Running an Unattended Micro-Market: Shrink, Lockdown, and Reconciliation

Micro-markets are no longer the experiment. Operators reported 55,770 micro-market locations in 2024, up from 42,900 the year before — roughly a 30% jump in a single year, according to Automatic Merchandiser's State of the Vending and Micro Market Industry report.

Most of the writing about small-location markets — including ours — argues about whether the economics work: how few people a location can have before a purpose-built kiosk stops paying for itself. That floor is genuinely contested. Industry guidance has put it anywhere from a hundred and fifty people to several hundred, operators pushing the format into smaller sites report workable numbers far below that, and our own breakdown of the small-location economics lands on a traditional break-even in the 150–250 range. The honest answer is that it depends on your product mix, your margins, and how far you drive. If that's the question you're on, start there — this page won't rehash it.

This page is about the question that comes next, and gets much less attention: once you've decided a location is worth it, what does it actually take to leave a market alone in it?

What "unattended" actually requires

Strip away the hardware and unattended retail is four problems:

1. A customer has to be able to ring themselves up with nobody there to help.
2. The device has to stay a checkout, not become a browser, a game, or a settings menu.
3. You need a record of what was sold that you still trust when the counts drift.
4. The money has to land somewhere you can reconcile against inventory.

Everything below maps to one of those four.

1. The customer rings themselves up

TallyTill turns an Android tablet or phone you already own into the checkout. There are three ways to get an item into the cart, and all three are started by the shopper:

• Scan the barcode or QR code. The device reads it and looks the item up in your catalog. This is the workhorse for packaged goods, and it runs locally on the device.
• Tap to snap a photo. For the things without a barcode — a pastry from a local bakery, loose fruit, a deli item — the shopper taps, the device takes one picture, and server-side AI matches it against your catalog.
• Tap the item on screen. Your catalog is browsable and searchable. For a market with 60 SKUs, this is often the fastest path.

One thing worth being blunt about, because the category is full of fuzzy claims: the camera does not watch the room. It is not passively monitoring the shelf, it is not tracking what a customer picks up, and nothing lands in the cart on its own. Recognition happens when — and only when — the shopper initiates it. That is a real constraint, and it's also why the format runs on an ordinary tablet instead of a ceiling grid of sensors.

Nobody is standing there to answer questions, so it helps that the kiosk can answer some itself. TallyTill includes a voice assistant, Tally, that you switch on in settings — with it enabled, a shopper can ask out loud what's in something, where an item is, or how to pay. It's off until you turn it on, which is the right default for a device sitting in a quiet lobby.

How the shopper pays — your staffing model decides

This is the part operators skip, and it's the one with an actual rule behind it. Stripe classifies retail environments by a single test: is a person on the property? Not whether anyone is watching the device — whether a human being is there at all.

• Attended — "staff support is available, either per device or per location."
• Semi-attended — "a customer independently interacts with the card reader, but staff may be available in the store or nearby if needed."
• Unattended — "no staff is on location to help the customer if needed."

Stripe's pre-certified card readers and Tap to Pay support attended and semi-attended retail environments, but don't support unattended. A market alone in a lobby overnight is the textbook unattended deployment — so for that site, the QR code on the checkout screen is the primary payment path, not a fallback.

This is not "no credit cards," and that's the assumption worth correcting. The QR opens a Stripe hosted checkout on the shopper's own phone — an ordinary online card payment, with no attendance restriction attached to it. Cards work. Wallets work. Buy-now-pay-later works. What's unavailable unattended is card-present capture on your tablet: Tap to Pay and Stripe readers. The shopper's phone becomes the reader instead, which is why this format needs no payment hardware at all.

Now the nuance that changes the answer for a lot of micro-markets: most of them sit inside a host building that does have staff. An office with a receptionist. A gym with a front desk. A plant with a facilities crew on shift. If someone is on the property and could help a confused shopper, that's semi-attended — and readers and Tap to Pay are permitted. The host building's staffing, not yours, is usually the deciding factor, and most operators have never framed the decision that way.

There's no attendance switch to flip in software — this is a placement judgement you make per location, and it decides whether you buy a reader for that site at all. If a location is borderline, or you're weighing overnight access against staffed hours, the classification rules and what each tier permits are laid out in Attended, semi-attended, unattended: how Stripe's rules decide your payment options.

2. Locking the tablet down

A consumer tablet left alone in a lobby is a consumer tablet until you lock it. Kiosk mode uses Android screen pinning to pin the app to the checkout screen — no home button out, no browser, no settings, and the right-click context menu is disabled. Getting out requires a hidden gesture plus your admin PIN, a manager QR code, or disabling kiosk mode remotely from the web portal.

Screen pinning is a meaningful lock, not a vault. It depends on a one-time hardening checklist on the device — a device PIN, and requiring that PIN to unpin — and until you've done that, pinning is much easier to escape than it looks. Do the checklist before the tablet goes out, not after. For a lobby or a badge-access break room, that's the right tier. If you're placing a market somewhere genuinely public and hostile, budget for a locking enclosure; the software lockdown and the physical one are separate problems.

The upside of using your own hardware is unglamorous but real: when a tablet dies, you replace it from any electronics store that afternoon instead of filing an RMA and losing a week of sales at that location.

3. Shrink, and the honor-system problem

This is the anxiety that stops most operators from opening a small unattended location, and it deserves real numbers rather than reassurance. Operators responding to the industry survey report about 4% shrink on average. The National Retail Federation's last shrink survey put retail shrink overall at 1.6% of total sales (FY2022 — NRF has since discontinued the series).

