MenuPass — AI multilingual QR menu for Japan tourist restaurants

1. One-liner

AI turns a photo of a Japanese paper menu into a 5-language QR menu with allergen and halal flags.

2. Trend signal — why now?

Three forces collided in the last 12 months and Japan’s small restaurants are stuck between them.

Inbound tourism hit an all-time record. Japan logged 39.06M international visitors Jan–Nov 2025, smashing the 36.87M record set in 2024. February 2026 alone: 3.47M arrivals, +6.4% YoY. South Korea, Taiwan, China and an increasingly halal-observant SE Asia/Middle East crowd are the fastest-growing segments. The yen is weak. Tourists keep coming.

Operators are explicitly saying language is the wall. A 2025 industry survey found 60.2% of Japanese restaurant operators cited language barrier as the top reason they don’t want more inbound guests, and 29% said multilingual menus are too hard to make. That’s a self-reported demand signal in plain English: they don’t want to lose tourists, but they cannot serve them.

The labor escape hatch just slammed shut. On 2026-03-27 the Immigration Services Agency announced the SSW Type-1 quota for food service — 50,000 over five years — would be exhausted by May 2026. New foreign restaurant worker visas frozen for the rest of FY2026. Operators who were going to “hire someone who speaks English” can no longer hire that person. Officials told operators publicly to “use technology and automation.”

At the same time, the money is sitting on the table. Tokyo’s “Inbound Response Capability Enhancement Subsidy” pays up to ¥3M per store, covering 1/2 to 2/3 of multilingual menu, signage, tablet and QR install costs. Osaka, Kyoto and Fukuoka run analogous programs. To qualify, the store must be listed on EAT Tokyo — a clunky government-built menu site that sets the floor for “what’s available” today. Operators want to spend the subsidy; nobody is making it easy.

And cheap multimodal AI just made the build trivial. A photo of a hand-written paper menu (think Showa-era izakaya in Asakusa or Shinsekai) → structured items, prices, allergen guesses, halal/vegan flags, 5-language translations including dish-explanation text — all under $0.05 per menu in late-2025 inference costs. Two years ago this was a research project. Today it’s an API call.

Provenance:

• Signal 1: 60.2% of Japanese restaurant operators cite language barrier as top reason for not wanting more inbound tourists; 29% say multilingual menus are too hard. — https://restauranttechhub.com/how-restaurants-are-using-ai-tools-in-2026-to-improve-operations/ — 2026-04-30
• Signal 2: Japan suspended new SSW food-service visa applications on 2026-03-27 as the 50,000 cap nears exhaustion by May 2026; officials tell operators to use technology. — https://japantoday.com/category/national/japan-suspends-new-visas-for-foreign-restaurant-workers-as-quota-nears — 2026-04-30
• Signal 3: Tokyo’s “Inbound Response Capability Enhancement Subsidy” pays up to ¥3M per store (1/2 of eligible expenses) for multilingual menus, tablets and QR install — store must be on EAT Tokyo. — https://www.shigyo.co.jp/en/search_post/business-subsidy/tokyoinboundgrant/ — 2026-04-30 Category: Geographic arbitrage

3. The opportunity

Japan has 1.41M restaurants. ~74% are independent, mostly mom-and-pop. The tourist-facing slice — Tokyo, Osaka, Kyoto, Fukuoka, Hiroshima, Sapporo plus train-station and onsen-town clusters — is roughly 200k–300k stores. Of those, the chain operators (Torikizoku with 637 outlets, Saizeriya, family-resto giants) already have multilingual tablet POS. The independents do not.

The independents have one of three things on the wall today:

1. A Japanese-only paper menu — the most common case
2. A clunky tablet running an old Toast/Square knock-off in Japanese
3. A printed English menu somebody at city hall translated badly in 2018

What they want: tourists scan a QR, see the menu in their language with photos, see clearly whether the item contains pork, alcohol, gluten, dairy, shellfish, peanuts; can ask “what’s in tsukune?” and get a real answer; can place the order without flagging down the one waiter who knows ten English words.

