AI Textile Sorting Explained: Bulk, Retail, and E-commerce Aren't the Same Job
Every donated garment gets a decision in about ten seconds. Keep or reject. Which grade. Which price. Which bin. Most operations make that decision millions of times a year, by hand, with whoever is on shift.
AI textile sorting promises to make those decisions faster and more consistent. But most of what gets sold as "AI sorting" treats one job as if it were three — and that's where operators waste money.
I spent three years on sort floors before I built SortIQ. Here's how the work actually breaks down, and where AI earns its cost.
The three jobs hiding inside "sorting"
A sort floor isn't one workflow. It's three, with completely different economics.
| Job | Volume per day | Value per item | The real decision |
|---|---|---|---|
| Bulk / export | Highest | Lowest | Grade and route — not price |
| Retail / in-store | High | Low to mid | The right floor price, every time |
| E-commerce | Low | High | What it will actually sell for online |
A tool that spends the same effort on a $1 T-shirt headed for a bale as it does on a designer coat headed for an online listing is either too slow for bulk or too shallow for e-commerce. Usually both.
Job 1: Bulk and export sorting
Recyclers, exporters and the back end of every thrift operation move product by the pound. Nobody prices a single item here. The questions are: what grade is it, which container does it go in, and who is moving how much.
What AI should do in bulk:
- Count production per sorter. Most recyclers lose visibility the second product leaves the Gaylord and hits the conveyor. You can't pay incentives on output you can't see.
- Log grade and category as items move, so the container manifest builds itself instead of being hand-tallied.
- Stay cheap. Bulk runs at the highest volume in the building. If AI costs more per item than the item earns, it's dead on arrival.
What AI should not do here: run a heavy image model on every rag. Trained sorters already grade by sight. Let them state the grade, log it, and move on.
Job 2: Retail and in-store sorting
This is where most thrift revenue is won or lost — and where the leak is quietest.
New sorters default to one safe price across an entire rack. Every blouse gets the "good" price, whether it's a fast-fashion piece or a premium brand in near-new condition. Nobody notices, because nothing looks wrong. The money just never shows up.
What AI should do in retail:
- Recognize brand and condition quickly, and apply a price tier — not a single default.
- Ask the right questions. Holes, stains, wear. A short condition prompt is what keeps good/better/best pricing from collapsing into "good."
- Learn from your own register. If an item sells in a day, the price was low. If it sat until markdown, it was high. That feedback is what separates a pricing system from a lookup table.
Job 3: E-commerce sorting
A small share of donations are worth far more online than on a rack. Pulling them out and pricing them correctly is pure upside.
This is the one place where heavy AI is worth the cost: multiple photos, full brand and material ID, and pricing against what comparable items have actually sold for — not what sellers are asking. Asking prices are noise. Sold prices are data.
The part most tools miss: routing
The value isn't three separate tools. It's the handoff. A quick grade or brand read at the bulk or retail stage should flag the handful of items worth escalating to e-commerce — and leave everything else on the fast, cheap path.
Get the routing right and you spend serious compute only on the items that justify it. Get it wrong and you either miss the upside or burn margin on every rag.
Why the sort station is the right place for AI
The sort station is the only point where every item passes through a human's hands with a decision attached. Capture data there and you get item-level visibility nobody else in the chain has: what came in, what grade, what price, what sold.
That's also why hands matter. Sorters can't stop to tap a tablet 3,000 times a shift. SortIQ runs on Meta Ray-Ban smart glasses so the sorter looks, speaks and keeps moving. The AI answers by voice. No screen in the way.
Why this matters now
Two pressures are converging:
- Labor. Training a new sorter to price well takes weeks most operations don't have. Turnover resets the clock.
- Regulation. California's SB 707 is putting textile tracking on the map, and other states and federal lawmakers are watching. Operations that already capture item-level data at the sort line will be ready. Everyone else will be building it under deadline.
The operators who start capturing sort data now set their own baseline. The ones who wait will be told what to report.
What to ask any AI sorting vendor
- Which of the three jobs is your tool built for — and what does it cost per item in each?
- Do you price off sold comps or asking prices?
- Does the system learn from my own sell-through, or just a generic database?
- What happens on my floor when the Wi-Fi is weak?
- Who owns my data?
If the answers are vague, the product probably is too.
See SortIQ on a real sort line. Book a 30-minute walkthrough and we'll map which of the three jobs is costing you the most.