The case for LLMs in the field

Whether it's resolving an outage or just keeping up with critical grid and asset maintenance, speed and efficiency is of the essence. So it can be frustrating (or if you are without air conditioning in the summer, infuriating) to see field workers like linemen, technicians, and electricians sitting in their trucks typing on a computer. 

Of course, surfacing, capturing, and sharing data is a critical role of field workers. For example, they need to review work histories, update work orders, and request parts or materials to do their job. Yet while digitization of these processes has unlocked meaningful efficiencies in the back office, it has also shifted new, time-consuming responsibilities onto field crews at a time when aging infrastructure demands more frequent maintenance amid rising extreme weather. As a result, the hours field crews spend off the assets comes with a significant and growing opportunity cost when jobs that could have been resolved in a matter of hours or days stretch to weeks.

At Powerhouse Ventures, we get excited by opportunities for AI to help unlock valuable time and cost savings while tapping into an acute pain point by automating the “boring stuff.” In the case of field crews, ultimately many of their tedious, time-consuming tasks revolve around pulling information out of and putting it back into fragmented sources, the kinds of workflows where large language models (“LLMs”) excel.  

In this piece, we explore in more detail the hidden bottlenecks preventing field crews from actually “turning a wrench,” as well as opportunities for LLMs to unlock new efficiencies and time savings out in the field.

Key takeaways

• Field crews spend a significant share of their day on essential but time-intensive data extraction and entry, such as looking up work histories, updating work orders, coordinating with HQ, and tracking apprentices’ progress, leaving limited time for their actual hands-on maintenance work. Even small inefficiencies of field crews’ time can translate into millions of dollars annually across a large workforce.

• Delays on field projects directly translate into reduced asset uptime and operational losses: even just a three-day outage at a 100 MW solar plant can exceed $100,000 in losses.

• LLMs are well-suited to reduce this burden, especially given that most of these tasks involve pulling text-based information from and back into fragmented sources.

• Field crews prefer mobile, voice-based tools that are simple and intuitive: crews don’t want to pull out a computer, take off their gloves, and type or click through screens. 

• There are four major pain points in energy field operations with clear opportunities for LLMs: accessing information in the field, documenting work in real-time, preserving institutional knowledge as veteran technicians retire, and tracking the progress of a growing apprentice workforce.

• LLM solutions should fit into the tools crews already use: field crews are fatigued by new standalone software, so the most effective solutions will integrate seamlessly into existing systems rather than require them to adopt another separate tool.

If you’re a seed-stage founder developing digital solutions to unlock efficiencies for field crews, or if you’ve been looking at the intersection of AI and field workers in the energy sector, we’d love to hear from you! And if you know someone building in this space, please tag them in the comments. Contact us.

First, some realities of building digital solutions for energy field crews

1. Don’t build another standalone SaaS tool

Not even a decade ago, many crews were still doing everything on paper. Since then, utilities and IPPs have adopted a patchwork of digital tools, leaving field crews and supervisors fatigued and understandably wary of promises that the next software or app will “magically” fix their problems. It is therefore critical that any new digital tool for field crews be simple, intuitive, and integrated into existing systems—not another standalone tool to learn and keep up with.

Taking this into consideration, the opportunities we outline below should be viewed not as ideas for standalone point solutions, but as features of an intelligence layer that sits on top of organizations’ existing field operations systems. Our conversations with industry experts have validated this approach, with some utilities and IPPs experimenting with in-house AI co-pilots that integrate with software they already use rather than adopting entirely new platforms.

2. Field crews need their hands to work

Across our expert conversations, we consistently heard that typing is a major friction point for field crews. Workers don’t want to pause, remove their gloves, and pull out a laptop or tablet while on site, and they’re not quite ready for augmented reality smart glasses either. Instead, there is a strong appetite for mobile, voice-based tools they can operate using natural language.

