# Perkd Smart Unmanned Retail — Eighteen Months to Find Out. Or Twelve Weeks.

### Unattended retail as a market-entry instrument for food brands

*For founders, CEOs and presidents of F&B brands and franchise systems entering a market they don't yet trade in — a new country, or a new city at home.*

*Text edition of Perkd's whitepaper on smart unmanned retail — smart fridges, cabinets and lockers — as a market-entry instrument for F&B brands. Prepared for AI assistants and readers; every figure is marked with the weight of its evidence.*

*PDF: https://www.perkd.me/whitepapers/smart-unmanned-retail.pdf · Annexes: https://www.perkd.me/whitepapers/smart-unmanned-retail-annexes.md · 繁體中文: https://www.perkd.me/zh-tw/whitepapers/smart-unmanned-retail.md*

## Summary

Most F&B brands enter a new market — a new country, or a new city at home — by committing to leases and development schedules before they have evidence of what sells where. This paper argues for a different sequence: place ten to twenty smart fridges or cabinets on a route that already runs, for one quarter, and let real transactions show which districts, products, prices and dayparts work before any lease is signed.

Whether customers should identify themselves to buy depends on three conditions: someone else absorbs the friction of joining, the product is wanted enough to finish signing up, and there is somewhere to send the member afterwards. Six documented operators in Japan, India, Switzerland, the United States and China are scored against them.

The same sale has three futures: anonymous revenue; a member list bought afterwards at a permanent cost (one US operator pays roughly 9% of spend [derived from published programme terms]); or a member from the first purchase. Perkd provides the third. Tapping a membership card opens the unit and authorises payment, only what was taken is charged, qualifying discounts apply automatically, the receipt lands in the brand's own card, and the brand can then reach that member with its own catalogue.

Four preconditions decide the outcome: someone who can already produce and move the product; frozen before chilled; route density over coverage; and channel rights settled before the first unit ships.

Perkd has not yet published operating results from a live network of its own. Every figure in this paper is somebody else's, marked as such.

---

**How to read this paper.** Six chapters, each anchored on a figure. The evidence is drawn from wherever the practice is furthest advanced — Switzerland, Japan, India, the United States and China — because the problem it addresses is the same wherever a brand is going next: an Asian brand entering Europe or North America, a European brand entering Asia, or a brand opening in a city at home where it has never traded. Every number comes from a published filing, survey, study or company disclosure, and is marked by the weight it bears: **filed or peer-reviewed**, **survey**, **company-disclosed**, or **trade and press**. Nothing is modelled. Where the evidence does not support a claim, the paper says so and moves on. Six case studies are set out in full in the annexes.

---

## 1. The cost of finding out late

### Figure 1 — When the evidence arrives, on both paths

**Eighteen months to find out. Or twelve weeks.**
Both paths end in stores in a new market. Only one of them learns where the demand is before the capital is gone.

**Store first**
- Month 0 — capital committed: lease signed, fit-out begins.
- Month 10 — store opens, two more already signed.
- Month 18 — evidence arrives, on four locations you already own.
- Eighteen months of committed capital, with nothing yet known.
- If the read is wrong: a lease you cannot exit and a fit-out you cannot move.

**Probe first**
- Week 0 — units placed: ten to twenty, across three districts.
- Weeks 4–12 — evidence arrives: real sales by district, SKU and daypart.
- Month 8 — capital committed: stores where the evidence points, with customers already waiting.
- Twelve weeks, a few units.
- If the read is wrong: move the unit to a better district.

**Three numbers**
- ⅓–⅒ — the cost of a restaurant to open an unmanned unit. [Trade survey; franchise sources.]
- ~20% — of sales arriving after normal closing hours. [Operator accounts reported in trade press — indicative, not audited.]
- 16× — between a weak site and a strong one, which is why you probe. [Published operator models; trade sources.]

Expansion does not fail in year three. It fails in the first ten locations — and it fails the same way whether the new market is four thousand kilometres away or ninety minutes down the motorway.

