The busiest year for South African solar installers was 2023. Load-shedding hit Stage 6 for weeks on end, and a rooftop PV system with a battery bank was the only reliable way for a suburban household in Fourways or Table View to keep a fridge, a router, and a kettle running through a nine-hour outage. Installers were quoting on the phone, sending crews out with tape measures, and turning away work that would not fit into the diary. The phone was the sales team.
By mid-2025 Eskom's fleet had recovered enough that whole months passed without a scheduled outage. The panic buyer disappeared. What arrived in their place is a slower, more comparative customer: someone who has finally got around to asking three installers for quotes, who has read the PV GreenCard consumer brochure, and who is being courted by GoSolr, Wetility and Alumo Energy on subscription terms with no capital outlay. That customer expects a reply within an hour. Two, at most.
If your installation business in Boksburg or Somerset West grew in 2023 by picking up the phone every time it rang, this is the year that stops working. The intake flow that fed a two-week backlog now has to work harder for every job. That is the real problem where AI can pull weight for a small solar installer: the lead-response gap, the quote turnaround, and the ten-year admin tail that starts the day a system commissions.
Where the hours actually go in a small SA solar business
Walk a typical five-person installer through a week and the pattern is consistent. Two days on active installs. One day the owner is on site meeting new prospects. Half a day is spent chasing SSEG paperwork, either with the City of Cape Town, City Power, Ekurhuleni or the local municipality. Half a day is warranty and service calls: inverter fault codes on the WhatsApp group, a Sunsynk that dropped grid at 03:00, a battery that reports 68% SoH when the client thinks it should still read 95%. The last day gets sliced between quote drafting, panel and inverter procurement from Rubicon or Segen, and the annual electrical declarations for existing sites.
The billable, technical work is the site design and the physical install. Almost everything else is administration that used to be absorbed by the sheer strength of the lead flow. It is not any more.
The intake gap: from bill photo to answer in under an hour
A qualified inbound solar lead in 2026 almost always arrives as a WhatsApp with a photo of an Eskom or municipal bill, sometimes a photo of the roof from Google Earth, and a one-line message saying "what would this cost, roughly."
Handled well, that message deserves an answer within an hour that includes: a plausible system size in kWp, an inverter and battery pairing at a sensible price point, an indicative rand range with clearly labelled assumptions, and a specific site-survey slot. Handled badly, it sits in the owner's phone until Saturday evening because Tuesday was on a roof in Randburg and Wednesday was three warranty callouts in Milnerton.
An AI-assisted intake flow can carry the whole first-pass response. It reads the bill photo, extracts the monthly kWh and the tariff, cross-references against the property location for a rough insolation figure, and produces a draft reply the installer approves in fifteen seconds before it goes out. The reply is honest about what it is: an indicative sizing based on the bill, subject to a physical roof survey and a full electrical assessment before anything is quoted formally. It offers two survey slots pulled from the calendar. It does not commit to a price.
This is the part most installers get wrong when they try it themselves. They either automate a generic "thanks for your enquiry, we will be in touch," which loses the customer to the next installer, or they attempt a fully autonomous quote from a photo, which produces numbers the business cannot honour. The useful middle is a well-framed draft the human approves. In my experience, that middle wins the deal a startling proportion of the time.
SSEG paperwork: the tail nobody quotes for
Every grid-tied installation in Cape Town, Johannesburg, Ekurhuleni and most major municipalities now requires SSEG registration before commissioning, and NRS 097-2-3 compliance from an approved inverter list. The paperwork itself is not hard. It is voluminous, slow, and full of small things that get bounced back: a missing single-line diagram signature, an SLD annotation the municipality wants in a specific format, a COC number that was not filled in on the application form.
A small installer with fifteen active installs will typically have four or five SSEG applications in various stages of stall at any given time. Chasing them costs half a working day per week. AI helps here in a quiet, unglamorous way. It tracks each application against the specific municipality's known turnaround, prompts the client for missing documents in the client's language of choice, and flags when an application has crossed a threshold that warrants an escalation email to the SSEG office. It does not fill in the compliance forms and does not sign anything the electrician of record must sign.
