In just a few clicks, an owner receives a “precise” estimate of their home. The algorithm crossed thousands of comparable transactions, analyzed the surface area, the condition of the property, the neighborhood and even the photos. The number seems solid. Until a developer comes along and offers 30%, sometimes 50% more. Or, conversely, significantly less. How is this possible? Because the price of a property intended for a developer cannot be calculated like that of a sale between individuals. And that AI, as powerful as it is today, does not yet have the means to fully understand it.
The promoter balance sheet, a backward calculation that the algorithms do not master well
An individual buys a house to live in. A developer buys a right to build. He thinks in reverse: he starts from the forecast turnover of future housing, subtracts construction costs, technical fees, taxes, financial costs and his margin, to determine what he can devote to land.
This “promoter balance sheet” depends on variables that traditional databases (DVF, announcements, past transactions) capture very imperfectly: the floor area actually authorized, the number of possible housing units, the type of program (free access, social, intermediate), the anticipated resale price of apartments or new houses, or even any architectural constraints imposed by the town hall. There are also all the standards for disabled people, accessibility, percentage of slope for access to the basement, minimum and maximum distance for the location of a 2-wheel premises, all obligations which are specific to each property and very complicated to anticipate by an AI.
Excellent AI tools on the legacy market struggle to model these multiple and hyper-local scenarios. They remain anchored in the logic of “comparable”while the promoter reasons in potential.
PLU, servitudes and informalities: the opaque territory of AI
A Local Urban Plan is often 300 to 400 pages long. It contains floor space coefficients, maximum heights, parking rules, architectural requirements, protection zones, easements, etc., not to mention current modifications or revision projects. Added to this are elements rarely formalized in open databases: the informal opinion of the town hall on a project, the attitude of neighbors, the possibility of sharing with a contiguous plot, or even the real state of the networks. Very specific data can also constrain construction and influence the developer’s study: presence of trees, which the town hall wants to preserve or which are of a species that must be preserved. These phytosanitary elements do not only concern century-old trees, sometimes very small trees cannot be uprooted or moved because of obligations. This information still circulates widely through dialogue, field work, human analyzes and experience. The AI can read a scanned PLU. It is much more difficult to extract all the operational nuances and to anticipate the political or technical trade-offs that take place behind the scenes.
Result: two plots of land of the same size, in the same neighborhood, can have very different resale values to a developer. A purely algorithmic estimation smooths out these differences.
Data + AI remain excellent… for the individual market
This is not to denigrate artificial intelligence. For a traditional sale between individuals, current tools already provide valuable assistance: realistic price ranges, rapid identification of comparable goods, detection of anomalies. They speed up diagnosis and reduce rough approximations. But as soon as we switch to the developer’s land logic, data alone is no longer enough. The price is no longer a snapshot of the past market. It is a projection, negotiated, conditional on obtaining a building permit, and influenced by dozens of human and regulatory parameters.. It is precisely for this reason that support remains central.
Analyze the real constructability, solicit several developers, compare the balance sheets, secure the suspensive conditions, dialogue with the town hall, sometimes the urban area, the road services and the notary… so many steps where professional expertise and human monitoring still make the difference.
Hybridization rather than replacement
The future probably does not lie in an AI that would completely replace the expert, but in tools that augment it. Models capable of ingesting town planning documents more precisely, of simulating several developer balance sheet scenarios, or of warning of recourse risks, will make the work faster and more precise. They will not eliminate the need to understand the local context, negotiate and support the seller over an average period of 12 to 18 months.
In the meantime, relying solely on an automated estimate for a sale to a developer means taking the risk of significantly undervaluing your property.…or commit to an unrealistic project. Data and AI have already transformed the real estate market.
In the very specific segment of land intended for development, they have not yet revolutionized everything. And maybe that’s just as well. This still leaves room for field intelligence… and human supportat the heart of the model that we defend at Sell to a Promoter.