The useful thing to understand is what that 4% is made of. In an unattended market it is rarely one person walking out with an armful. It's the accumulation of small leakage: the "I'll get it next time" that never gets gotten, the item that never made it into the cart, the count that quietly drifts between visits. Deliberate theft exists, but it's usually not the biggest line.

Two things address different parts of that:

Paying at the point of sale — by QR on the shopper's phone, or on a reader at a semi-attended site — removes the honor step entirely for most of the leakage. There is no "pay later" to forget.

Cart photos, when you enable them, are captured at checkout and attached to the transaction. When a shopper disputes a line item, or when one location's counts keep drifting and you want to know whether it's theft or a catalog error, you have a picture of what was on the counter alongside what was rung up. Two caveats. First, what this is: an audit trail and a mild deterrent, not prevention — it tells you what happened after the fact, which is what you need when deciding whether a location is worth keeping. Second, it's a choice, not a default behaviour: you set which checkout events capture a photo, and you can turn it off entirely. Photographing people at a snack shelf is a trade some locations won't want to make, and a workplace especially may have a view on it. Decide it on purpose.

If you take cash, that's switchable per device. Attended locations get change-due tracking; unattended ones can run honor mode, which hides the change calculator and shows the shopper a short envelope code to write on a drop-box envelope, so you can match cash to the sale later.

4. Stocking a location you visit weekly

The real operational constraint on a small location isn't shrink, it's service frequency. A market that needs daily restocking can't pay for itself at thirty people. A market you visit weekly can.

That works only if you plan the route from what sold rather than from a shelf count. Sales report to the web portal in real time, per device and per location, so before you load the van you already know which SKUs moved, which have been sitting for three weeks, and which locations don't need a visit this cycle. Sell-through per item is what tells you to stop carrying the protein bar nobody buys — and at a 60-SKU market, cutting dead inventory matters more than it does at a 400-SKU one.

Tax is applied by jurisdiction, which starts mattering the moment your locations cross a city or county line.

Running several locations from one portal

Each tablet registers as a device. From one web portal you manage the catalog and pricing, see payments and reconciliation across all locations, and pull sales reports. Cash and check are set per device, so the lobby market can be card-only while the staffed location takes cash; the Stripe payment methods you accept — wallets, buy-now-pay-later — are set once for your account rather than per site.

That's also how a mixed estate works in practice. The unattended sites run the QR path with no payment hardware in them at all, and the semi-attended ones — the locations inside a staffed building — can add a reader on top. Same catalog, same portal, same reporting; the difference is which payment path is available at the shelf.

Manager PIN, discounts, and refunds are gated and logged, so a part-time route driver can fix a mis-ring without you handing over the keys, and you can see afterward who did what. Adding location number four is adding a tablet, not procuring and installing a terminal. That is the whole argument for this format.

When the connection drops

A lobby closet with one weak access point is a normal deployment, so this matters more here than in a staffed store. Sale records are captured on the device and sync when connectivity returns. What keeps working varies by function:

The card row is the one people get wrong in both directions. Card sales can complete offline, but only through Stripe's store-and-forward on a Bluetooth Stripe reader — never Tap to Pay — and only when you've enabled offline card payments for your account and that reader has already connected online at the location beforehand. Queued sales forward automatically once connectivity returns. The full conditions and the risk that comes with them are in Taking card payments when the internet goes down; read it before you rely on it.

Note what that implies for a genuinely unattended site: store-and-forward runs on a Bluetooth reader, a reader is card-present, and card-present isn't available unattended. So the offline card path belongs to your semi-attended locations. It's a good reason to know which of your sites is which before you buy hardware for any of them.

Practically, for an unattended location with unreliable wifi: make sure your unbarcoded items are also tappable on screen so the shopper always has a working path, and treat the connection as the thing to fix — a cheap mesh point or a cellular hotspot in the closet does more for that site's revenue than any payment accessory.

Where this fits — and where it doesn't

Good fit: locations under the headcount your kiosk economics require; buildings where you already have a tenant relationship and want to test demand before committing hardware; apartment and coworking sites; a secondary market inside a location that already has a main one; anywhere you'd otherwise place vending and leave revenue on the table.

Not the right tool: if you need to gate physical access to the product — a locked cooler that opens only after a card pre-auth — that's a different category and TallyTill doesn't do it. Very large markets with long queues will be better served by a bigger, purpose-built screen. And if your model depends on passive, walk-out-style detection, that isn't what this is.

Pricing and getting started

TallyTill is transparent about pricing — current numbers, in full, are on the pricing page. We don't put figures in articles like this one, because articles outlive price changes and go stale quietly.

To open a location: load your catalog in the portal, get barcodes on everything you can, put the rest on screen with a photo, connect Stripe, decide whether the site is unattended or semi-attended so you know whether you need a reader, do the kiosk hardening checklist, and mount the tablet. Then let the first two weeks of sales data tell you what the location actually wants to buy.

Keep reading

• Attended, semi-attended, unattended: how Stripe's rules decide your payment options — the classification test in full, and what each tier permits.
• Micro market "lite": when an unattended market pays off in a small location — the economics and payback side of this decision.
• How to start a self-serve fridge without a $10,000 smart fridge — if the market you're picturing is one cooler rather than a shelf run.
• Taking card payments when the internet goes down — the full offline card rules.

---

Sources

• Technology and convenience fuel growth — 2024 State of the Vending and Micro Market Industry, Automatic Merchandiser / Vending Market Watch — micro-market location counts and operator-reported shrink.
• National Retail Federation, National Retail Security Survey 2023 (FY2022) — 1.6% retail shrink benchmark; series since discontinued. Also cited in William Blair, "Overcoming Theft and Shrinkage in Unattended Retail Operations".