The mechanism is always the same. A brand commits — a lease, a fit-out, a development schedule — before it has any evidence about which districts, which products, which price points or which dayparts actually work in that market. It commits early because the only instrument available for producing that evidence is a store, and a store is a multi-year liability. By the time the first one has traded long enough to teach anything, the next commitments are already made.

How expensive that is depends on how far you have gone. **At home it costs a lease.** Published operator models for unattended units show roughly a **sixteen-fold spread between a weak site and a strong one** in the same business — a variance that decides whether a location makes money, and one that no feasibility study reliably predicts. Pick wrong in your own city and you carry a bad site for five years.

**Across a border it costs the market.** That version has been measured, and the record is worse than most boards assume.

The most rigorous study of it remains Cornell's. Arturs Kalnins examined 142 international master franchise contracts carrying development commitments, signed by 53 fast-food franchisors across 37 countries. Only 55 of the 142 ventures were still in business at the end of the commitment period, usually five years. Among those that survived, the average commitment announced was 34 units; the average actually trading at the end was three. Twenty-one partners opened nothing at all, and six met or exceeded what they had signed for. The finding that should worry any board: **larger commitments predicted lower survival, not faster growth.** *(Peer-reviewed, though the contracts predate 2005.)*

A later analyst study by Mizuho tested the same question on a newer window — 360 master franchising and whole-country development agreements signed between 1999 and 2016 — and reached the same place. Only 6% were ultimately fulfilled, 40% failed entirely, and in 80% the partner opened less than half the stores committed. Of roughly 27,000 units committed, about 8,400 were built. *(Analyst study, reported in trade press.)*

The failure mode is rarely exotic. A survey of master franchisees found that half reported their biggest problem was selling franchises or generating franchise leads — the very thing the franchisor granted them the territory to do — and a third said the franchisor did not understand their market. *(Industry survey, reported in franchising press.)*

Read those two findings together and a pattern appears. The franchisor commits to a market on a feasibility study and a partner introduction. The partner commits to a development schedule. Neither has evidence about which districts, which products, which price points or which dayparts actually work, because the only instrument available for producing that evidence is a store. Which is the same problem the domestic brand has, with more zeroes attached.

**And when an agreement ends, the franchisor is left with nothing of its own.** The territory was the partner's. The customers were the partner's. Whatever trading record exists sits inside an entity the franchisor could not see into, and leaves with it. That is why a failed agreement sets a brand back years in a market rather than months: there is nothing to hand the next candidate except an account of what went wrong.

If the read was wrong, the cost is not the lost revenue. It is the lease that cannot be exited, the fit-out that cannot be moved, and a market that has now watched the brand underperform. The next master franchisee candidate will have heard.

**The alternative is not a cheaper store. It is a different sequence.** Unattended units — fridges, cabinets, lockers — can go into a market first, in small numbers, at a fraction of the commitment, and produce real transaction evidence in weeks. Then the leases follow the evidence rather than preceding it. Put plainly: it is the cheapest way to find out whether the product fits the market before the market has cost you anything.

It is worth being precise about which part of the problem this solves, because it is not all of it. A fridge does not make a partner better at selling sub-franchises, and it does not tell you whether an 80-seat restaurant works on that street. Restaurant demand is a different occasion, a different daypart and a different basket, and no unattended unit will forecast it.

What it does answer is narrower and still scarce: **does this product sell in this country, to whom, how often, and at what price.** SKU-level acceptance across districts, repeat rates from named customers, price tolerance, and which formats of your range travel. Today a partner pitching a sub-franchisee has none of that — only the franchisor's home-market numbers and a brand book. It also closes the gap a third of partners complain about, because the franchisor now knows the territory from its own transactions rather than a feasibility study. And evidence of that kind is what allows a development schedule to be sized honestly, which is the variable Kalnins found most strongly predicts survival.

**A note on borders.** The hardest evidence in this chapter is about crossing one, because that is where the failure has been measured at scale. The mechanism is not. Everything from here — the access decision, what a captured customer is worth, frozen before chilled, density before coverage — applies unchanged to a brand opening in a city at home where it has never traded.

