AI for Solar Installation: Five Jobs a Small Installer Can Automate

A small solar installer can use AI today for lead qualification, survey-note write-ups, proposal wording, permit paperwork drafts and customer updates. Start with whichever job eats most hours or loses most enquiries. Engineering, design sign-off and safety checks stay with qualified people, because the roof and the electrics do not forgive a confident mistake.

Most pages on this subject cover AI in the energy grid or in panel manufacturing. A four-person installer has none of those problems. It has a phone that rings during a roof job, a pile of survey photos on a phone, proposals that take an evening each and customers asking when the scaffold is coming.

Follow me on Instagram@sagnikteaches

The hours below are illustrative, for a made-up team of four with about 20 enquiries a week, eight site surveys and proposals a month, six installations a month and around 15 signed jobs waiting to be installed at any one time. Time your own week before you believe any of them.

Connect on LinkedInSagnik Bhattacharya

The five jobs at a glance, with hours saved

That adds up to roughly 50 to 60 hours a month on paper, which is more than a full-time admin week. In practice you will recover less, because every draft needs a read, and the first month is spent setting up. A sensible aim for the first quarter is half of the table. The order of the table is the order I would work in, with one adjustment: if proposals are your bottleneck, jump to job 3 early, with the number-checking rule in place.

Subscribe on YouTube@sagnikteaches

Job 1: qualify leads before anyone drives anywhere

A solar enquiry is only worth a site visit if a few facts line up: the roof, the shade, the electricity use, who owns the property and whether the customer is serious. Each one can be asked by message and sorted by AI, and the answer decides whether a surveyor spends two hours on site.

The qualifying questions, in the order a homeowner can answer them:

1. Do you own the property? (Tenants and leaseholders need the owner's or freeholder's agreement.)
2. What is the roof made of, and roughly which direction does the main slope face?
3. Is there shade from trees, buildings or chimneys during the day?
4. What is your typical monthly electricity bill, or can you send a recent one?
5. Are you thinking of a battery, an electric car charger or a heat pump in the next two years?
6. When would you like it installed, and have you had other quotes?

You help a small solar installer sort enquiries. For each enquiry I paste, return: (1) a one-line summary, (2) a score of hot, warm or cold using my rules below, (3) the two facts still missing, (4) a draft reply of under 90 words that asks for those facts and offers a 15-minute call. Never quote a price, a saving or a payback time. Never say a roof is suitable. If the property is not owned by the writer, say we need the owner's agreement.
Rules: hot = owner, usable roof described, monthly bill over $120, wants install within six months. cold = tenant without owner agreement, or no roof access, or "just browsing".
Enquiry: [paste, with name and address removed]

An illustrative case. An enquiry reads: "Semi in a village, big south-facing roof, bills about $180 a month, thinking of an EV next year, want it done before winter." The output scores it hot, lists "roof material" and "any shade" as missing, and drafts a reply asking for a photo of the roof and a recent bill, offering Tuesday or Thursday for a call. A person reads it, approves and sends it in under a minute. Before, the same lead waited until the evening, by which time leads like it had sometimes booked another installer's survey.

The check: after a month, compare how many hot leads converted to surveys against your old figure. If the scores are wrong, change the rules, not the model. A longer treatment sits in the chatbot tutorial for qualifying leads by bill, roof and shade.

Job 2: survey notes into a clean report

A surveyor comes back with 40 photos and a page of shorthand: "roof N 32deg, tiles concrete, 2 broken, DB in garage, space for 2 ways, meter ok". The report that goes to the designer and the customer needs those turned into full sentences and ordered sections. AI is good at ordering and wording. It is not good at knowing whether "space for 2 ways" means what you meant.

Turn these site-survey notes into a report with headings: Roof, Electrical, Access, Shading, Photos needed. Use only what I wrote. Where something is unclear or missing, list it under "Questions for the surveyor" instead of guessing. Keep measurements exactly as written.
Notes: roof N 32deg, tiles concrete, 2 broken near ridge, DB in garage, space for 2 ways, meter ok, scaffold front only, big tree at rear 8m approx, no loft access hatch seen.