3. Digital transformation is hard

LLMs only reach their full potential once an organization has completed the foundational work of harmonizing its data. This is no small task for utilities and IPPs, which contend with data silos, strict security, safety and compliance requirements, and a web of legacy systems to integrate. These broader digital transformation challenges are important, but they are not the focus of this article.

How LLMs can unlock valuable time for field crews

While the specifics vary by team, highly skilled field workers spend a significant share of their day 1) accessing information across fragmented sources and 2) inputting information into multiple systems. In addition, they face growing workforce challenges, including 3) the loss of institutional knowledge as experienced workers retire, and 4) the need to safely and efficiently train a new generation. Across these four pain points, there are several clear opportunities for LLMs to unlock value and give crews more time for the field work itself.

Pain Point #1: Accessing information from fragmented sources

While out on a job, field crews often need to look up and reference information to complete their work. This ranges from surfacing the asset’s work history to see where the last technician left off, to interpreting error codes on equipment, or reviewing instructions from engineers at HQ. The information they need is scattered across disparate sources, such as manuals, work histories, work orders, contracts, or emails, and is sometimes living inside people’s heads.

In some cases, experts shared it can take weeks of back and forth with HQ and several visits to a remote site before field crews have all the information they need to do a job. What could have been resolved in a matter of days stretches to weeks because of hard-to-access information. These delays can translate directly into significant losses at the grid or asset level. For example, using Lawrence Berkeley National Laboratory’s Interruption Cost Estimate Calculator, we estimated an incremental day of outage for 100,000 residential customers and 2,000 businesses in California would have an economic cost of over $30 million. At an asset level, if a 100 MW solar project is down for just three days longer than it needs to be, losses could exceed $100,000, assuming a 25 percent capacity factor and a power price of $60/MWh.  

Incomplete information is not only a time bottleneck, but also a safety hazard. Especially at high-risk sites like substations or storm-damaged assets, even a small lapse in information could lead to error-prone decisions with serious consequences. 

a. Opportunity: LLM-powered field knowledge assistant

Just like white-collar workers can reap the benefits of ChatGPT-like tools to surface and summarize key information, so can field workers. A field supervisor at a large utility told us it would be “amazing” if his crew had a ChatGPT-like assistant to help them get the information they need faster. As noted above, the ideal interface would be voice-based, allowing crews to speak to the tool in natural language.

There is also a meaningful opportunity for further time savings if this assistant can generate briefings for crews to listen to during their long drives to job sites, such as a history of recurring issues or the most recent updates to the asset.

To ensure accuracy and avoid hallucinations, an LLM assistant must ground its responses in an organization’s internal systems and documents.

Real-world examples have quantified the potential efficiency gains from implementing a solution like this. For example, a large renewable energy operator worked with Field Service AI by BCG X to deploy a GenAI-powered knowledge assistant, and they measured a 15 to 20 percent reduction in job duration, increasing asset uptime, and a five to ten percent boost in daily productivity for technicians and field leaders.

Startup spotlight: 

• TeamSolve provides a digital “Knowledge Twin” to operations teams across utilities, industrial facilities, and commercial and residential building managers, offering field teams on-demand access to procedures, manuals, and site information. According to a case study with a leading private water operator in the Philippines, implementing the Knowledge Twin led to over 20 percent asset performance improvement and 50 percent time savings. 

b. Opportunity: Improve communications with back-office teams at HQ

LLMs can also help reduce unnecessary back-and-forth between field crews and back-office teams. When crews arrive on site, they may be gathering new information that engineering teams at HQ need to diagnose the issue, or they may find conditions are different than they expected. Either way, they require input from HQ to do their job.

LLM tools (and computer vision tools—though not the focus of this article) can help field crews capture key data and send it quickly back to their back-office teams, getting responses faster, reducing downtime and truck rolls. LLM-powered tools may also be able to help back-office teams make sure that subtle details or time sensitive items aren’t lost in emails or buried in systems.