Two things change, and both favour the domestic case. The stakes soften, because a misjudgement costs one lease rather than a territory and a development schedule. And the hardest precondition disappears: at home the kitchen exists, the routes already run, the licences are held and the cold chain is proven — which is the single largest obstacle discussed in chapter 5. **A brand expanding at home is the easier version of this, not the marginal one.** It is also the faster one: no import, no customs, no local co-packer, no waiting for a partner.

The rest of this paper is about whether that works, what it requires, and what it leaves you holding afterwards.

---

## 2. The decision inside the decision

### Figure 2 — The one decision that determines the outcome

**Should the customer have to identify themselves to buy?**
Requiring sign-up costs you sales. Not requiring it costs you the customer. Three conditions decide which is the cheaper loss.

- **Condition 1 — Friction absorber.** Does anyone other than the shopper bear the cost of joining — an employer who has contracted the service, or a payment app they already open daily?
- **Condition 2 — Product pull.** Is the item wanted enough to finish signing up rather than walk away? A named restaurant's signature dish clears this. A generic soft drink does not.
- **Condition 3 — Destination.** Is there somewhere to send the member afterwards — stores, a shop, a subscription, a benefit scheme? If not, identity is a cost with no return.

| Documented case | Condition 1: Friction absorber | Condition 2: Product pull | Condition 3: Destination | What they chose | What it cost them |
|---|---|---|---|---|---|
| Japan, unmanned frozen shops | Does not hold | Holds | Does not hold | Cash box, no gate | Grew tenfold as a category [trade survey] and knows nothing about a single buyer |
| Japan, restaurant-owned frozen machines | Does not hold | Holds | Holds | Cash, no gate | The shop is ten metres away. ~10,000 operators [industry body and manufacturer statements] capture nothing |
| Daalchini, India | Holds | Holds | Holds | App required | Nothing. Reports 68% repeat customers [CEO statement, no published methodology] and a profitable year [filed accounts] |
| Felfel, Switzerland | Holds | Holds | Holds | Employer contracts | Persuaded 600 companies instead of 150,000 people [company-disclosed] |
| Genki Forest, China | Holds | Does not hold | Holds | Scan to open, join after | Capital on cabinets to win a following it could message |
| Farmer's Fridge, United States | Partly | Does not hold | Holds | Open, rewarded later | Pays roughly 9% of spend [derived from published programme terms] to attach identity after the sale |

**Three conditions hold: require it. None hold: do not. In between, the population in front of the machine decides.**

Before the market question comes a design question that most brands treat as a technical detail: **must the customer identify themselves in order to buy?**

Requiring sign-up costs sales. Not requiring it costs the customer. Three conditions decide which is the cheaper loss.

**Does anyone absorb the friction?** In Switzerland, Felfel placed roughly 800 fridges with about 600 corporate clients, serving around 150,000 people a day. The employer contracts the service and sets the subsidy; the employee pays the balance in-app. Felfel did not persuade 150,000 individuals to register — it persuaded 600 companies, and the individuals arrived enrolled. The alternative it displaces, a conventional staff canteen, is described as requiring investment on the order of CHF 1 million. *(Company-disclosed; press.)*

**Is the product wanted enough?** Japan's unmanned frozen shops grew from 131 stores nationally in 2020 to 1,282 in 2022, with about 60% opened in a single year; one brand went from roughly 19 shops to more than 400. A freezer, a cash box and a camera, often at around ¥1–2m for a small site — one-third to one-tenth the cost of a conventional restaurant. *(Trade survey; franchise sources.)* Product pull was overwhelming, so the operators asked nothing of the buyer and grew anyway.

**Is there somewhere to send the member?** India's Daalchini requires the app: no app, no purchase. It turned profitable in FY25 on a net profit of INR 2.5 crore, against a loss of INR 8.58 crore the year before, and reported FY26 revenue of INR 140 crore — double the prior year — with EBITDA rising 5.7 times to about INR 21 crore while the fleet grew 62% to roughly 5,500 machines. Over 95% of those machines are owned by franchise partners rather than the company, and its published repeat rate is 68%. *(Profit and loss figures reconcile to filed consolidated accounts; the revenue and EBITDA figures are stated by the CEO to a trade outlet and not yet filed. The repeat rate is a CEO statement with no published methodology.)* Mandatory identification did not suppress that business. It is that business.