What they get from incumbents: Oshinagaki and EAT Tokyo are unstyled directories that turn the menu into a static PDF in three languages and call it done. Trangle (Google Play) is a tourist-side phone app, not an operator product. Slang AI / Loman / Certus do voice phone answering — wrong product for in-store. POS giants (Square, Recruit AirREGI) have multilingual modes but their setup process assumes the operator can type the entire menu in Japanese first, which 70-year-old izakaya owners will not do.

Wedge: photo-the-paper-menu → 5-language QR menu live in 10 minutes, with allergen and halal/vegan auto-tags the operator can correct in 2 taps. Subsidy-fundable. POS-agnostic. Voice-explanation chat for tourists who don’t know what natto is. Sold to the operator at a price the Tokyo subsidy mostly covers.

Pork-in-ramen is the killer feature. A halal Indonesian tourist looking at “shoyu ramen” sees a clear ❌ pork-stock flag. The izakaya owner did not have to know to add that flag — the model knew. That’s the AI doing real work.

4. Target market

• Primary customer: 1–3 staff independent izakaya, ramen shop, sushi-ya, teishoku-ya, café in Tokyo / Osaka / Kyoto / Fukuoka tourist wards (Asakusa, Shinjuku, Dōtonbori, Gion, Hakata, etc.). Owner is 45–70 years old, Japanese-only, runs the place themselves.
• Why they buy: Tourists keep ordering wrong items, asking confusing questions, leaving low Tabelog reviews because the experience felt awkward. Owner can no longer hire an English-speaking part-timer (visa freeze). Subsidy money is being left unspent because the application looks like work.
• Rough TAM reasoning: 200k–300k tourist-area independents in Japan. Convert 3% in 24 months = 6,000–9,000 stores. At ¥4,980/mo per store ≈ ¥360M–¥540M ARR (≈ $2.4M–$3.6M USD).
• Why now for them: Record tourist volume + foreign-worker visa freeze + active multi-prefecture subsidy programs all hitting the same fiscal year. The window is the FY2026 subsidy cycle.

5. Product sketch (MVP)

• Photo-to-menu in 10 minutes. Operator photographs the paper or whiteboard menu with a phone. AI extracts every dish, price, and category. Operator confirms in Japanese.
• 5-language QR menu in English, Mandarin (Simplified), Korean, Traditional Chinese, and Indonesian/Malay (halal-conscious markets). Each dish gets a short tourist-friendly description, not a literal translation.
• Auto allergen + dietary flags. Pork (incl. hidden in broth/dashi), shellfish, dairy, egg, wheat/gluten, peanut, alcohol; halal-friendly / vegan / vegetarian / pescatarian icons. Operator can override.
• “What’s in this?” chat. Tourist taps a dish, asks any question in their language, gets answer pulled from the menu’s structured data. No model hallucination — it answers from confirmed ingredients only.
• Order from phone, ticket prints in Japanese to the kitchen. Optional. Works without POS integration via thermal printer or bridges to Square / AirREGI.
• Price-parity guarantee. Same menu, same prices, no two-tier pricing — explicit anti-Osaka-ramen-shop-incident feature, used as a trust marker for tourists.
• Auto-listing on EAT Tokyo, GuruNavi-EN, byFood and Tabelog-EN so the subsidy paperwork is one click.
• Subsidy concierge. App pre-fills the Tokyo “Inbound Response Subsidy” application PDF with the operator’s data; we partner with one Tokyo gyosei-shoshi (administrative scrivener) to file it for ¥30k flat.

6. AI angle — what’s load-bearing

AI is the entire product, not a sticker.

