Sold Price vs. List Price: Why Most Resale Pricing AI Gets It Wrong
The president of a Goodwill region put it to us bluntly on a call: list price is irrelevant. It's inaccurate. He only trusts what things actually sold for.
He's right, and it's the single best filter for judging any AI pricing tool on the market right now.
The problem with asking prices
Anyone can list a jacket for $200. That doesn't mean it's worth $200. Online marketplaces are full of optimistic, stale and duplicate listings. Price off that pool and you'll overprice — then mark down, then mark down again.
The errors are big. TRUSS, a resale pricing platform built on sold-transaction data, reports that its own benchmarks found half of items on major resale platforms mispriced by 25% or more. It also found general AI chatbots disagreed with real market prices by about 23% on average (TRUSS). Those are a vendor's own numbers, but the direction matches what operators see every day.
A chatbot that "knows" what a brand is worth is guessing from the open web. That's not pricing. That's vibes with a confident tone.
What good resale pricing looks like
The serious players have converged on the same approach:
- Sold comps only. Price from what comparable items actually sold for, not what sellers are asking.
- More than one marketplace. One platform gives you one audience's view. Pricing only from one creates gaps.
- Weighted matches. A comp with the same brand, model, size and condition counts more than a loose match.
- A range and a confidence score. Not a single magic number. If the system found three weak comps, it should say so.
If a vendor can't explain how its prices are built along these lines, be careful.
But comps aren't the whole answer
Here's what most AI pricing pitches skip: everyone can license sold-comp data. The image recognition is commodity too. Several vendors can identify a brand from a photo in seconds.
So where's the real advantage? Your own register.
Your stores have their own shoppers, their own neighborhoods and their own price sensitivity. An item that sells in a day was probably underpriced. An item that sat until markdown was probably overpriced. That signal — what actually sold, at what price, how fast, in your stores — is data no competitor can buy.
Savers made the same bet. It built ThriftIQ on its own sell-through data from more than a billion pounds of goods a year, combining brand with seasonality and sell-through to set prices (Gizmodo).
Match the pricing method to the item
Not every item deserves the same pricing effort. That's how you keep costs sane.
| Item type | Share of volume | Right pricing approach |
|---|---|---|
| Everyday apparel for the floor | Most of it | Fast brand and condition read, priced from tiers calibrated to your own sales |
| Branded or premium pieces | Small | Pulled for a closer look; priced against sold comps |
| High-value items for online sale | Smallest | Full photos and ID, sold-comp range with a confidence score, draft listing |
Spend heavily on the few items that justify it. Keep the rest fast and cheap.
How SortIQ approaches pricing
SortIQ starts on the sort floor, where the price decision actually happens. Sorters wearing Meta Ray-Ban glasses get a suggested price by voice, built from your own pricing logic and brand tiers.
Our pricing methodology follows the rules above: sold data over asking prices, ranges over single guesses, and a feedback loop from your point-of-sale so recommendations get sharper as your stores sell. Items that look like online candidates get flagged instead of disappearing into the general rack.
Five questions to ask any pricing AI vendor
- Do you price from sold transactions or listings?
- How many marketplaces feed your comps?
- Do you give a range and a confidence score?
- Does the system learn from my sales, or only a generic database?
- What does it cost per item at my volume?
See how SortIQ prices on a live sort floor. Book a walkthrough.