How we estimate prices

Short answer: this is not just ChatGPT. Below we openly explain what the model weighs and how we verify its accuracy.

Where the price comes from

Two separate systems work in every analysis. The price does not come from a ChatGPT-style model: it is computed by a statistical model trained on tens of thousands of real Baku listings: it weighs 40+ factors and produces a fair price range. The AI (language model) does not set prices: it explains the result in plain language and verifies the claims made in the listing text. The number you see is not a chatbot's opinion - it is a computation grounded in market data.

What the model weighs

40+ factors across five main groups:

  • Location: district, nearest metro, distance to the center and the sea
  • Apartment: area, room count, floor and building height
  • Building: new or old construction, residential complex
  • Condition and documents: renovation, title deed, mortgage eligibility
  • Market: district-level average prices and supply

Accuracy: the independent test

To validate the model we held out 12,832 listings it never saw during training, then compared its predictions against those listings' real prices.

Range accuracy
83%
of listings had their real price inside the predicted range
Median deviation
6.4%
half of the predictions deviated from the real price by less than this

These figures are based on asking prices and may shift over time. No prediction is a guarantee: the model shows you the market picture, the decision is always yours.

What the AI does and doesn't do

The AI does
  • Reads the listing text and verifies the claims: renovation, floor, documents
  • Explains the model's result in plain language
  • Flags risks worth your attention
The AI doesn't
  • It doesn't set the price: that is the statistical model's job
  • It doesn't make guarantees: markets move
How we estimate prices - Methodology | Mənzil.ai