An illustrative output lists the roof pitch and covering, flags "broken tiles near the ridge" as a pre-install repair question, notes that the notes say "N" without saying whether that is the roof direction or the slope facing, and lists "loft access: not seen, confirm" under questions. The surveyor fixes the direction note, which could have put the array on the wrong slope. That one catch is the argument for the questions list.

Two safeguards. First, never ask the assistant to complete missing measurements. Second, keep the photo list as a checklist the surveyor ticks on site, so nothing is discovered missing at the desk. A matching approach with a full checklist is in the site survey checklist tutorial.

Job 3: proposal wording, with numbers you supply

Design tools produce the layout, the system size and the yield estimate. What they leave you is the covering email, the explanation of each option, the answers to common questions and the follow-up. That is the job for AI, and it is the only one in this list where the biggest risk is a number that looks plausible and is wrong.

The rule is simple and strict. The assistant gets numbers from you and never produces them. The text it writes explains those numbers in plain language.

Write a customer-friendly explanation of this proposal in under 250 words. Use only the figures below; do not calculate or estimate anything else. If a figure you need is missing, write [MISSING].
Figures from our design tool: 12 panels, 5.4 kW system, estimated yearly generation 4,900 kWh, system price $14,200, battery option 5 kWh at $3,900 extra, estimated first-year saving $820 (assumes current tariff, customer's usage profile).
Tone: calm, no hype, no promise of savings. Include one sentence saying estimates depend on weather, usage and tariffs.

An illustrative output explains what 5.4 kW means in terms of a typical household, states the two prices, describes the battery as optional and says the savings figure is an estimate under stated assumptions. What you would fix: it says "should pay for itself in about 17 years". That was not in your figures, since $14,200 divided by $820 is a calculation it did on its own, and it ignores tariffs and degradation. Delete it, or add your own payback method if you use one and are prepared to explain it.

The mistake to guard against is a unit slip: kW for kWh, or kWp for AC output. After each draft, read every number and its unit against the design tool. If you want more on adding AI to proposals you already send, see AI-written cover letters and FAQs for solar proposals. For batteries, never let a draft imply backup power unless the design supports it.

Job 4: permit and grid paperwork drafts

Permit and connection applications are forms and short narratives. They ask for the system size, equipment models, inverter details, roof type, site address and a description of the work. Much of it is copy-and-paste from design output, and AI can assemble a draft from your own data. It must never be the author of a technical fact.

The workflow is deliberate:

1. Collect the data from your own sources: design output, equipment datasheets, survey report.
2. Paste it into the assistant with the form's field list, and ask for a draft in the form's own order, with unknown fields marked [NEEDED].
3. Ask for a second list: any field where two sources disagree, for example a panel count in the design and in the order.
4. A named person checks every field against the source documents, completes the [NEEDED] items and signs or submits.

An illustrative case: the assistant fills 22 of 26 fields from the pasted data, marks four as [NEEDED] (inverter serial prefix, main breaker rating, conduit route, property type), and notes that the survey says "12 panels" while the order says 14. The discrepancy alone could save a rejected application.

What stays human is large. Requirements differ by authority and change over time, so the checklist of required documents comes from the authority's current page, not from an AI. Electrical compliance statements are signed by the person responsible. Treat the assistant like a keen junior who prepares the pack and never submits it. A closer look at narratives and checklists is in the solar permit application tutorial.

Job 5: customer updates that stop the "where are we" calls

Between signing and switch-on, a solar customer waits through site survey sign-off, permit decisions, equipment delivery, scaffold, installation, inspection and connection. Each gap produces a call. A short update at each stage prevents most of them.

Define the stages in your CRM or a simple spreadsheet, and write a template for each. AI helps in two ways: drafting the templates in your voice, and tailoring each send with job details.

Write a short, friendly text for the stage "install date set". Fields: [first name], [date], [arrival window], [crew lead first name], [what to clear]. Under 70 words. Mention that dates can move for weather and say we will tell them as soon as we know. No technical jargon.

An illustrative output: "Hi [first name], your installation is booked for [date], with the crew arriving between [arrival window]. [crew lead first name] will lead the team. Please clear [what to clear] beforehand. Dates can move if the weather turns, and if that happens we will tell you as soon as we know." A person reads it, checks the date and sends.