However, one important constraint is that some energy assets, such as utility-scale solar, wind, storage, and transmission infrastructure, are often located in remote areas with limited connectivity, so “lightweight” solutions that can work with limited cell service or have offline capabilities are key. 

Startup spotlight:

• Twindo leverages AI and intelligent automation to streamline operations for renewable energy teams, helping reduce frictions between the field and back office. One of Twindo’s customers, a wind turbine services provider, explained that Twindo helps their back office teams “receive [reports from field technicians] faster and can react…[helping to] avoid unnecessary revisits and reworks on assets which is a huge plus for us and the end clients.”

Pain Point #2: Inputting information into multiple systems

Documentation is critical to the success of field crews but it can be a tedious process. To ensure projects run smoothly from inception through closeout, workers must document a wide range of activities. For example, tracking their daily progress, updating and closing work orders in a timely fashion, ordering spare parts and materials, and communicating with engineering, accounting, finance, and other teams at HQ.

Today, many field crews are manually typing messy notes and clicking through slow, clunky, complex workflows in legacy field software. If their system is mobile-compatible, they might be able to use a table or smartphone, but often it requires pulling out a laptop—making accurate data capture on the fly nearly impossible. As a result, crews frequently log information after the fact, sometimes from memory, which leads to inadequate records and leaves the next crew with gaps that force them to rely on guesswork or start from scratch. 

Moreover, field crews’ time is expensive, making slow and inefficient data entry a significant cost driver for utilities. For example, the median annual wage for electrical power-line installers and repairers is almost $100,000, or nearly $50 per hour. Ideally, those wages should mostly be spent on skilled field work, not data entry. Yet just 30 minutes per shift lost to inefficient documentation can cost utilities hundreds of dollars per crew per week, or millions annually when scaled across a large workforce, before accounting for downstream delays and reworks caused by poor data quality. At a national level, the costs are substantial: with approximately 127,400 electrical power-line installers and repairers, utilities spend roughly $12 billion per year on their salaries alone. And because this figure does not capture other field roles (such as solar and wind technicians) as well as non-labor costs like truck rolls, it understates total fieldwork costs.

a. Opportunity: Field documentation agent

LLMs are uniquely well-suited to process, structure, and clean up incomplete, messy or typo-ridden textual information. Field crews no longer need to spend valuable time making sure they enter the right details into the right fields or click through a series of boxes just to move a workflow forward.

Here, an LLM-powered agent could auto-draft work order updates from voice notes recorded on-site, transcribing them and structuring the information directly into the required fields within existing software. And if this agent has access to the same sources as the field knowledge assistant described above, it could retrieve missing details from internal systems to enhance the documentation. For instance, it could identify the technical name of a particular part from a spoken description and automatically generate a corresponding parts order request for HQ. As a real-world example, Twindo’s AI platform (mentioned above), also helps field crews simplify and automate data entry, with a customer claiming the platform helped them reduce reporting time by up to 50 percent.

Looking further upstream, more advanced agentic AI could help back-office teams orchestrate their workflows, such as ordering spare parts, more quickly so materials reach field crews sooner.

Startup spotlight:

• Partium helps technicians and supply chain teams across industries make ordering spare parts smooth and hassle-free by providing instant part identification using computer vision, enriching messy and missing data, and sourcing parts replacements. Partium claims their platform enables up to 70 percent faster part identification, 40 percent fewer service errors, and up to 35 percent cost savings by identifying cross-compatible alternatives.

Pain Point #3: Loss of institutional knowledge to retirement

Another major challenge utilities and IPPs are facing is that institutional knowledge is literally “walking out the door” as droves of their experienced, highly skilled field technicians retire. According to the International Energy Agency, advanced economies have 2.4 workers within a decade of retirement for every worker under 25 years old, and North America has the highest share of energy workers aged 55 or older. A lot of mission-critical institutional knowledge around how to safely perform maintenance or resolve specific quirks of aging infrastructure mostly lives in the heads of these veteran operators who have spent their careers working on these highly complex assets. This is an urgent issue in the energy sector, but it is challenging to capture all the nuances of tacit knowledge and instinct accumulated over decades of lived experience in a scalable way.

