We want to sell an apartment, we type the address into an online simulator and get a price in less than a minute. This reflex has become almost systematic even before contacting an agency. These simulation tools exploit public databases to produce a quick estimate, but their accuracy depends on parameters that most users are unaware of.
DVF Data and Geographic Limitations of Real Estate Simulators
The most reliable simulators do not rely on the prices displayed in listings. They query the DVF database (Demand for Property Values), fed by the DGFiP from notarial deeds. Thus, we obtain the price actually signed at the notary’s office, not the price desired by the seller.
This distinction changes everything. An apartment listed at a certain amount can sell significantly below that after negotiation. The simulator that relies on completed sales starts from a more solid data point than an aggregator of listings.
The problem is that coverage is not uniform. The DVF data does not cover Bas-Rhin, Haut-Rhin, Moselle, or Mayotte. In these areas, a simulator that still displays a result without indicating this misleads users. When using Econozen’s online simulators, we save time on the initial framing, but it remains useful to check that the geographic area has enough comparables.
In a recent and homogeneous condominium, where the units are similar, the gap between the automatic estimate and the final price remains moderate. In a single-family home, two properties of identical size on the same street can show a considerable value gap because the database does not describe the interior condition of the property.
EPC and Energy Data: A Criterion That Models Now Integrate
Recent simulation tools no longer just cross-reference surface area, location, and price per square meter. The energy performance diagnosis is becoming a full-fledged estimation factor. A property classified as F or G suffers a depreciation that models are beginning to quantify based on past transactions associated with an EPC.
The difficulty arises from methodological discrepancies between diagnostics. An EPC done using the old method and another established after the decree published in the Official Journal on August 26, 2025, do not produce the same results for the same property. The simulator that cross-references this data without distinguishing the date of the diagnosis can therefore overestimate or underestimate a property.
In practice, when entering the characteristics of an apartment into an online tool, there is rarely an option to specify the date of the EPC or the method used. The model works with what it finds in public databases, and feedback on this point varies across platforms.
What the Simulator Does Not See in the Property
No algorithm detects a structural crack, an annoying overlook, or a recently renovated kitchen. The DVF data records the transaction, not the actual condition of the property. Here are the elements that systematically escape automated models:
- The interior condition of the property (recent renovation, obsolescence, quality of materials) does not appear in any database used by simulators
- The exposure, brightness, and view from the property are not coded in the transaction data
- Local nuisances (noise, nearby urban planning projects) are not included in the model’s variables
- The quality of the condominium (charges, voted works, condition of common areas) remains invisible to the tool
These factors can represent a significant gap in the final price. That is why a simulator provides a ballpark figure, not a selling price.
Online Simulator and Professional Estimate: Two Distinct Uses
Online estimation and the opinion of a real estate agent are often opposed as if one should replace the other. In practice, they meet different needs and intervene at distinct moments in the sales process.
The simulator serves to frame the situation. We check in a few minutes if the property is within a range consistent with the local market. It is a pre-estimation tool that prevents contacting a professional with completely unrealistic expectations.
The on-site estimate complements what the algorithm cannot capture. An agent visiting the property incorporates the condition of the property, the dynamics of the neighborhood, and local competition at the precise moment of the sale. These elements do not appear in any database exploitable by an automated model.
Why Different Simulators Give Different Results
Each tool uses its own calculation model. Some weigh surface area more heavily, others the geographic area or the date of the selected transactions. The historical depth of the data also varies: a simulator that incorporates sales from the last twelve months does not produce the same result as another that goes back three years.
- The geographic perimeter of comparables differs (same street, same neighborhood, same city)
- The weighting of criteria (surface area, floor, age) changes from one model to another
- The frequency of updates to DVF data is not the same across platforms
Testing two or three simulators allows you to identify the common range. A significant gap between results indicates an atypical property that models struggle to categorize.
Reliability of Online Estimates: What Can Be Reasonably Expected
A simulator works well when the property resembles other properties sold recently in the same area. The more the property deviates from the local standard (atypical surface area, rare configuration, sloping land), the less relevant the estimate becomes.
For an apartment in a recent building with several comparable sales nearby, the automatic estimate provides a reliable working basis. For a house with specific characteristics, it serves as a starting point, not a conclusion.
The automated analysis is progressing with the integration of new data (EPC, proximity to transport, enriched cadastral data), but the model remains dependent on what public databases contain. The perfectly documented property in the data does not yet exist.
