why report on two accuracy metrics?

This report includes both metrics for several reasons. Last year, our product development teams focused on simplifying our Aurora AI product offering and expanding its use cases. We then used Relative Accuracy to evaluate how our two Aurora AI products (i.e., AI SmartRoof and AI Roof faces) performed relative to humans and to each other.

Percent Match serves two purposes. First, it allows us to measure the general incremental improvements Aurora AI has made in response to the significant increase in training data and improved model infrastructure over time. It also helps us determine whether obstruction detection has improved, and by how much.

Speed

Speed is another critical requirement for solar sales and designs. Across the solar sales and design market, 3D models can be drawn manually, purchased as a service, or — for Aurora customers — generated automatically through AI. Solar design services vary in turnaround times from 30 minutes to 24 hours, depending on the service and how much a customer is willing to spend per project. Our engineers closely monitor how quickly Aurora AI can generate a roof model when a customer uses the feature.

How well Aurora AI performed

Relative accuracy

AI SmartRoof’s performance has significantly improved this year due to recent upgrades, and it now exceeds the performance of AI Roof faces. AI SmartRoof performed better than, equal to, or within a narrow margin of human performance 79.1% of the time, compared to 77.1% for Roof faces (Table 1). 

Version	Product	Better or Equal	Withing a Narrow Margin	Total
v2	AI Roof Faces	59.6%	17.5%	77.1%
v2	AI SmartRoof	60.1%	17.8%	77.9%

Table 1: realtive Accuracy of AI Roof Faces and AI SmartRoof

Percent Match for obstruction detection

Aurora AI’s ability to detect and classify obstructions also improved significantly (Image 1).

Thanks to the latest upgrades, Aurora AI demonstrated a 26% improvement in classification and a 27% improvement in its ability to detect obstructions relative to the previous version (Table 2).

Metric	v1	v2	Improvement
Misclassification	11%	8%	26%
Missed Obstructions	25%	18%	27%

Table 2: Relative Accuracy of AI Roof Faces and AI SmartRoof
Note: Obstruction Detection uses the same neural networks for both AI Roof faces and AI SmartRoof, so the results apply to both products.

Speed

With regards to speed, our ongoing analysis of Aurora AI’s run times shows that recent upgrades have had a significant impact. It now takes less than 10 seconds to run Aurora AI, on average, compared to almost 30 seconds just over two years ago (Figure 1). This represents an improvement of over 65% during that time period.

Analysis of how sales and design teams use Aurora AI

Aurora AI continues to exceed expectations. As its accuracy and speed improve, more customers are adopting and using Aurora AI for an increasing number of projects. It has now been run over 3 million times and grew more than 35% last year, despite a 25% contraction in the market.

• 3M

  runs total

• 35%

  growth YOY

• 47%

  Sales Mode

• 53%

  Design Mode

What’s even more interesting is that not only is the overall usage of Aurora AI growing, but that usage is shifting from being mostly sales-driven to mostly design-driven. Last year, most customers used Aurora AI in Sales Mode, taking advantage of its speed and accuracy early in the solar sales cycle to provide homeowners with the quickest, most accurate, and most affordable 3D models. However, this year, the ratio has flipped to a majority in Design Mode usage.

Customer case study

Aurora AI’s evolution has been remarkable. What once required CAD expertise is now accessible to our entire team, empowering newer sales reps to create stunning system designs.

The company

In its 20 years serving the San Francisco Bay Area, Simply Solar is a local leader in providing custom solar solutions. Committed to delivering premium systems, the company focuses on exceptional customer service and ensures every project meets the highest standards. Supported by a team of in-house professionals, Simply Solar is dedicated to precision, quality, and sustainable energy solutions.

The challenge

Despite their expertise, Simply Solar faced significant operational challenges. Without AI-powered tools, design tasks were time-intensive and prone to errors, which limited their ability to scale and slowed project timelines. Additionally, the absence of automation in proposal and design workflows created inefficiencies, causing delays in critical stages of projects and impacting overall productivity.

The solution

With the implementation of Aurora’s AI-powered tools, Simply Solar dramatically increased their efficiency. Design tasks that once took 15 minutes were reduced to just 15 seconds, dramatically speeding up project timelines. AI automation also seamlessly integrated into their workflow, streamlining proposals and designs, which not only accelerated project turnarounds but also enhanced accuracy. Most importantly, these AI-driven processes enabled Simply Solar to handle peak demand effortlessly, ensuring quality and consistency across all projects.

• 60x

  Sales reps can design a system 60x faster with Aurora AI

• 20x

  Designers can design a system in 15 seconds, 20 times faster than the pervious process

• 1

  With Aurora AI, one designer can accomplish the workload of 2-3 individuals

With Aurora AI, what once consumed 15 minutes of a salesperson’s or designer’s time can now be accomplished in just 15 seconds. This remarkable boost in efficiency has streamlined our operations and empowered our team to meet customer needs and operate more effectively.

The vision for AI at Aurora

2024 vision review

2024 was a challenging year for the solar industry. However, we have faced tougher times and will continue to persevere as we find new and better ways to meet the growing demand for affordable and reliable electricity. Aurora’s goal is to empower a future of solar for all. The only way to achieve this is by measuring our progress. Here’s a look at how we performed relative to the promises we made last year:

Objective	Status
Applying improved land parcel data to improve automatic identification of property boundaries	🟢
Updating the US LIDAR database with the most recent and high resolution data available	🟢
Increasing model training efficiencies and accuracy by consuming more training data faster	🟢
Streamlining product features by consolidating AI SmartRoof and AI Roof faces	🟡
Expanding Aurora Al’s capabilities to more markets outside of the US	🟢
Exploring and investing in more solar sales- and design-specific Al use cases	🟡

What we will achieve in 2025

As Aurora continues to lead the way in solar innovation, we remain steadfast in our commitment to making Aurora AI as fast and accurate as possible. We know this is key to installing more solar on more roofs, faster. Looking ahead, in addition to our ongoing focus on speed and accuracy improvements, we will explore new features and capabilities that will bring even more value to our customers and homeowners. Here are just a few things you can look forward to in 2025:

• Making Aurora AI available via API (released)
• Continuing to update and maintain LIDAR databases with the most recent and highest resolution data available
• Finalizing the consolidation and streamlining of the AI SmartRoof and AI Roof faces products
• Exploring and implementing (if appropriate) the ability for customers to work around outdated LIDAR
• Exploring and investing in more solar sales- and design-specific AI use cases

Success is a work in progress. It’s not about achieving a goal; it’s about constantly improving and pushing boundaries