1. Multimodal extraction from messy paper menus — vertical script, hand-written calligraphy, smudged thermal printouts, mixed kanji+katakana — into structured JSON. Three years ago this required a contractor. Today GPT-4o / Claude / Gemini do it for cents.
2. Allergen inference — the model knows that tonkatsu, tonkotsu, shoyu ramen with pork stock, miso soup with dashi, and takoyaki sauce all contain animal products an Indonesian Muslim tourist needs to avoid, even when the operator did not list them. This is the value proposition. A naive translator gets it wrong; a good model gets it right and lets the operator confirm.
3. Tourist-facing chat — grounded on the structured menu only, so it cannot hallucinate “yes this contains chicken” when the dish is actually pork. Voice mode for accessibility.
4. Cross-language consistency — same dish description tone across English / Chinese / Korean. Important for trust.

Strip out the AI and the product collapses into “make a PDF” — which is what EAT Tokyo already is, and nobody uses it.

7. Localization angle

Japan-first by construction. The localization is the wedge.

• Language pairs: Japanese ↔ EN/ZH-CN/ZH-TW/KO/ID. Spanish, French, Thai later.
• Payment rails: PayPay, Suica/IC, Rakuten Pay tourist-side; Stripe/JCB on the operator side for the SaaS sub.
• Distribution: EAT Tokyo registry, prefecture commerce associations, Japan Restaurant Association (JRA), JNTO partner programs.
• Compliance hooks: The product registers the store on EAT Tokyo automatically, which is the gating requirement for the Tokyo ¥3M subsidy. That alone is worth more than any feature.
• Cultural quirk built in: “no two-tier pricing” feature, vendor-of-record contracts in Japanese, Japanese-only operator UI, English/CN/KR/ID tourist UI. Operator never sees a foreign character.

8. Business model — path to $1M–$5M ARR

• Pricing: ¥4,980/mo per store base ($33/mo). ¥9,980/mo with order-routing + thermal printer hardware + subsidy concierge. One-time setup ¥19,800 (covered by subsidy).
• ACV: ¥60k–¥120k per year per store, average ¥75k ($500/yr).
• Path to $1M ARR: 2,000 stores × ¥75k = ¥150M ≈ $1M. ~1% of tourist-facing independents.
• Path to $5M ARR: 6,000 stores in Japan + light Korea/Taiwan expansion of the same product (mirror flow: paper menu → JP/EN/KR/ZH for inbound tourists in Seoul/Taipei). Korea has the same pain — 8M Japanese tourists/yr — and no subsidy gates.
• Expansion path: orders → POS hooks → review-management add-on (Tabelog, Google) → ad-spend on tourist platforms → halal/vegan certification marketplace. Each adds ¥2-5k/mo seat upsell.

9. Go-to-market wedge — first 100 customers

Concrete. Subsidy is the channel.

1. Cold-walk Asakusa, Ueno, and Shibuya yokocho alleys with a tablet. Three founders, two days, pitch each shop in person: “I will set up your menu in 5 languages while you watch, the Tokyo government will pay for half, and your first month is free.” The Showa-generation owner in a 10-seat izakaya does not buy on email — they buy on a handshake and a demo. Tokyo has ~20k of these owners within walking distance of a JR station. Aim for 20 paying first month from this sprint.
2. Three Tokyo gyosei-shoshi (administrative scrivener) firms that already file subsidy applications for restaurants. Pay them a ¥10k bounty per signed customer; they bundle MenuPass into every Tokyo Inbound Subsidy application they touch. Each firm does 30–80 subsidy applications per year. 50 customers in 90 days from this channel alone.
3. JNTO + Tokyo Convention Visitor Bureau partner listing. EAT Tokyo signups already require a multilingual menu — push for a “MenuPass-powered” badge on the EAT Tokyo directory. Government press release does our marketing for free.
4. Halal/vegan tourist creator partnerships. 5 mid-tier YouTubers/Instagrammers in the halal-Japan and vegan-Japan niches (Halal Navi, Eat Halal Tokyo, vegan Kyoto creators each have 50k–500k followers). Pay them to feature 3 MenuPass-equipped restaurants; restaurants get free month, creators drive the demand pull that makes operators answer the cold knock.
5. Tabelog operator-side ads. Tabelog runs paid placement for restaurants. We buy slots in Asakusa/Shinjuku/Dotonbori targeting “menu translation” / “多言語メニュー” search intent; cheap because nobody bids in 2026.