Site survey coordination
Two crew members, three surveys booked in a day, a client who has not confirmed access to the DB box, and a Ford Ranger stuck in Voortrekker Road traffic. Every solar installer knows this Tuesday.
An automated survey-coordination loop is not clever. It confirms the appointment the day before, asks the client to send a photo of the DB box and the meter box in advance, checks whether there is a body corporate involved and if so whether trustee approval will be required, and reschedules automatically when the surveyor's morning runs long. Small change. Real hours back.
After-sales: where the ten-year relationship lives or dies
A photovoltaic system is a fifteen-year asset. The client bought it once. The installer either keeps them as a service customer for those fifteen years or hands them to a competitor at the first battery replacement.
The service tail is where AI does its most durable work for an installer. Three concrete patterns.
Inverter-monitoring triage. Sunsynk, Deye, SolarEdge and Growatt all publish live telemetry to their apps. A fault code at 03:00 wakes the client, who WhatsApps the installer, who often does not see the message until sunrise. An AI layer reading the telemetry and the fault-code catalogue can send an accurate, plain-English message to the client within a minute ("your inverter reports an F16 grid-loss event, this is the municipal supply, no action needed on your side, your system will resync automatically at daylight") and only pages the on-call technician for codes that actually require human attention.
Annual PV service reminders. SAPVIA and the PV GreenCard scheme recommend annual servicing. Most installers do not chase it because it is thankless admin. An AI-driven annual reminder, timed to the anniversary of commissioning, offering two service-visit slots and a specific per-site checklist, converts at rates that make the whole subscription-solar competitive threat look manageable.
Battery state-of-health nudges. A lithium battery that has drifted from 100% to 82% SoH after four years is normal. A client who watches it drift and gets no explanation reads it as failure. A quarterly automated update, plain language, honest, drawn from telemetry, turns a support call into a client who trusts you to install their neighbour's system.
What AI must not touch on a solar site
Worth naming plainly, because the market conversation tends to skate over it.
- The final system design and the panel-inverter-battery pairing. Sizing is professional judgement anchored in the site survey, the load profile, the roof orientation, the shading study and the client's actual usage pattern. A large-language model is not a solar engineer.
- The Certificate of Compliance and the electrician-of-record signature. Both are legally the responsibility of a registered person of record and must not be automated, delegated, or drafted by a bot.
- Live fault diagnosis on a hot roof. If a technician is standing next to a live DC string with a multimeter, the last thing they need is an AI hallucination on their phone. Diagnostic AI is a support tool for the office. It is not a decision tool for the technician on the ladder.
- Warranty adjudication. Inverter and battery warranty calls turn on specifics of manufacturer terms, installer records and site conditions. That is a human conversation.
- The condolence call. A client whose battery bank was stolen in a burglary does not want a chatbot's empathy. Pick up the phone.
POPIA, the PV GreenCard, and the paperwork trail
Solar installers hold sensitive customer data: municipal bills with account numbers, ID copies for SSEG applications, banking details for finance forms, geolocation on the property. POPIA applies in full. Every automated flow that touches this data must be built with retention, access-logging and deletion in mind from day one. Ask the vendor where the data sits, who at their end can see it, and what happens on a deletion request. Vague answers are the answer.
On the accreditation side, the PV GreenCard installer register carries a reputational weight in the SA market that the marketing versions of "certified" do not. Automation should support the paperwork trail the register asks for, not shortcut around it.
Where to start
The single highest-return first project for most SA solar installers is the WhatsApp bill-photo intake with an AI-drafted first reply, approved by the owner before it goes out, plus a site-survey booking flow. It costs almost nothing relative to what a single lost deal is worth. The impact is visible in the first week: the "did anyone reply to me" message that used to come three days later stops coming.
The natural extensions, in order, are SSEG paperwork tracking for the current in-flight applications, annual-service reminder automation for the existing installed base, and only then inverter-telemetry triage once the office can absorb the alerts the telemetry generates.
Installers who have run this stack for a full year in 2026 report the same two things. The Saturday evening WhatsApp storm gets absorbed. And the client who would have compared five quotes to a subscription operator picks the installer who replied first, honestly, with a sensible number and a specific survey slot.