Where all three hold, require identification. Where none hold, do not. In between, the population in front of the machine decides: a captive workplace repays a one-time sign-up across hundreds of visits, and an airport concourse never will. This is a site question, not a market question — the office tower in your own city and the office tower in a foreign one give the same answer. Farmer's Fridge, with more than 2,000 locations across roughly 20 cities in transient sites, leaves its machines open and buys identity instead through a loyalty scheme worth roughly 9% of spend. *(Company-disclosed; programme terms.)*

That is a rational answer to the same question — and it sets the price of the alternative, which is the subject of the next chapter.

---

## 3. Three futures for the same sale

### Figure 3 — The same sale, three futures

**Every machine sells something. Only one of these leaves you a customer.**
Day 0: a stranger walks up to your unit and buys your product for the first time. What you do in this moment decides everything after it.

**Future 1 — Sold anonymously.** Cash or a tap. Nothing is asked, nothing is kept.
- Day 0 — Buys, pays, walks away.
- Week 1 — You know a unit moved. Not who moved it.
- Month 3 — They may have come back. You cannot tell, and cannot ask.
- You end up with revenue: no second sale you can influence, and no evidence for the store decision you are trying to make.
- Evidence: roughly 10,000 restaurant-owned machines in Japan work this way. [Industry body and manufacturer statements.]

**Future 2 — Identified afterwards.** Open to all, then paid to come back and enrol.
- Day 0 — Buys at the screen with nothing asked of them.
- Day 0+ — Offered a standing discount to join. Some take it.
- Month 3 — A real member list, assembled at a permanent cost per sale.
- You end up with customers you rent: it works, and it is the right answer where people pass once. It is margin traded for data.
- Evidence: one US operator pays roughly 9% of spend for this. [Derived from published programme terms.]

**Future 3 — A member from the first purchase.** Opening the door is the payment. Buying is joining.
- Day 0 — One tap opens the unit, charges only what was taken, and the receipt arrives on your card.
- Week 1 — You can reach them — your products, your prices, your offers, in your own card.
- Month 3 — The next purchase can happen at the unit, in a store, or without either.
- You end up with customers you own: the unit stops being a sales channel and becomes the front door of the market you are entering.
- Evidence: 68% repeat customers where the purchase is the sign-up — one operator, company-disclosed. [CEO statement, no published methodology.]

**In a market where nobody knows you yet, the first purchase is the only introduction you get. Anonymous is the expensive option.**
*Outcomes as documented by operators in Japan, the United States and India. Timelines are illustrative.*

Here is the finding that should change how a franchisor thinks about this category.

Sanden's frozen vending machine has passed 10,000 cumulative units shipped in Japan, and roughly 90% of buyers are restaurants and food manufacturers selling their own products. *(Industry body and manufacturer statements.)* These are not technology companies. They are food businesses that decided an unattended box selling their own product was worth around ¥2m of capital — and the operator economics support them: published trade models put monthly operating profit at roughly ¥50,000 at 300 meals and ¥230,000 at 900, against fixed costs of about ¥40,000 a month. Operators have reported that about 20% of sales arrive after normal closing hours. *(Trade models and operator accounts — indicative, not audited.)*

Now the part that matters. Almost all of those machines take cash. The restaurant is often ten metres away. The operators satisfy the destination condition completely — and capture nothing, because nothing in the transaction records who bought.

That is not negligence. A vending stack is built to move stock and settle money; nothing in it knows who opened the door, and nothing downstream of it can reach that person afterwards. Ten thousand careful operators reached the same outcome because the equipment they could buy made no other outcome available.

**Ten thousand food businesses built a satellite storefront in front of their own shop, and threw away every customer who used it.**

Worth noticing where those machines are: beside the operator's own restaurant, in the operator's own city. Nothing about this loss requires a border. A brand that puts units in the next district over, on its existing route, loses the customer in exactly the same way unless it decides otherwise.

