Selling what does not exist yet
Kitchens, windows and facades are sold before they exist. AI now makes the picture during the meeting. What that does to decisions, returns and trust.
A kitchen is sold before it exists. So is a new window, a facade, a staircase or a fitted wardrobe. The customer is asked to commit a large sum to something nobody can show them, on the strength of a sample, a drawing and a conversation.
The trades know where that sale is lost. A proper visualisation takes time, so it usually arrives after the meeting, for a customer who has already started comparing. The doubt grows in the gap between the conversation and the picture.
AI can now close most of that gap. The question for a company is no longer whether it can show the finished result. It is how honestly it does it.
Why the picture sells
People read pictures fast. In 2014 a team led by Mary Potter at MIT showed that the brain can process an image seen for as little as 13 milliseconds. A description takes minutes and still leaves every listener with a different kitchen.
Augmented reality offers some of the best evidence. A study in the Journal of Marketing found that letting shoppers try products in augmented reality had a positive effect on sales. The effect was small overall and stronger in some places: for less popular brands, for more expensive products, and for customers new to the category or to buying online. The authors concluded that it works best when uncertainty about the product is high.
Made to order goods sit at the far end of that scale. They are expensive, they are made once, and nobody can try them. Real estate shows the same pattern with a simpler tool. In the National Association of Realtors’ 2025 report on staging, 83 percent of buyers’ agents said that staging made it easier for a buyer to see the property as their future home.
What AI changed
Three tools have become ordinary. The first is the product picture without the shoot: a real product photographed once, then placed in any scene, light and format, in a style the brand has fixed in advance.
The second is the customer’s own space. A photo of a room, a garden or a facade becomes a photorealistic picture of the finished project. The technique estate agents use for virtual staging now shows a kitchen that has not been built, in a room that has not yet been cleared.
The third is the customer’s own body. Since July 2025 Google has let shoppers in the US upload a full length photo in Search and Google Shopping and see clothes from billions of listings on themselves.
Configurators were the older answer, and they keep one advantage: they only show what the company can build. A generated picture can show anything, which is exactly why it needs a real catalogue behind it.
Returns and the honest picture
It is tempting to promise that better pictures mean fewer returns. The evidence is thinner than the promise. A review of the research on virtual try on, published in June 2026 by Cristina Yun and Jung Hwa Seo, finds that the evidence on returns is promising but less mature than the evidence on purchase intent. Try on reduces returns mainly when it improves the accuracy of fit and narrows the gap between what the customer expected and what arrived. A picture that only entertains does not.
The authors draw a conclusion every seller should take seriously: aim for sales that stay sold, not conversion alone. For made to order trades the point is sharper still. A render that shows a finish the company does not offer, or a worktop that will not fit the wall, does not produce a return. It produces a dispute.
So the honest picture is also the profitable one. It shows the real product, in the real dimensions, from the company’s own catalogue.
Labels and provenance
Regulators and platforms are moving in the same direction. Since 2 August 2026 the transparency rules in Article 50 of the EU AI Act apply. Providers of generative systems have to mark their output so it can be detected, with time until 2 December 2026 for systems already on the market, and deployers have to label deepfakes. In June the Commission published the final code of practice that shows how, together with a set of EU icons for labelling AI content. Following the code is voluntary, and by the end of July about 190 organisations had signed it.
The platforms are adding their own rules. Google Merchant Center requires product images made with generative AI to carry an IPTC metadata tag that says so, and asks sellers not to strip it. At its developer conference in May, Google began rolling out checks for C2PA Content Credentials in the Gemini app, with Search and Chrome to follow. They show whether an image is an unaltered original from a camera or has been changed, and by which tools.
The direction is clear. A generated picture is allowed. A disguised one is not. Whether a render of a customer’s own kitchen counts as a deepfake is a legal question. Our practical answer is simpler: label it, and keep the original photo one click away.
How the trades use it
In the trades that sell by consultation, the change is mostly about timing. The picture used to arrive after the meeting. Now it can be made in the meeting, from a photo the customer took on a phone, and changed while they watch: another front, another stone, another frame colour.
That changes the conversation. The customer stops trying to imagine and starts choosing. The consultant stops describing and starts showing. And the quote that follows refers to a picture both sides have already agreed on.
The picture does not replace the consultant. Knowing what fits the wall, what lasts and what the budget will carry is judgement built over years, and the customer buys that as much as the worktop.
Where Beform, Reel and Proof fit
Beform is a product and brand of crue GmbH, with its own brand and its own site at beform.ai. It turns a photo of a client’s kitchen, bathroom, living room, facade or garden into a photorealistic picture of the finished project in about twenty seconds. Walls, windows, doors and the camera stay where they are, and only what a renovation would change, changes. On a company’s website, the Showroom lets a visitor see their own room and leave a request with the picture attached. In the meeting, the consultant changes materials live from the company’s own catalogue, so every picture is something the company can build and price. Beform runs in sixteen languages for fifteen trades, and every picture carries a visible label and IPTC metadata. The before and after pictures on its site are AI illustrations, not client projects.
Reel does similar work for brands that sell from a catalogue: product photos, mockups and short films in styles locked from real references, made on the best model for each job, with a person picking every image and every clip that is used.
And because a picture is an argument, it should be tested like one. Proof runs two versions of a page, an ad or an email against each other and keeps the one that sells, so the choice between a photograph and a render is made on the numbers.
Selling what does not exist yet is the oldest problem in the made to order trades. The picture now arrives in time. The work is to make sure it tells the truth.