10. Build complexity — justification

Low to Medium. Two people, 12 weeks to pilot v1. Off-the-shelf multimodal model API, Next.js or SvelteKit web app, Cloudflare-hosted QR menu pages, Stripe/JCB billing, LINE for operator support (Japanese SMB lifeline), thermal printer SDK on the order add-on. The hardest two things are (a) the allergen-flag prompt engineering for Japanese cuisine — solvable with one Japanese-speaking food consultant and a 200-dish ground-truth dataset, ~3 weeks of work; and (b) the EAT Tokyo + subsidy-application PDF integration — pure ops + one gyosei-shoshi partnership. Neither is research; both are work.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Restaurant SaaS, no licensing needed; Tokyo subsidy participation is operator-side.
Ethical — no harm / dark patterns	✅	Anti-two-tier-pricing feature is explicitly anti-discrimination.
Market exists (evidence above)	✅	60.2% operators self-report the pain; ¥3M/store subsidy is live.
1–5 person team can build this	✅	2 builders + 1 Japan-side ops, 12 weeks to v1.
Launchable with <$50K / ₹40L	✅	API costs + Tokyo SIM + ¥200k of cold-walk leg work.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	60.2% operators self-cite this as their top inbound problem; visa freeze raises urgency. Not literal “hair on fire” daily but felt every shift; subsidy makes it actionable now.
Demand evidence	15	13/15	Govt subsidies funded; multiple incumbents with weak product; explicit operator survey numbers. Skeptic nods.
Build feasibility	15	13/15	Pair ships pilot in 8–10 weeks. Allergen prompt-eng + thermal printer SDK are the only soft spots.
Distribution clarity	15	12/15	Subsidy-paperwork channel via gyosei-shoshi is novel and high-conviction; cold-walk works in Tokyo dense wards; English not required for sales. Risk: requires Japan-on-the-ground operator.
Revenue mechanics	15	12/15	¥4,980/mo subsidy-funded is a no-brainer ACV; path to ~$3M ARR Japan-only is realistic. Ceiling without Korea/Taiwan expansion limits to ~$5M without geographic stretch.
Time to first revenue	10	8/10	First paying store in 4–6 weeks of cold-walk; revenue at ¥4,980/mo from week 1 of pilot.
Defensibility	10	6/10	Allergen ground-truth dataset + EAT-Tokyo + gyosei-shoshi partnerships compound; no patent, but the trust + subsidy relationships are not 1-week-clonable.
Total	100	81/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (multimodal + allergen prompt eng) · operations-heavy (Japan ground game, gyosei-shoshi, cold-walks)

Best founder shape: one technical builder + one Japan-resident or fluent Japanese-speaking ops/sales lead. The ops person is non-negotiable. A pure remote US/India team will lose this.

Key assumptions to validate (3–5)

1. Assumption: Tokyo izakaya owners will pay ¥4,980/mo (subsidy-funded) for a multilingual QR menu. How to test: 30 in-person interviews + LOI form across Asakusa/Ueno/Ebisu in week 1; target ≥10 signed LOIs at price.
2. Assumption: Multimodal AI hits ≥95% accuracy on dish + price extraction from real Japanese paper/whiteboard menus, and ≥90% on allergen flags before operator review. How to test: Build the extraction pipeline against 50 real menus collected from Asakusa shops; have a Japanese food consultant grade output.
3. Assumption: A Tokyo gyosei-shoshi will agree to bundle MenuPass into Inbound Subsidy applications for ¥10k per signed store. How to test: Sign one paper partnership in week 2; commitment to 5 referred stores in month 1 is the kill criterion.
4. Assumption: Tourists actually use the QR menu (not just install-and-die). How to test: Pilot 5 stores 4 weeks; track scan-to-order rate. Target ≥35% of foreign guests scan, ≥20% complete a question in chat.
5. Assumption: The Tokyo subsidy will keep flowing through FY2026 and FY2027 (the whole pricing model bends if cut). How to test: Confirm with TCVB liaison + read FY2026 budget allocation; have a backup ¥2,980/mo unsubsidized tier ready.