The delay message is the one never to automate. Customers forgive a late install and do not forgive silence, and the wording depends on the reason. Give the assistant the reason and the new plan, and have it draft the message for a person to send.

What the five jobs look like for a team of four

Here is how the table's hours were reached for the illustrative installer: an owner, a designer-surveyor, a part-time office person and an installation lead. They take about 80 enquiries a month, survey and quote 8, install 6 and carry around 15 signed jobs in the queue.

• Lead qualification. 80 enquiries at about 22 minutes each, counting the reply, the back-and-forth for missing facts and the CRM entry, comes to roughly 30 hours. With a draft to read and approve, about 8 minutes each, it falls to around 11 hours: a saving of about 19. On top of that, the surveyor used to visit around 12 roofs a month; screening out tenants without the owner's agreement and heavily shaded roofs by message brings it to 8, and four fewer visits at about 2.5 hours each with travel is another 10 hours of surveyor time.
• Survey notes. 8 reports at 2 hours of write-up drop to about 1 hour each with a draft to check. Savings: about 8 hours.
• Proposals. 8 proposals at 3 hours of writing fall to about 1.5 hours each with the numbers pasted in and checked. Savings: about 12 hours.
• Permits. 6 packs at about 2 hours 20 minutes fall to about 1 hour 25 minutes. Savings: 5 to 6 hours.
• Updates. 15 live jobs at roughly 80 minutes a month each of calls and messages is 20 hours. Stage messages that arrive before the customer has to ask cut the calls by about half. Savings: about 10 hours.

These are made-up figures for illustration. What they show is where the effort lies. The biggest single line is lead qualification, and part of its value never shows up in office hours at all: it is the surveyor not driving to a roof that was never going to work. Time your own steps for two weeks, and compare.

Two edge cases that test your rules

Rules look fine until an odd enquiry arrives. Run these two through your set-up before you trust it, and write down what you want to happen.

The after-hours enquiry from a tenant. A message arrives at 10pm: "Can I get solar? I rent the flat and pay the bills." The qualification rules say a tenant without the owner's agreement is cold, but a flat-out refusal wastes a lead who might be a landlord's contact. A better rule: score it cold for a survey, but reply with a polite note that you would need the owner's agreement, and offer to talk to the owner directly. The draft should say that in 60 words, with no promise it will be possible. If your assistant scored it hot because the bill was large, tighten the rule so ownership is checked first.

The customer who already has a quote. "Another company quoted $13,000 for 5 kW, can you beat it?" Never let an assistant respond with a price or a comparison. The right draft thanks them, asks to see the quote's equipment list and warranty terms, and offers a call, because two quotes at the same headline price can differ in panels, inverter, mounting and warranty. If your assistant drafts a discount, delete it; pricing is the owner's decision.

Choosing between a chat assistant and AI built into your tools

You have two routes, and they are not exclusive. A general chat assistant such as ChatGPT or Claude is flexible and cheap, but it only knows what you paste in and cannot update your CRM. AI built into a CRM or design tool sits closer to the data, so it can pull job details itself, but each feature has its own limits, and some come with a higher plan or an extra fee. Where a vendor page does not print a price, treat the feature as not yet costed and ask for it in writing.

A practical way to choose: start with a chat assistant and a Project for the job you are testing, because it costs nothing in switching, and see whether the output earns its place. Once the rules are stable and you are pasting the same data every day, look at whether a built-in feature does the same thing with less copying. The rules you wrote for the chat assistant move with you, since they are plain text. That is the real asset in this exercise: a written set of rules about how your firm qualifies, quotes and communicates, which any tool can follow.

What stays with an engineer or a qualified person

• System design sign-off: string sizing, voltage limits, inverter matching, protection devices.
• Roof assessment: structure, load, condition, fixings and weatherproofing.
• Electrical work and its certification, including the final inspection and commissioning tests.
• Any statement about fire safety, building regulations or safe working at height.
• Savings and payback promises. State assumptions, and never promise outcomes.
• Anything about warranty terms. Copy from the manufacturer's documents and your own terms.

An assistant can help check a design against a written checklist you supply, and spot items you forgot, but it does not replace the person who signs. If an AI-generated sentence touches any of these topics, delete it or have the right person rewrite it.

