← Perspective AI & Tools

A model that learns from listings learns wishes

Home-pricing tools get better every year. How close they land to the real price depends on what they learn from.

Message me on WhatsApp

It has never been easier to ask a machine what an apartment is worth. You paste the address, the square meters and three photos, and a few seconds later you have a number with decimals and a paragraph explaining it.

In Panama, that number can land as far from the final price as whatever your neighbor guessed a few years ago. The tool has improved a lot. What it has to learn from, not so much.

Can artificial intelligence price a home in Panama?

Artificial intelligence can estimate what a home in Panama is worth, but the result depends on the data behind it. Asking prices on listing portals are not the prices buyers actually paid, and a model needs comparable, verifiable transactions to calibrate itself. Without them, it estimates what sellers are asking.

Where sales are on record, the model has something to work with

In markets where sale prices are published in an orderly way, with dates, an automated valuation compares your home with what buyers recently paid for similar homes nearby. It gets things wrong, like anyone. But you can measure how wrong, because every new sale tells it whether it was right.

When all the model has is listings, that loop never closes. There is nothing to check its answer against. And a listing is an aspiration with photos.

The bias of what doesn't sell

There is a subtler problem than mistaking the asking price for the price paid.

A well-priced unit sells and its listing disappears. The overpriced one stays up. Take a snapshot of everything on the portals on any given day and the expensive units carry more weight than they do in the market, because they are the ones that didn't go.

And when a listing comes down, the model doesn't know why. It may have sold at the asking price, sold well below it, or the owner gave up. To the machine it's the same story: a listing that is no longer there.

Sorting, yes

To be clear: I use these tools, and they save me hours.

For sorting hundreds of listings by area, size and price per square meter, a machine is faster and more patient than any person. It can help flag possible duplicate listings of the same apartment, point out the one priced outside its building's range, and in minutes give you the list of what your property is competing with today.

What you get is a map of supply. A useful one, though it can overstate the market when the sample is dominated by listings that stay up.

What it can't see

What moves the price inside a single building is rarely in the listing. The original 2008 kitchen next to one that was just redone. The eighth floor that faces another tower and the twentieth that faces the ocean. The bedroom on the avenue side that you hear all night. The maid's room turned into a study, which adds square meters on the listing sheet and maybe not on the title.

Then there's the paperwork. Whether the finca (the property's registration number at Panama's Public Registry) carries an old mortgage, whether the registered square meters match the listing, whether the building has approved a special assessment. None of that comes from a portal, and all of it changes what a serious buyer will pay.

How to use it

Ask the machine for the map; build the price from something else. A defensible range needs references to what was paid, not what is asked: appraisals (which aren't the price, but are done by someone who walked the unit), sales a colleague can confirm, how long similar units have been on the market. That was true before AI (a price without a method is still a rumor). What changes is how fast you get to the hard part, which is exactly the part that leaves the agents who only did the easy work without a job.

If you asked an AI what your property is worth and you liked the number, be a little suspicious of that feeling. A model trained on asking prices learns wishes, and you already knew yours.

Frequently asked questions

Can I use ChatGPT to check whether a price is reasonable?

As a first filter, yes. Ask it to separate asking prices from prices actually paid, and to tell you what it can't see. If it can't point to a single verifiable sale, what it gave you is an average of listings.

Dario Jhangimal
Dario Jhangimal
Licensed real estate broker · PN-1240 · SpotOne Realty

Did an AI give you a price for your property? Send it to me. We'll look at what it's based on, what information it's missing and how well it holds up against the market. Let’s talk.

Perspective is editorial, informational content. It is not legal, tax or investment advice. Every transaction is assessed in its own context.