Importing a listing does not verify its claims
Let’s explore this page together
Why partner feeds should preserve the source of each claim, surface conflicts and stop when a material property condition cannot be reconciled.
This article reflects the named expert’s practical perspective. See NovAsia’s editorial policy for how material is prepared and reviewed.
Automated importing can make a property record look finished very quickly. A partner sends a feed or file, the title lands in the right field, the price appears, the area is mapped and the photographs are attached. From a technical perspective, the listing is ready to render. None of that proves that the claims themselves have been verified.
Once partner data appears on our page, the wording changes. A partner's statement that “parking is included” can easily become the platform's own unqualified “parking included.” The data has not merely moved between systems. Its apparent owner has changed.
The source should travel with the value
I do not like treating a property field as a bare string. A useful field also has provenance. The price may come from a dated price list. The floor area may come from a brochure. Completion status may be supplied by the developer. Availability may come from a feed refreshed once a day.
Those sources can all be legitimate inputs. Their origin still affects how the information should be presented and updated.
If an import stores only the latest value, a future editor may be unable to tell whether an earlier number was wrong, simply older or attached to a different unit. The problem becomes worse when two sources disagree. Choosing whichever value arrived last is an operational rule, not an evidence rule.
A later feed can be more accurate than an earlier one. It can also be less specific, accidentally mapped to the wrong phase or based on a different definition. The conflict itself is information. A mature import process should preserve it long enough for somebody to understand it.
Material conflicts need a different response from formatting differences
Not every mismatch deserves human review. A spelling variation, extra whitespace or image order can often be normalised safely. Other conflicts go to the heart of the offer: two unexplained prices, different area definitions, contradictory availability, or uncertainty about which unit the record actually describes.
I would not require every incomplete listing to disappear. “Not confirmed” can be a useful state. A buyer may still benefit from seeing a project while one secondary detail remains unknown. But some gaps are too central to hide behind a polished card.
If the identity of the unit is unclear, properties from different units can be merged into a fictional offer. A view from one apartment, a price from another and a floor plan from a third may combine into something that no seller ever offered. At that point, completeness of the page is a liability.
Normalisation should not silently strengthen the claim
Imported data often needs cleaning before it can be shown consistently. Dates, currencies, areas and place names may arrive in several formats. Normalisation is useful until it changes the meaning.
If the source says “approximately 40 square metres,” converting it to 40.00 square metres creates false precision. If a project is advertised “from 120,000,” applying 120,000 to every unit turns a starting price into a unit price. If a description says “near the beach,” a mapping process should not invent an exact walking distance without separate evidence.
The same rule applies to status. “Expected completion” is not “completed.” “Available on request” is not “available.” “Furnished package offered” is not “furniture included in this unit.” A uniform data model should make these distinctions easier to see, not erase them.
Automation is successful when it knows when to stop
There is a natural temptation to add more and more rules until the import appears fully autonomous. That works well for syntax and many obvious inconsistencies. Property context is harder. The same field name can mean different things across partners and projects.
This is where I want a clear handover threshold. If a conflict affects a material buying condition, the system should flag it for review rather than inventing confidence. The person reviewing it can inspect the source, request clarification or decide that the field must remain unknown.
Automation has reached its limit here, not failed. Detecting the point where context is required is useful automation. A system that publishes everything without interruption is efficient only if speed is the sole objective.
Corrections need their own provenance
Once a conflict is resolved, the record changes. I would also preserve why it changed. A price moving from 120,000 to 125,000 could be a new price, a corrected typo or a different furnishing package. Those are not the same event.
Internal notes can stay internal, but the product should know whether an update is a correction, a new commercial term or a remapping to the correct unit. Otherwise the next automated feed can overwrite the reviewed value and recreate the same problem.
Several systems make this harder. A human editor may fix a field today, while tomorrow's import treats the partner feed as more recent and restores the old claim. Without a reason attached to the correction, the platform has no basis for deciding which change is authoritative.
Publishing carries responsibility for presentation, even when the source is external
A property platform will always rely on external information. The limitation is useful because it shows where human checking begins. Developers, sellers, managers and partners are natural sources for many facts. The mistake is not using their data. The mistake is allowing the import process to erase who said what and how strongly it is supported.
I would avoid the opposite extreme as well. A partner source is not suspect simply because it is external. Imported data deserves verification without assuming every source is wrong. The standard should be traceability and proportionate review.
Importing should reduce manual transcription. It should not convert receipt into verification. The strongest system can accept thousands of records efficiently and still stop on one important contradiction. For the buyer, that single pause may be more valuable than a perfectly populated catalogue whose most important conditions no longer have a visible source.