The same sale has three futures. Sold anonymously, it is revenue and nothing else: no second sale you can influence, and no evidence for the store decision you are trying to make. Identified afterwards, it becomes a member list assembled at a permanent cost per sale — which works, and is correct where people pass once. Or the purchase itself is the enrolment, and you end the transaction holding a customer rather than a receipt.

In a market where nobody knows you yet, the first purchase is the only introduction you get. Anonymous is the expensive option.

---

## 4. What that is worth

### Figure 4 — What the first purchase makes possible

**Each thing you know at the door opens a business you did not have.**
The unit sells the same product in all four columns. What changes is how much of the customer's next year you can reach.

| What the brand can actually do | Tier 1: Anonymous sale | Tier 2: Known buyer | Tier 3: Rewards & offers | Tier 4: Member commerce |
|---|---|---|---|---|
| Know that stock moved, and what sold where | Yes | Yes | Yes | Yes |
| Know who bought, and tell a first visit from a fifth | No | Yes | Yes | Yes |
| Prove the product sells in this country — named repeat buyers, not unit counts | No | Yes | Yes | Yes |
| Reach that customer again without buying the audience back from a platform | No | Partly | Yes | Yes |
| Give a reason to return that is not simply walking past again | No | No | Yes | Yes |
| Put the right price in front of the right person, without discounting everyone | No | No | Yes | Yes |
| Sell the range the cabinet cannot hold — gift sets, bulk packs, seasonal, higher-priced lines | No | No | No | Yes |
| Take the order when the unit is sold out, closed, or nowhere near them | No | No | No | Yes |
| Move a machine customer into a store, or a store customer into a repeat order | No | No | Partly | Yes |
| Carry the customer into the next market you open, before it opens | No | No | Partly | Yes |

**The four levers**
- **Frequency** — a returning customer is a decision you can influence, not weather you wait for.
- **Basket** — the catalogue is no longer limited by what fits behind one door.
- **Acquisition cost** — members arrive through sales you were already making, not through media you buy.
- **Measurement** — incremental or cannibalised stops being an argument and becomes a number.

Knowing who bought is not a data-collection exercise. Each tier of knowledge opens commercial activity that is simply unavailable below it.

**Tier 1, anonymous.** You know stock moved. You cannot tell a first visit from a fifth, so you cannot prove to a prospective master franchisee that a district has repeat demand — only that units left the machine.

**Tier 2, known buyer.** Repeat becomes measurable. The district case stops being a unit count and becomes named customers returning. For a franchisor recruiting a local partner, that is the difference between a projection and an argument — and for a brand taking a domestic site to its own board or its landlord, it is the same difference.

**Tier 3, rewards and offers.** You can give a reason to return that is not proximity, and put a price in front of one customer without discounting everyone. China's Genki Forest, locked out of convenience-store cooler space by incumbents with vastly larger fleets, deployed its own smart cabinets into offices and apartment buildings specifically so that buyers would flow into its own messaging channel. The brand reports outlets with a cooler selling roughly 118% above those without, and smart cabinets about 74% above conventional coolers. *(Brand-supplied, no methodology.)* Notably, its cabinet business reportedly reached full-year profitability in part by not expanding aggressively.

**Tier 4, member commerce.** The catalogue stops being limited to what fits behind one door. You can sell gift sets, bulk packs and higher-priced lines the cabinet cannot hold; take the order when the unit is sold out, closed or nowhere near the customer; and move a machine customer into a store. The general-retail evidence on this direction of travel is encouraging: a large industry study found that opening a physical location lifted online sales in the trade area by about 6.9% on average, and about 13.9% for emerging direct-to-consumer brands. *(Large industry study — general retail, not unattended.)*

There is a second-order prize. Once a brand's network exists, it becomes a channel other brands will pay to use: Starbucks' ready-to-drink line launched a new product through Genki's cabinets. *(Reported with a named executive.)*

The four levers underneath all of this are plain: **frequency** becomes a decision you can influence rather than weather you wait for; **basket** is no longer capped by the cabinet; **acquisition cost** falls because members arrive through sales you were already making; and **measurement** turns incremental-versus-cannibalised from an argument into a number.