1. Opportunity: Voice-based knowledge capture agent  

Utilities and IPPs urgently need easy, scalable ways to capture the institutional knowledge that lives in their veteran workforce’s heads. As noted earlier, technicians don’t want to (and don’t have time to) sit at a computer typing long, comprehensive notes.

Here, we see another opportunity for LLMs to engage with field crews more naturally by “talking to them” through voice interviews, extracting nuanced details with targeted questions and clarifying follow-ups. For example, at the end of the day, a technician could answer a short series of curated questions generated by an LLM-powered voice agent to record what they did, what they tried, what worked, what didn’t, and importantly why. By allowing crews to speak freely, there’s a higher chance they’ll surface small nuances or key details they might otherwise leave out when constrained to typing. This collected knowledge could then be consolidated and structured, creating an evolving, organization-wide understanding of how assets behave and how experienced operators diagnose issues, while making it easy for others to query and retrieve those insights later.

Startup spotlight:

• Intuigence AI, while focused on industrial manufacturing, offers capabilities such as using an AI copilot to conduct expert interviews with retiring engineers to capture institutional knowledge before it walks out the door and structures it into dynamic, queryable knowledge graphs.

Pain Point #4: Training a new generation of field workers

As seasoned field technicians retire, utilities and IPPs face an urgent need to maintain a strong pipeline of qualified workers to take their place. However, the energy industry is grappling with a major skills gap, as more than 76 percent of energy employers report difficulty finding qualified workers. This shortage underscores the need to train and develop a new generation of field crews through expanded training programs and apprenticeships.

However, building a workforce that can safely operate independently takes years, and long timelines are further strained by attrition. For instance, recent apprenticeship completion rates in the US have fallen below 35 percent overall.

1. Opportunity: Apprentice training progress agent

There are countless ways LLMs can support upskilling, training, and workforce development, from generating personalized training modules to simulating real world troubleshooting scenarios and curating micro-quizzes and lessons. Applying AI to workforce development could be an entire deep-dive article on its own.

Here, however, we highlight a perhaps overlooked opportunity: using LLMs to help crew leaders keep track of the many apprentices they oversee and where each one stands in their training.

Experts emphasized that when apprentices are out in the field (often at high-risk sites), the top priority for crew leaders is keeping everyone safe. Yet as more veteran workers retire, crews increasingly find themselves imbalanced, sometimes with more apprentices on-site than trained operators. This dynamic frequently slows down work, as experienced technicians must spend more time teaching and monitoring safety than completing the job itself. With so many apprentices to support, crew leaders struggle to keep track of who has learned what and which skills each apprentice has mastered, leading to redundancies and delays.

An LLM-powered assistant could track each apprentice’s development by capturing what they completed each day, such as through short voice-based interviews, and translating those insights into a live training record that updates automatically as skills are mastered. Crew leaders would gain a real-time view of who is progressing and who needs support, without relying on memory or scattered notes to understand who is qualified for which tasks.

Summarizing where we see opportunities for LLMs in energy field operations at Powerhouse Ventures

At Powerhouse Ventures, we believe AI tools can generate value and find strong customer pull by automating the “boring stuff”: the behind-the-scenes work that feels tedious and isn’t core to a worker’s expertise but is still essential. 

In the energy sector, we see enormous potential for LLMs to reduce the time field crews spend on data extraction, data entry, and managing a rapidly changing workforce. The founders who understand these pain points deeply and build solutions that are simple, intuitive, and seamlessly integrated into existing systems will help define the AI-enabled era of field operations.

If you’re a seed-stage founder developing digital solutions to unlock efficiencies for field crews, or if you’ve been looking at the intersection of AI and field workers in the energy sector, we’d love to hear from you! And if you know someone building in this space, please tag them in the comments. Contact us.