Risk flags

1. Platform dependency: EAT Tokyo / Tokyo subsidy bureaucracy. If Tokyo cuts the program in FY2027, ACV needs to drop or the value prop softens. Mitigation: Osaka/Kyoto/Fukuoka programs as fallbacks; unsubsidized tier.
2. Operator demographics: Many target owners are 60–75 years old, do not use email, and are LINE-only. Onboarding has to be 100% LINE-driven and Japanese-only on the operator side. Slow ops execution kills this.
3. Incumbent wakes up: Recruit’s AirREGI and Square Japan could ship a “multilingual mode” any quarter. Wedge: speed + subsidy-paperwork ownership + allergen depth, none of which is in their roadmap today.
4. Tourism slowdown: A yen reversal or geopolitical event could cut inbound 20–30% in a quarter. Demand is volume-elastic; pricing tier may need to flex.
5. Pork-in-broth liability: If MenuPass mis-flags a halal-friendly dish that actually contains pork dashi, a tourist eats it, and a viral post follows, brand damage. Mitigation: explicit operator-confirmation step on every allergen flag, conservative defaults, ¥0 marketing claim of “guaranteed halal” — only “operator-verified ingredients.”

14. Structured verdict

Score:                  81/100
Verdict:                STRONG GO
Confidence:             High
Best-fit builder:       1 multimodal-AI engineer + 1 Japan-fluent ops/sales lead, ideally Tokyo-resident; food-domain advisor on retainer
Time to revenue:        4–6 weeks from cold-walk start
Capital to launch:      ¥3–5M / $20–35K (API credits, Tokyo SIM/scooter, gyosei-shoshi retainer, 50-menu ground-truth dataset)
Top 3 assumptions to validate first:
  1. Tokyo izakaya owners will sign LOI at ¥4,980/mo subsidy-funded (30 cold-walk interviews, target 10 LOIs, week 1)
  2. Multimodal AI hits ≥95% extraction + ≥90% allergen accuracy on real paper menus (50-menu pilot dataset, week 2)
  3. One Tokyo gyosei-shoshi commits to 5 referred stores in month 1 (paper partnership, week 2)
Kill criteria:
  - Abandon if <5 of 30 cold-walked stores express interest at price
  - Abandon if multimodal extraction accuracy <85% on real menus after 3 prompt iterations
  - Abandon if Tokyo cancels the Inbound Subsidy in the FY2027 budget without an Osaka/Kyoto equivalent stepping up

15. Next step — 1-week validation sprint

• Day 1: Build the photo-to-multilingual-menu pipeline against 10 real menu photos collected from Asakusa/Ueno (publicly visible). Stand up a single QR-menu page template. Translate into EN/ZH/KO/ID.
• Day 2: Recruit a Japanese food consultant for 1 day to grade allergen-flag accuracy on 50 dishes. Iterate the prompt twice.
• Day 3: Cold-walk 30 izakayas in Asakusa and Ueno in a single afternoon. Show the live demo on iPad. Pitch ¥4,980/mo, first month free, “Tokyo subsidy will cover most of next year.” Collect LOI signatures or rejections, on paper.
• Day 4: Meet 2 gyosei-shoshi firms in Shinjuku. Pitch the ¥10k bounty bundling. Ask for a one-page MoU.
• Day 5: Decide go/no-go on these measurable outcomes:
  • GO if ≥10 signed LOIs from Day 3 AND ≥1 gyosei-shoshi MoU AND extraction-pipeline accuracy ≥90% across the 50-dish test set.
  • VALIDATE-MORE if 5–9 LOIs OR accuracy 80–90% — extend sprint by 1 week, retest with prompt fixes.
  • NO-GO if <5 LOIs OR no gyosei-shoshi committed OR accuracy <80% — the channel or the model isn’t there.

Falsifiable. Cheap. One Tokyo trip with a tablet and a translator does it.