**And none of this stops being useful once the stores open.** It is tempting to read unattended units as scaffolding — something you put up to get a market started and take down once the real estate arrives. The opposite is closer to true. A store reports monthly and tells you what a catchment bought in aggregate. A network of units reports continuously, by district, by SKU, by daypart, from named customers whose repeat behaviour you can watch. That makes it the cheapest place in the business to test an assortment change or a price, and the earliest place a shift in demand shows up — months before it reaches a store P&L. Fit is not something a brand establishes once on entry and then owns. It moves, and the units are how you keep watching it.

---

## 5. What has to be true

### Figure 5 — The product decision, before the market decision

**Frozen buys you room to be wrong. Chilled buys you frequency.**

| | Frozen & packaged — where a central kitchen should start | Chilled ready-to-eat — earned later, in dense workplaces |
|---|---|---|
| Waste | Near zero — one to three months of shelf life | Low double digits is common [industry characterisation; no operator publishes this figure], against 4–10% in restaurant foodservice [industry sources] |
| Replenishment | Weekly or less — one route can carry many units | Daily — plus a markdown routine, designed before launch |
| Licensing burden | Lighter — pre-packaged, sealed at the kitchen | Heavier — fresh handling rules, market by market |
| Basket | Larger — customers buy to stock up | Smaller — one meal at a time |
| Visit frequency | Lower — a planned purchase, not an impulse | Higher — a daily habit, which is what builds members |
| Cost of a bad site | Small — stock keeps; move the unit | Daily — unsold stock is written off every day it is wrong |

**Enter a market frozen. Add chilled only where the same people walk past every day.**
*Waste benchmarks: industry sources for unattended fresh and restaurant foodservice. Shelf life: Japanese operator practice.*

### Figure 6 — Where this business is actually won or lost

**Twenty units, two outcomes.**
The same capital, the same machines, the same product. The difference is whether one person can restock them in one run.

**Coverage — every unit is its own trip.** Twenty units spread across a country, because every site looked good on its own. Replenishment, site work and warehousing consume close to half of revenue at the large operators [filed prospectus, 2026 Hong Kong listing]. Spread thinly, the route cost eats the margin before the product does.

**Density — one run serves them all.** Twenty units clustered on one route that already runs — yours, a partner's or a 3PL's. At scale, one person covers on the order of 120 units [filed prospectus, 2026 Hong Kong listing]. Attached to logistics that already run, the marginal unit costs minutes rather than a journey. A dedicated van and driver for fifteen units is the version that does not work.

Four preconditions decide whether any of the above survives contact with operations. None is about technology, and a brand that gets one wrong will conclude the category does not work when what failed was the setup.

**Somebody must already be able to produce and move the product.** This is the first question to settle and the one most likely to stop a programme before it starts. Expanding at home, it is already settled — your kitchen, your routes, your licences — which is why the domestic version of this can start next month. Crossing a border, there are three honest answers, and only three. **You export your own packaged product** from an existing central kitchen into cold storage in the target market — the cheapest path, viable for frozen and long-shelf-life ranges, and the reason this paper argues frozen first. **You buy from a local co-manufacturer**, which works when your range fits something they already make, and becomes expensive when it requires bespoke formulation and compliant packaging for a short run. **Or your partner already has kitchens and cold chain in that market** — the case where this moves fastest, and the one worth seeking. What does not work is building a commissary to serve fifteen units: that signs the lease this paper argues against, in a smaller form. If none of the three applies, the honest answer is that unattended retail is not your entry instrument for that market, and the alternatives are covered at the end of this chapter.

**Frozen buys you room to be wrong. Chilled buys you frequency.** Frozen and packaged forgives almost everything: one to three months of shelf life means near-zero waste, weekly rather than daily replenishment, a lighter licensing burden, and a larger basket because customers stock up. Chilled ready-to-eat reverses all four and returns a daily habit instead — which is the better business once a location is proven, and an expensive way to learn before it is. Unattended fresh is commonly described as running low-double-digit waste against 4–10% in restaurant foodservice, and a peer-reviewed grocery study found automated markdown cut waste by about 21% while improving gross margin about 3%. *(Industry sources; peer-reviewed, adjacent category.)* Enter a market frozen. Add chilled where the same people walk past every day.

