The building isn't finished. The visual has to be.
A global manufacturer of aluminium composite panels needed to show architects and developers what a half-built building would look like once it was clad. Getting there took three departments and up to a month. We built a visualiser that does it in one session and produces the bill of quantities with it.
- From site visit to a visual in the customer's hands
.svg?2026-08-31T07%3A37%3A09.904Z)
From site visit to a visual in the customer's hands
- Departments involved in one visualisation
Departments involved in
one visualisation
- To process a site photo into a cladvisualisation
.svg?2026-08-31T08%3A14%3A59.494Z)
To process a site photo into a clad
visualisation
- Panel counts, accessories and wastage, generated with the visual
Panel counts, accessories and
wastage, generated with the visual
A global panel range, sold through people standing on building sites.
A global manufacturer of aluminium composite panels the cladding that gives a finished building its face. Their range runs to a wide catalogue of finishes, colours and profiles, specified by architects and developers on projects that are, by definition, not yet built.
They sell through their own team and through a network of installers and dealers. Which means the person in front of the customer is frequently not in the same building as the designer who can show what the product will look like.
The product is chosen on appearance. And nobody could see it until the building was already finished.
Three departments, and a decision being made
somewhere else.
The company sells standard signage across 22 categories and offers customisation in every one of Cladding is specified visually. An architect or developer chooses a panel finish because of what the finished building will look like which means the manufacturer who can show that first has an enormous advantage, and the one who shows it last is quoting against a decision already made. Producing a single visualisation ran through three separate departments.
Someone went to site.
A team member travelled to the building and photographed it. Since these are buildings under construction, the photographs showed exactly that scaffolding, cement sacks on the stairs, site debris, missing windows and doors, machinery parked against the facade.
A designer rebuilt the building by hand.
Back at the office, a designer spent anywhere from a couple of days to a week manually cleaning the photograph, painting out obstructions, adding the windows, doors and fixtures that weren't installed yet, then applying the panel finishes. Time depended entirely on the designer's queue.
Then it went back to sales.
The finished visual returned to the sales team to present, and the bill of quantities was worked out separately.End to end, from site visit to a visual in the customer's hands: a week to a month.
The bottleneck was never one person. It was three departments handing work between them, each with its own queue and a specification decision being made somewhere else while they worked.
We didn't wait for the building. We finished it in software.
The obvious way to show a clad building is to wait until it exists, or to model it from architectural drawings. Neither works here. The decision is made mid-construction, and the drawings a manufacturer receives are rarely complete enough to render from.
So we started from what actually exists on day one: a photograph taken on somebody's phone. The software reads the real building out of that photograph removes what is temporary, restores what is missing, and treats the result as a surface it can clad. No survey equipment, no drone, no measured drawings, and no CAD department in the middle.
The consequence is that the work stops being a relay. One person photographs, clears, completes, clads and costs in a single session, without raising a request with anyone.
The installer already visits the site. We made the visit the only step.
One photograph. Everything downstream.
Around fifteen minutes of processing per photograph, and the work sits with one person from start to finish.
- Photo
- Clear
- Complete
- Clad
- Verify dimensions
- BOQ
Upload the site photograph.
Taken on an ordinary phone or camera.
Automatic clearing.
The software removes what it can identify on its own: people, trees, scaffolding.
Guided clearing.
For everything else, the operator circles it cement sacks on the stairs, debris against the wall, a parked vehicle. Generative fill reconstructs what should be behind it.
Complete the building.
Windows, doors, light fixtures and landscaping that aren't installed yet get added, so the customer sees a finished building rather than a construction site with cladding on it.
Apply the panels.
Finishes from the manufacturer's actual range, on the actual building.
Generate the bill of quantities.
The software estimates facade dimensions from the photograph. The operator verifies them, and the BOQ follows: panel counts, accessories and wastage allowance.
A bill of quantities, measured from a photograph.
Most visualisation tools stop at the picture. A picture is a marketing asset; a picture that produces a costed material list is a sales tool.
The software estimates the facade dimensions from the image itself, and the operator verifies them against what they know of the building. From those verified dimensions come panel counts, the accessories required, and a wastage allowance.
Estimating real-world measurements from a single uncalibrated photograph is not a feature you can buy. It is a computational problem, and it is the reason this tool produces a quote rather than a rendering.
Supervised, not automated.
Automatic removal is limited to what the software identifies reliably: people, trees, scaffolding. Everything else is operator-directed the person circles what needs to go, reviews what the fill produces, and can undo the step and try again.
Approved by a person, every time.
This is generative AI doing work a designer used to do by hand, and a hallucinated window on a real building is a commercial problem, not a technical curiosity so nothing reaches a customer without a human having looked at it and approved it.
We use AI where it removes hours of manual work, and we keep a person in the loop wherever the output has your name on it.
A global panel range, sold through people standing on building sites.
The tool sits with the manufacturer's own team and with their installer and dealer network and that is the part with the largest commercial consequence. A dealer standing in front of a half-built building can photograph it, clad it in the manufacturer's panels, and put a visual and a material list in front of the developer. Without raising a request with head office, joining a designer's queue, or waiting on three departments. The manufacturer keeps control of what is shown, because the panel range, the finishes and the BOQ logic are all defined centrally. The channel gets to sell with it.
The head office design team stopped being the bottleneck, because it stopped being in the path.
"Nobody makes software for what we do" is sometimes true.
Standard configuration software assumes you are assembling a product from options. This starts from a photograph of a real, unfinished building and has to reason about it what is construction debris and what is architecture, what is missing, and how large the facade actually is. There was no product to buy for this, and no template to adapt. It was built from the problem, which is the only way work like this gets done. If your product is chosen on how it will look somewhere it hasn't been installed yet, and someone in your business spends hours helping customers imagine it, that work can usually be built into software.
Is your product specified visually?
If your customer has to imagine what your product will look like before they'll buy it, and someone in your business spends hours helping them imagine it, this is the same problem.
Thirty minutes. Show us how you do it today.