Where each of the five installer jobs slips

• Qualification scores drift. You tweak the rules and forget to record why. Keep a dated note of each change, and re-test on ten past enquiries.
• Survey notes lose a caveat. "Possibly asbestos cement sheets, not tested" becomes "cement sheets". Add a rule: never remove words like possibly, not tested or unclear.
• Proposals carry a made-up number. Check every figure against the design tool, every time. A mismatch means the whole draft is suspect.
• Permit fields copied from an old job. The model reuses a previous inverter model. Start every pack from a blank prompt with only this job's data.
• Updates go to the wrong customer. Merge-field mix-ups happen. Send the first three of any new template to yourself, then to a customer.

An installer's first four weeks, one job at a time

1. Week 1. Time the five jobs for one real week. Choose the one with the worst ratio of hours to value.
2. Week 2. Write the rules block for that job, with two real examples. Run it on ten past cases and compare with what you actually did.
3. Week 3. Go live on new cases, with every output read before use. Note edits.
4. Week 4. Update the rules from the edits. Add the second job if the first is calm.

A general chat assistant with a Project (a saved space for instructions and files) is enough for all five jobs. Some CRMs and design tools offer built-in AI; check what each feature does, what it costs and whether it needs a higher plan. OpenSolar's core app is free, for example, but its API access (a per-project fee) and its connectors (a monthly subscription) are paid add-ons with prices that vary by region, and other design and CRM tools also price AI features by plan, so confirm before you budget. To compare the tools themselves, see the tool comparison for CRM, design and tracking. The AI hub lists the rest of the series.

Questions small solar installers ask about AI

Can AI design the solar system?

AI can help with the paperwork around a design, but string sizing, roof loading, electrical protection and compliance sign-off stay with a qualified designer and installer. Design software does the layout and yield modelling; a person checks the output against the roof and the standards you work to.

Which of the five jobs should I automate first?

Start with lead qualification if you lose enquiries to slow replies, or customer updates if you are drowning in how-is-it-going calls. Both are low risk and easy to check, and within a few weeks you can see whether they are working. Leave permit drafts until your rules and templates are clean.

How do I check AI figures in a proposal?

Never let the assistant calculate yield, savings or payback from scratch. Take those numbers from your design tool, paste them in, and ask the assistant only to explain them. Then compare every number in the final text against the tool's output before it goes to the customer.

Do I need special software for any of this?

No. A general chat assistant, a saved set of instructions and your existing CRM or inbox cover most of it. Some CRMs and design tools have AI built in, which can be handy, but check what each feature does and whether it costs extra before you rely on it.

Is it safe to paste customer details into AI?

Leave out what is not needed. For drafting wording you rarely need a name, address or phone number. Use placeholders, check your plan's privacy settings, and use a business plan if you handle many customer records. Electricity bills and roof photos often carry names, addresses and account numbers, so crop or blank those before you upload them.

Further reads

• AI Solar Design Software Compared: Cost per Proposal for Small Installers — What design software costs per proposal for a small installer.
• Can AI Design a Solar Array? What Design Software Does and Doesn't Do — What design software does and does not do for arrays.
• Solar Installation Quality Checklist: Sort Photo Evidence With AI — Sort installation photo evidence for handover packs.
• Solar Installer Marketing: Local Pages and Case Studies Written With AI — Fill the top of the funnel with local pages and case studies.
• AI Automation for Roofers: Six Workflows Worth Setting Up First — Roofing workflows that overlap with solar installs.
• Solar Panel Cleaning Business: Quote and Route Jobs With AI — Quote and route an add-on cleaning service.
• Solar Battery Quote Template: Options Explained With AI, No Backup Promises — Battery option wording that never overpromises backup power.
• Solar Lead Generation: Cost per Lead, Cost per Survey and AI Follow-Up — What solar leads cost and why a fast reply protects them.
• Can AI Write a Solar Sales Script? Discovery Questions and Objections — How to have AI draft solar discovery questions and objection handlers that stay honest, with role-play prompts and a claims checklist.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenSolar's free-software and API add-on pages, OpenAI and Anthropic help pages for Projects and plan limits, and general installer workflow practice. Hours saved are illustrative, not measured.