Solutions / Deep custom
A ranked shortlist of the roofs to quote first.
An agent loop for solar companies: it reads satellite imagery for roof size and shading, estimates energy use, checks public company records, and hands sales a shortlist of properties that fit.
What outside research says
190% is how far South Africa’s national average electricity tariff has risen since 2014, mostly above inflation.
Source: CSIR, 2025 (opens in a new tab)4,412 MW of rooftop solar was installed in South Africa by June 2023, up from 983 MW in March 2022.
Source: The Presidency, 2024 (opens in a new tab)
Findings from the research linked under each figure. They are not Yielde results.
The loop
How it works
From an area on the map to a ranked shortlist
Six steps, from the first event to the result. Choose a step to read what happens in plain words, how it is built, and what keeps it right.
Step 1 of 6
You choose where to look
In plain words
You tell it where to look and what a good site means to you: the suburbs or industrial parks, the kinds of buildings, and the smallest roof worth your time.
Safeguard It works only inside the areas and rules you set. Each rule change is logged, so every shortlist can be traced to the rules it was made with.
Under the hood
Your rules are agreed with you and stored as settings, not loose instructions: areas as boundaries on a map, building types, the smallest usable roof, and the tariff categories you want and the ones you avoid. The run is designed to start on the schedule you set, usually weekly. It keeps a record of every site it has already assessed, so each run adds new sites and re-checks old ones only when newer imagery is available.
It reads the roofs
In plain words
It looks at satellite and aerial photos of each building and estimates how much roof could carry panels, which way it faces, and what casts shade on it.
Safeguard Roof area, direction and shade are estimates until someone visits the site, and every brief labels them that way.
Under the hood
An image model outlines each roof and its faces from the licensed imagery, then removes what panels cannot use: edges, skylights, water tanks, air-conditioning units. Direction comes from the outline, and shade is estimated from trees and taller buildings nearby. Pitch and roof condition cannot be read reliably from above, so they stay unknown. Every estimate is stored with the date and resolution of its image. Where an image is old, cloudy or too coarse, the site is marked not assessed instead of guessed.
Works with
- Satellite and aerial imagery
Image too poor: goes straight to step 05, Your team approves.
It scores each site
In plain words
Each site gets a score: how big and clear the roof is, how much power the building probably uses, the tariff it is likely on, and which business is in it.
Safeguard A site on your do-not-contact list is removed before it is scored and stays out of every later run.
Under the hood
First each site is checked against your CRM: existing customers, open deals and anyone who asked not to be contacted are taken out. The rest are scored by plain code with weights agreed with you: usable roof, shade, likely use for the building type and size, and the tariff from the tariff tables. Public company records link the building to the business in it, matched on address and name, and a weak match is marked unconfirmed. Each score shows its parts, so you can see why a site ranks where it does.
Works with
- Your CRM
- Tariff tables
- Public company records
It writes a brief per site
In plain words
For each site near the top it writes a one-page brief for your sales team, with a first message they can edit, so they start with facts instead of a cold call.
Safeguard The brief may not state a saving, a payback or a production figure. Those wait for a site visit and a full design.
Under the hood
The language model writes only from the stored facts for that site: each statement points to the image, tariff row or company record it came from. The first sizing is a range worked out by code from the usable roof and the estimated use, not by the model. The likely decision-maker comes from the public company record. The draft approach follows a brief written with you and makes no claim about savings or production, because those need a site visit and a full design.
Your team approves
In plain words
Your salesperson reads each brief, checks the sizing, edits the first message or drops the site. Nothing goes near a prospect until they approve it.
Safeguard No draft is released and no site becomes a lead without a named person’s approval. The agent cannot approve its own work.
Under the hood
The shortlist arrives ranked, with the sites that could not be assessed from imagery in a separate list for your team to judge from a drive-past or skip. Each site can be approved, edited or rejected, and a rejection asks for a short reason. Those reasons are reviewed with you to tune the rules for the next run. A draft is released only with a named person’s approval attached.
Leads land in your CRM
In plain words
Approved sites appear in your CRM as leads, each with its brief and the approved first message, ready for your salesperson to send from their own email.
Safeguard The system never contacts a prospect. It writes the lead and the draft after one last check against the do-not-contact list, and your salesperson sends.
Under the hood
On approval, the site is written to your CRM as a lead with its brief, score, sources and approved draft, and assigned to the salesperson you chose. The write is matched against existing records first, so a site already in your CRM is updated, not duplicated. The do-not-contact list is checked once more at release, because someone may have opted out since the draft was written. Sending stays with your salesperson.
Works with
- Your CRM
What it knows
Your rules for a good site, licensed satellite and aerial imagery of the areas you chose, tariff tables, public company records and your CRM. Every fact it uses is stored with its source and date, so any brief can be checked.
What it may do on its own
Read imagery, estimate roof area, direction and shade, score sites against your rules, match buildings to company records, and draft briefs and first messages. It reads your CRM, and writes to it only after approval.
What waits for your yes
Every site before it becomes a lead, every first message to a prospect, and every site it could not assess from imagery. Sending any message stays with your team.
How we test it
Before the first run it scores sites your team already knows, including ones you visited and turned down, and its roof estimates are compared with what your team measured on site. The first shortlists are checked with you before any draft is released.
What we measure
Once it is live, these are tracked and reviewed with you.
- Shortlisted sites your team approved
- Roof estimates corrected after a site visit
- Site visits booked from the shortlist
What you get
- Roof and shading estimates from imagery
- A fit score built on your own criteria
- A weekly shortlist with a first system sizing
- Draft outreach per site
Works with
- Satellite and aerial imagery
- Tariff tables
- Public company records
- Your CRM
Who it is for
Solar installers and EPCs who want to choose their clients instead of waiting for them.
Built to order. We scope it with you on a call, price it before any work starts, and run it in draft mode until you trust it.
Related
- Quoting agent
Quotes drafted the same day, priced the way you would price them.
An agent that turns an enquiry, a site photo or a drawing into a priced quote using your own rates, materials and margins, ready for you to check and send.
Built to order - 3D energy modelling
Proposals that show the projected saving on the client’s own roof.
A system that builds a 3D model of a site, places the panels, and projects generation against consumption hour by hour, so the proposal shows its working, with every assumption stated.
Built to order - Research and reporting agent
Decisions made on this week’s numbers, not last quarter’s.
An agent that joins your financials to market research and outside data, and writes a weekly brief the way an analyst would: what changed, why, and what to look at.
Built to order