How Kartalens works

Asking prices: what sellers and landlords ask in public property listings in and around Addis Ababa. Today they come from public Telegram property channels; other sources will be named here as they are added. Not sale prices, and not valuations. Under Proclamation No. 1357/2024 valuations are done by certified valuers; Kartalens helps you ask better questions before you get one.

Every post is stored as published, including later edits and deletions. An AI model reads each post (Amharic or English) and extracts whether it is for sale or rent, the type of property, bedrooms, size, price and location words. Prices are converted to birr (sale: total price; rent: monthly), and values that don't make sense, such as a size read as a price, are flagged and left out of every statistic.

Locations are matched against a list of Addis Ababa neighborhoods and sub-cities with their English and Amharic spellings. Posts rarely give an address, so we record how precise each location is and never invent a map pin. When the model returns Amharic location words that don't appear in the post, we ignore them and match the post itself.

The same property is often posted many times. Reposts of the same text, including price changes, are grouped into one property with a price history, and every statistic counts each property once. Listings with similar specifications but different wording are kept separate, because many units in one project look alike.

Apartments and houses for sale are compared on price per m², since almost every sale post states a size. Rentals are compared on monthly rent, since few rental posts state a size. We show the median and the middle half (25th to 75th percentile), and we don't show a figure with fewer than 5 properties behind it. Each table also shows how much of the data comes from the largest source channel.

Off-plan units (developer prices for apartments not yet finished, often paid in instalments) are a different market from resale, so they get their own table and never enter resale figures, the index or yields. Listings in Sheger City (Legetafo, Tafo, Furi) are shown as their own group but left out of the Addis index.

The sale price index compares like with like. Each quarter, the median price per m² of each group (property type × sub-city) is compared with the previous quarter, using only groups with at least 3 properties in both, and weighted by how common each group is. A quarter is published only if at least 3 groups covering half its listings support it. The rent index works the same way over half-years and is experimental: it does not yet adjust for the mix of bedroom counts.

For a listing you check, we look for at least 5 similar properties asking in the last 12 months: first in the same neighborhood with the same bedrooms and a size within 20%, then with looser specs, then in the whole sub-city. We never compare across the whole city; if there aren't enough similar listings, we say so. For sales the range is built from price per m² and scaled to the property's size. Off-plan units are compared only with other off-plan units, and resale only with resale.

We test Price Check on listings we already know: hide one property, price it using only listings seen in the year before it, and compare with its actual asking price.

No test results published yet.

• Asking prices usually sit above final sale prices; by how much is not known.
• Coverage depends on the channels we follow; one agency currently supplies much of the data.
• An agency re-posting near-identical text for similar units can be grouped as one property.
• Locations are only as precise as the posts; many name a sub-city, not a neighborhood.