Interior design with AI: where it steps in, where it stops and what people decide
How we have used artificial intelligence in the concept phase since 2024, and what that means for the budget, schedule and decisions in your project.
Almost every design studio works with AI today. What sets one studio apart from another is when it uses it and what question it asks.
For you, as the client, this comes down to two very concrete numbers: how many directions you saw before signing the cost estimate, and how many of the project's risks were found on paper rather than on site.
Key takeaways
- We have been working since 2018. AI entered our process in 2024 and sits in one place only: the concept phase. Projects delivered before then are entirely the result of decisions made by people.
- The obvious layer of AI, renders and layout variants produced in a few seconds, changes the speed of delivery without touching the quality of the decisions in your project.
- The value appears when the tool is used to widen the range of directions, challenge the project's assumptions and check whether the initial request is framed correctly.
- The most expensive moment in a project is when you get a coherent, convincing answer to the wrong question.
- The decision stays with people. Professional liability, building regulations and on-site measurements cannot be delegated to a model.
What AI does in most studios
"Generate a facade." "Make this render better." "Write my concept text." "Give me a furniture layout."
Useful, fast and completely irrelevant to your project, because it touches none of the decisions that will determine your budget.
A furniture layout generated in thirty seconds looks good in a presentation and says nothing about how two waiters move along the same aisle at 1:00 p.m. This is the visible part of the tool and, at the same time, the least important part.
Where the options come from
A single proposal, nicely presented, hides the fact that someone else has already chosen for you.
That is why, since 2024, the first thing we ask a model to do is mechanical and, honestly, boring. We write down the real constraints of the space: the available light, staff flow, opening hours, the maintenance budget, the customer who walks in at 11 in the morning and the one who walks in at 7 in the evening. Then we ask for a large number of strategies that respect all of them at once, each with its trade-off stated.
What comes out is a long list that is largely unusable. Out of twenty directions, three hold up to a careful reading. Our work only starts with the filtering.
The constraint that decides everything else is rarely the one written in the request. At Bloom, in Romexpo, it was foot traffic passing the storefront at a constant speed. Any direction that does not respond to that fails, however good it looks in the image.
For you, the gain is simple: you get a set from which the weak directions have already been removed, and the remaining ones come with their trade-off written on them.
Who is missing from the project
"Make the design better" is a request with no verifiable answer. "Which assumptions in this project fail for a 78-year-old patient, for a child, for someone in a wheelchair, for a first-time visitor and for the woman who cleans at 6 in the morning?" is a list of concrete problems.
The last category is the one almost everyone ignores. A material that looks flawless in a render and stains on its third contact with professional detergent becomes, within two years, the most expensive element in the project.
Cross-examining our own proposal
Before we present a direction to you, we ask for the strongest argument against it. What is superfluous. What contradicts what is at stake in your project. What we are assuming without evidence.
It is a cheap form of cross-examination. It costs a few minutes before the presentation, instead of a project revision afterward.
Criteria are set before and measured after
A design project can be debated in terms of taste forever. The discussion gets shorter if, before the first sketch, we agree together on what exactly needs to change in your numbers.
Here the tool does something very concrete: it puts each direction up against the agreed criterion and shows under what conditions it wins. A direction that is good for occupancy can be bad for maintenance, and that shows up on paper months before it shows up on the invoice.
The criteria we set at the start are the same ones we track after opening. A few, from projects delivered between 2020 and 2023, so before AI entered our process:
|
Project (year) |
Criterion set at the start |
Measured result |
|---|---|---|
|
Captured foot traffic and length of stay |
+18% foot traffic in the first 3 months; +22% average time spent |
|
|
Public perception and weekend bookings |
+1.2 points in review ratings |
|
|
Conversion per square meter |
+15% sales per square meter; +22% average time spent in the showroom |
|
|
Occupancy and revenue per available room |
+17% occupancy after the redesign of the common areas; +24% RevPAR |
|
|
Team adoption of the space |
+25% employee satisfaction; +20% use of collaboration areas |
None of the projects in the table went through AI. The numbers are the result of design decisions made by people, using the method we have used since 2018. What we have gained since 2024 is the speed with which we check, before we draw, whether a direction has a real chance of meeting the agreed criterion.
The question behind the request
A client comes in with "I want more seats at the tables." Behind that there is usually a drop in sales per square meter.
Sometimes the answer really is more seats. Other times it is table turnover at lunch, the path from the checkout to the exit, or the fact that the terrace empties at 5:00 p.m. because the sun shines straight in and nobody put up any shade.
Before we draw, we ask three questions: what problem the client thinks they have, what problem they actually have, and in how many ways the request can be reframed. A model delivers five reframings in two minutes. We choose between them together with you, looking at your numbers.
Where the tool stops
AI generates, compares, critiques and reframes. The decision stays with people, for three practical reasons:
- Professional liability. A technical design is signed by a licensed architect, who is accountable for it. A model is accountable for nothing.
- Building regulations. Fire safety, accessibility and sanitary requirements for medical spaces are checked against the legal text in force, not against the memory of a model trained on data from two years ago.
- Measurements. A beam missing from the survey, a floor level that does not match, an existing utility route: these become visible on site, tape measure in hand.
The risk we manage
A model's obvious mistakes are easy to catch. Flawless reasoning built on a false premise gets past almost everyone, including a professional in a hurry.
That is why the first step in any project is still checking the premise, with you, on real data: sales figures, flows, opening hours, the complaints your customers keep repeating.
Our perspective
A design trend is not an argument for a project decision. Neither is a new tool.
What has concretely changed is that generating directions costs almost nothing, and that shifts the value of a design studio from production to judgment: what you ask, what you eliminate, what you take responsibility for. The practical effect shows in budget decisions made earlier.
Since 2018 we have delivered more than 85 projects and designed more than 50,000 square meters in HORECA, offices, residential and medical clinics. Most of the portfolio was done without AI. The tool came into our studio in 2024 and works in one place only: the concept phase, where we widen the range of directions and challenge our own assumptions before they reach you.
What goes into the presentation has already passed through a cross-examination filter. What you sign is still a project made by people who are accountable for it.
Do you have a space in the works and want to see what a direction tested before it is drawn looks like? Let's talk.
