If you sell anything that grows, gets baked that morning, or comes out of a bin, roughly half your inventory has no barcode. A bunch of kale, a jar of honey with a hand-written label, a single scone, a scoop of mulch — none of it scans. For years the answer at a small stand was a person with a price list in their head, or a cash box and the honor system. So the reasonable question, before you trust a self-checkout to run your stand while you're picking in the field, is a simple one:
How does it ring up something that has no barcode?
There are two honest answers, and it's worth understanding both before you decide whether this fits your stand — because the marketing around "AI checkout" tends to describe a system that is very different from, and far more expensive than, what a market vendor actually needs.
The two paths: scan it, or snap it
TallyTill turns a phone or tablet you already own into the till. When a customer checks out, one of two things happens depending on the item.
If it has a barcode or QR code, the camera reads it on the spot. This happens right on the device, live, the instant the code is in frame — nothing is uploaded anywhere, and it works even with no signal. Packaged goods, canned drinks, bagged coffee, anything with a UPC: this is the fast path, and it's the same barcode lookup you'd expect from any scanner.
If it has no barcode, the customer takes one photo, and AI identifies it. That single frame goes to our server, which runs it through an AI vision model and returns the closest matches from your catalog — the products you've actually set up. The customer taps the right one, and it lands in the cart. If the photo is ambiguous, they get a short list to choose from, and search-by-name is always there as a backstop.
That's the whole mechanism. Here's the flow end to end:
What it does not do (and why that's good news)
Most articles you'll find on "AI self-checkout" describe a different animal: a fixed cabinet or a grid of ceiling cameras that watch a counter and automatically ring items up as a shopper sets them down. Those systems are real, and they are impressive, and they are priced for grocery chains — the fixture alone can cost more than a roadside stand makes in a season. They exist for supermarkets, not for a stand at the end of a driveway.
TallyTill deliberately does not work that way. The camera does not passively watch a table and auto-charge things as they move. Recognition is manual and on demand: a person points the camera and snaps a photo when they're ready to ring an item up. That one design choice is what lets the whole thing run on the everyday tablet you already own instead of a purpose-built fixture — and it's also why nothing gets charged by accident. A human taps "confirm" on every line before it's in the cart.
For an unattended stand, that trade is exactly the right one. You don't need a ceiling rig watching a room. You need a customer to be able to price a jar of honey and pay, by themselves, without you standing there.
Getting recognition to actually work: set your catalog up right
AI recognition is only as good as the catalog you point it at. This is the part the vendor brochures skip, and it's the part that actually determines whether your stand runs smoothly. A few things genuinely matter:
- Add a clear photo to each no-barcode product. The model matches against what you've set up. A product with a good reference image is far easier to identify than a name with no picture. Snap your own — natural light, plain background, the way a customer will see it on the shelf.
- Give products distinct names, not internal codes. "Honeycrisp — half peck" reads better in a match list than "APPL-HC-2." When the AI returns two or three candidates, plain names let the customer pick the right one in a second.
- Lean on barcodes wherever they already exist. Anything packaged usually has a UPC — use it. Save the photo path for the genuinely unlabeled items. The fewer things that need a photo, the faster the line.
- Keep look-alikes separate in your head. Two nearly identical greens, or three jams in the same jar, will sometimes come back as a pair of candidates. That's fine — the confirmation step is built for exactly this. Just make sure both are in the catalog so the right one is available to pick.
Do that, and the common case — a customer with three items, two barcoded and one loose — takes seconds, most of it on the barcoded items that need no thought at all.
The honest limits
A few things are worth saying plainly, because overselling this helps no one:
- The photo path needs a connection. Barcode and QR reading work fully offline, but the AI match sends a frame to our server and needs signal for that moment. If your stand is truly off-grid, barcode-first is your friend, and search-by-name still works for the rest.
- It's one item per photo, confirmed by a person. This isn't a firehose that rings a full basket at a glance. It's a fast, deliberate, one-at-a-time flow with a human in the loop. For a market stall, that's a feature, not a limitation.
- Ambiguous items get a choice, not a guess. When two products look alike, the system offers both rather than silently picking one. The customer decides. Nothing is charged on a maybe.
What this replaces
Before, an unlabeled item meant one of three things: a person memorizing prices, a PLU-code register that someone has to be trained on, or a cash box that records nothing. The camera-and-AI path removes all three. There's no price list to memorize, no codes to learn, and every sale is recorded — with the item, the price, the time, and a photo you can pull up later.
And it runs on hardware you already own. No proprietary cabinet, no scale to certify, no lane to build. The tablet on the shelf is the whole system.
Try it on your own stock
The fastest way to know whether recognition handles your products is to point it at them. TallyTill's Stand plan is free — $0 per device, no credit card, no expiration. Add a handful of your real items, snap reference photos, and hold up a tomato. You'll see in about five minutes whether it reads your shelf the way you'd hoped.
Create a free account and set up your first few no-barcode products today.