**The outcome is decided by the route you pick, not the machine you buy.** Replenishment is the largest cost in this business — close to half of revenue at the largest operator to have filed public accounts, where one person covers on the order of 120 units. *(Filed prospectus, 2026 Hong Kong listing.)* That ratio is the whole argument for density: twenty units scattered across a country are serviced twenty times, twenty on one route are serviced once. It is also why a dedicated van and a full-time driver for fifteen units will not work — the route has to attach to logistics that already run, whether yours, a partner's, or a 3PL's existing drops. When density is achieved, the effect shows up in the accounts: Daalchini's fleet grew 62% in FY26 while reported EBITDA rose more than fivefold, which is what operating leverage looks like in this business. Density also explains the sixteen-fold spread between weak and strong sites in published operator models, and why no honest operator quotes a single payback figure.

**Channel rights must be settled before the first unit ships.** Unattended retail is a channel, and in most franchise agreements it is either unaddressed or reserved in language written before this format existed. Decide explicitly: does the territory grant cover unattended units, does the master franchisee have the right to operate them, and what happens to units the franchisor placed before the partner signed. An institutional partner will ask on the first call, and the answer *we hadn't considered it* is worse than either answer on its own. The simplest arrangement gives the incumbent partner first right to operate units in their territory, and transfers any the franchisor placed earlier on agreed terms.

Two further judgements belong to the brand rather than to a rule. **Positioning:** a branded fridge is the native format for a packaged range, and it is a poor first impression for a concept whose value is hospitality and theatre — a brand whose export product is a chilled bowl loses nothing; a brand whose promise is a dining room should lead with something else and let units follow. **The gate:** require membership where the population returns, leave it open where it does not, and run both in the same pilot so the conversion difference settles the argument rather than the opinion of whoever is loudest in the room.

**And if none of this fits**, the established alternatives are real and worth saying plainly: a food-hall or department-store concession, a residency in an existing licensed kitchen, or a delivery-only trial with a local operator. Each tests hot food and full menu in a way unattended units cannot. Each also ends with the customers belonging to the landlord, the host or the platform — which is the trade being made, and the subject of chapters 3 and 4.

---

## 6. Where we fit, and the first move

Franchising's own trade literature now describes non-traditional formats — kiosks, carts, compact units — as vehicles for testing new markets and piloting before wider rollout, chosen for flexible investment and fast setup. Brands have built this rung deliberately: kiosk formats at under US$90,000 and a few hundred square feet, express licensing concepts aimed at convenience and travel venues, campus express units used as testing grounds. *(Trade press and company announcements.)* The instrument is established. What has been missing is the part that makes it compound.

**Perkd makes the purchase and the membership the same event.** The customer taps a membership card to open the door; that tap is the payment authorisation. Items are reconciled after the door closes and only what was taken is charged. Whatever discount the customer qualifies for applies automatically, with no code. The receipt lands in that brand's own card.

**And the member then has somewhere to go.** A message from the brand arrives inside its card carrying real products, real prices, the customer's own qualifying offers, a basket and a checkout — running on the catalogue the brand already maintains. Local payment methods are supported market by market where integrated, and each unit settles to the party that owns it, so units can be funded by the brand, a local entity, or individual franchisees without becoming an accounting problem.

This combination is harder to assemble than it looks, which is why the ten thousand operators in chapter 3 did not. A machine vendor would need a consumer wallet and a payments stack to reach it; a loyalty platform would need authorisation-then-capture with items reconciled after the door closes. The two halves sit on opposite sides of the industry.

The arithmetic is in the paper already. Farmer's Fridge pays roughly 9% of spend, on every transaction, for a member list assembled after the sale. Enrolment at the door is paid once, in seconds, by a customer who was buying anyway. Whether that trade favours you depends on the site — which is the whole of chapter 2 — but it is a calculation a brand can run on its own numbers before it speaks to anyone.

