What INSEE actually says about AI in French real estate
The share of real estate businesses using an AI technology went from 7 to 14 percent in a year. The figure is striking, and it says something rather different from what it is made to say.
The share of real estate businesses using an AI technology went from 7 to 14 percent in a year. The figure is striking, and it says something rather different from what it is made to say.
INSEE's 2024 ICT survey produced a figure the sector quotes constantly. Among French companies with at least ten employees, the share of real estate businesses declaring use of at least one artificial intelligence technology rose from 7 percent in 2023 to 14 percent in 2024[1]. A doubling in a year.
A doubling sounds good. It has to be read both ways. It also means 86 percent of real estate businesses used no AI technology at all in 2024. Across a sector of roughly 30,000 agencies in France[2], the gap between the ambient discourse and what sits on actual desks remains considerable.
A second qualification, and it matters. The survey covers companies with at least ten employees. A large share of French estate agencies sits below that threshold. So the figure describes the upper part of the market, the networks and the structured agencies, rather than the independent seven-person office.
A third, on size. Across all sectors, INSEE measures 9 percent usage among companies under 50 employees, against 33 percent above 250[1]. Adoption is not spread, it is concentrated. What holds across the economy very likely holds in real estate too.
The survey wording is broad, as public statistics should be. The most widespread technologies are written language analysis, cited by 44 percent of users, and machine learning for data analysis, cited by 41 percent[1].
Translated into an agency's daily life, the first category mostly covers one behaviour: someone opens a consumer tool and has it produce listing copy. It is a real use, it counts in the statistic, and it looks nothing like an installed automation. It does not fire on its own, it connects to no agency software, and it depends entirely on one person's discipline.
This is where the figure misleads anyone trying to place themselves. Two agencies can answer yes to the same survey question and live in unrelated realities.
The first has a negotiator copying property details into a tool, retrieving a paragraph, pasting it back into the software. They saved ten minutes of writing and lost five in the back and forth. The net gain is thin and it disappears the week that person is on holiday.
The second has connected listing production to its transaction software. The listing exists when the negotiator opens the file, calibrated on how the agency writes, and all that is left is to approve it. The gain is structural, it depends on nobody's discipline, and it survives the holidays.
Both count as one point in the INSEE statistic. Only one changes the agency's week.
The doubling INSEE measured is real and it will continue. What we distrust is the reading applied to it. A sector moving from 7 to 14 percent is not a laggard that must urgently catch up, it is a sector at the very start of its curve, where the question is not whether to adopt but what to adopt. One more subscription is not adoption. A task that has left human hands is.