How the real value of a second-hand listing is calculated

Most second-hand valuation tools hand you a number and never say where it came from. This guide walks through the calculation behind Sowld's estimate, step by step, because a number you cannot check is useless to you.

The problem: there is no price list for used goods

A new car has a list price. A 2019 road bike with 8,000 km and a replaced derailleur has nothing of the kind. The only honest reference is what reasonably similar items are being asked for, right now, in the same market.

That is why Sowld does not consult a price database. It builds the reference every time, from the listings themselves.

Step 1 — Group the comparables

Listings are grouped by brand + model. A group is that item's market at that moment.

When only the brand can be extracted from a listing and not the model, the group is far noisier. On a Vinted grid, for instance, the card shows brand, size, condition and price — never the actual product name. A "Chanel" group could be pairing loafers against boots with nothing in the extracted text to tell them apart. Brand-only groups therefore need more listings before they are trusted.

Step 2 — Normalise for condition

Two listings of the same model are not comparable if one is "like new" and the other is "used, visible wear". Every price is therefore brought back to a shared reference condition before anything is compared.

The correction applied is roughly 10% per condition point away from the average. An item in better-than-average condition is worth more, a worse one is worth less, and the same scale applies to every listing in the group.

Step 3 — Median, not mean

The group's reference value is the median of the normalised prices, not the average.

This is deliberate. A single absurd price — a typo, a seller aiming high, a collector's piece that landed in the wrong group — moves the mean a lot and the median almost not at all. With the handful of comparables you usually get in second-hand, that difference is enormous.

Step 4 — Back to the listing's own condition

The median is expressed in the reference condition. For each individual listing it is adjusted back to its stated condition, using the same scale as step 2. That result is the estimated value of that specific listing.

Step 5 — The threshold

A listing is flagged as a potential deal when the asking price is at least 25% below the estimated value.

The threshold is not that high by accident. Below that gap, the difference sits inside the normal noise of second-hand: different shipping, one extra accessory, a better photo. That is not a deal, it is variance.

When more listings are needed

A group with one other listing in it is not a market. A minimum number of comparables is required for the median to mean anything, and that minimum rises again for brand-only groups. When there are not enough comparables, Sowld does not estimate — saying "I don't know" is more useful than an invented number.

The limits, stated plainly

• The reference is asking prices, not sold prices. Those are what is publicly observable on an active listing. If a whole group is overpriced, the estimate inherits that error.
• Condition is whatever the seller declared. No software can see a scratch that is in neither the photos nor the text.
• Authenticity is not verified. A well-described replica looks like an exceptional deal precisely because it costs less than the real thing. An anomalous discount is a reason to look harder, not to buy faster.
• The seller is not verified. Feedback, history and payment method remain your check to make.

How to use the estimate

Treat it as a filter, not a verdict. Its job is to turn fifty listings into the five that deserve two minutes of attention. The decision to buy stays a human one, made by looking at the photos, reading the description and asking the seller a question.