We are direct about our position: we have not yet published operating results from a live network of our own. Every figure in this paper is somebody else's, marked as such. What we offer is a capability and a framework, and the right way to test both is small.

### The first move

**Ten to twenty units. One route that already runs. Two or three types of location. Membership required at some sites and open at others. One quarter.**

This is not a substitute for a master franchisee, and it is strongest when it is not run alone. A partner who already holds kitchens and cold chain in the market can stand this up in weeks, which makes unattended units the first channel of the partnership rather than a rehearsal for it — the partner earns from day one, the brand learns from day one, and the store programme that follows is sized against real local trading rather than a negotiated promise. Where no partner is in place yet, a brand exporting packaged product can run the same quarter itself and arrive at that conversation holding something no competitor brings to it. At home, where the kitchen and the routes already exist, there is no partner to wait for at all — the quarter can start on your own logistics next month.

Either way the quarter ends with three things rather than one: evidence of what sells where, to whom and at what price; a base of members who belong to the brand before a single store opens; and a local trading record — the absence of which, on the evidence in chapter 1, is where most expansions quietly come apart, abroad and at home. And the units do not come down when the stores go up. They become the brand's closest read on the market it just entered.

The alternative is the sequence most brands still follow. Commit the schedule, sign the leases, and find out afterwards.

*To discuss a pilot, contact [merchant@perkd.me](mailto:merchant@perkd.me).*

---

## What the evidence does not support

**None of the six cases in this paper is an international franchisor probing a new market with unattended units.** They are workplace catering, a domestic vending platform, a beverage brand's own cabinets, restaurants selling beside their own shops, a low-capital franchise category, and a venture-funded fresh brand. Taken together they evidence two things well — that the format works at scale for food businesses, and that capturing identity at the point of sale changes the customer economics. They do not evidence the entry sequence itself, which is reasoning built on top of them. We would rather say so than let the cases carry more weight than they can.

**Cannibalisation of a brand's own stores by unattended units has not been measured in any published study we could find.** The closest proxy is store-on-store: an analysis of the 2008 Starbucks closures put the average cannibalisation from one neighbouring outlet at about 1.2% within a mile and 0.4% at one to three miles — dated, drawn from a single event, and not a benchmark. Measure it in your own pilot; do not accept an estimate from anyone, including us.

**No unattended operator publishes a chilled-food waste percentage.** The low-double-digit figure is an industry characterisation, not a disclosure. Treat it as a planning assumption to be replaced by your own number in the first quarter.

**The franchise-rung argument is evidenced for kiosks and compact formats, not specifically for fridges.** The reasoning transfers; the data does not yet.

---

## Notes on sources

**Filed or peer-reviewed:** the Cornell study of 142 international master franchise contracts; the Chinese operator's Hong Kong listing prospectus; the grocery markdown study. **Survey or analyst study:** the 360-agreement master franchising analysis; the master franchisee problems survey; Japanese unmanned shop counts. **Filed accounts:** Daalchini's profit and loss figures. **Company-disclosed:** Daalchini's FY26 revenue, EBITDA, fleet count and repeat rate; Felfel's fleet and client figures; Farmer's Fridge's footprint and loyalty terms; Genki Forest's sales-lift figures. **Trade, vendor or press:** machine costs, operator profit models, after-hours sales share, market-share characterisations, waste ranges. Figures are in original currency.

---

## Annexes — the six cases in full

Ordered by weight of evidence, not by size or fame.

- **A.** Daalchini, India — identity on every sale, at the cost of the casual buyer
- **B.** Japan's restaurant-owned frozen vending machines — after-hours revenue, and not one customer kept
- **C.** Felfel, Switzerland — frictionless enrolment, bought by contracting with the employer
- **D.** Farmer's Fridge, United States — every passing sale, and about 9% of spend to attach identity later
- **E.** Japan's unmanned frozen shops — the cheapest franchise rung, with no customer knowledge and no moat
- **F.** Genki Forest, China — an uncontested channel, where capture happens after the sale if at all

*Each annex opens with a fact strip, sets out both halves of the trade that operator made, and closes with a source-confidence table and source list separating filed accounts from company statements and press reporting.*