Gugan

1. One-liner

Tells Korean restaurant owners which delivery-fee bracket they are sliding into while they can still change it.

2. Trend signal — why now?

Korea’s two dominant delivery apps price their commission by percentile rank against every other merchant on the platform, not by an absolute revenue number. That single design choice created a measurement window nobody sells tooling for.

배달의민족 (Baemin, ~54–60% share) charges 중개수수료 of 7.8% to the top 35% of merchants by revenue, 6.8% to the 35–80% band, and 2.0% to the bottom 20%. The band is computed from the trailing three months of daily-average delivery revenue, fixed during the following month, then billed for the next three months (세무사신문, 2025-01-22; 경향신문, 2025-03-03). February–April behaviour sets the June–August bill. The owner is told the verdict only after the window has closed.

쿠팡이츠 runs the same ladder on a different clock: it bills everyone at the top 7.8% rate first, then computes the band from actual monthly revenue and refunds the difference within five business days of the following month (뉴시스, 2025-02-18). Every Coupang Eats merchant below the top band is systematically overcharged and then made whole by a rebate they have no independent way to check.

The band also decides whether a whole order channel is free or taxed. Packaging orders stopped being free on Baemin in April 2025, and Coupang Eats completed the sweep in April 2026 at 6.8% on 포장 (pickup) orders — while waiving it entirely for traditional-market stores and for bottom-20%-band merchants, a waiver just extended through March 2027 (뉴스핌, 2026-02-26; 스타뉴스, 2026-02-26). So the band is not merely a rate spread; for a Coupang Eats merchant near the bottom line it is the difference between free pickup orders and 6.8% ones.

And the money involved is now documented at the national level. KBIZ (중소기업중앙회) surveyed 1,250 platform merchants between 18 June and 3 July 2026 and published on 15 September 2026: delivery-app merchants hand over 26.2% of revenue in total transaction cost, up 5.2 percentage points in a single year, peaking at 45% for individual stores. Commission is 42.0% of that load, delivery fees 29.9%, advertising 15.6%, payment processing 12.4%. Asked whether the tiered pricing helped, 49.4% said nothing changed and 35.4% said it did not help — 84.8% combined (supple.kr, 2026-09-15; 시대, 2026-09-15).

Owners do not accept their assigned band, and they say so on the record. The complaint is always the same: the percentile denominator is invisible and looks padded with dead stores.

Provenance:

• Signal 1 (demand): KBIZ survey of 1,250 merchants — delivery-app cost 26.2% of revenue, +5.2pp YoY, max 45%; 84.8% say tiered pricing did not help — https://supple.kr/news/cmu24ah9w00715eeru4sqc8cs — 2026-09-15
• Signal 2 (feasibility): Baemin built a merchant-consent login channel and a third party is live on it — Cashnote’s notice states “2026년 1월 1일부터 배달의민족 ‘ID/PW 방식 연결’ 지원이 중단될 예정입니다”, with existing users converted to 배민비즈회원 consent login — https://cashnote.kr/notice/ — effective 2026-01-01
• Signal 3 (economic): DailyPay took pre-Series A funding in Jan 2026, grew Q1 revenue 529.8% YoY while profitable, and launched Baemin/Coupang Eats advance settlement in Aug 2026; separately the KFTC rejected ₩360B in settlement offers, leaving fines of ₩239–510B pending against Baemin — https://wowtale.net/2026/08/27/263536/ — 2026-08-27 Category: Regulatory arbitrage

3. The opportunity

The platform publishes the rule but not the number — and says outright that it will not. Reporting the band mechanics, 경향신문 states flatly: “다만 영업비밀이라는 이유로 매출액 구간은 공개하지 않았다” — Baemin does not disclose the revenue cutoffs, citing trade secret (경향신문, 2025-03-03).

That is a permanent, stated refusal, not an oversight awaiting a feature release. Baemin says “top 35%.” It never says “top 35% means ₩X of daily delivery revenue this quarter.” A merchant cannot compute their own band because the denominator is every other merchant’s revenue — data only the platform holds, and has declared confidential.

So owners are left arguing from anecdote, and the anecdotes are remarkably consistent — every one of them is a small store discovering it has been ranked as a big one:

• “‘배민1’(배민 직접 배달)로 하루 8만원 파는 매장인데 수수료율이 상위 35%다. 이게 상생인가요?” — a store selling ₩80,000 a day, rated top 35%. (경향신문, 2025-03-03)
• “홀 위주로 운영해서 하루 배달은 10건 될까 말까인데 상위 35%라고 한다” — mostly dine-in, barely ten delivery orders a day, still top 35%. (same)
• “최근 세 달 총 매출이 70만원인데 2구간(매출액 상위 35~50%)이다” — ₩700,000 of revenue across three months, placed in band 2. (same)
• “하루 30만원도 못 파는 날이 허다해 그간 매출 규모 하위 10%에 들어가기 만을 간곡히 바라고 있었는데 구간을 확인하니 상위 35%에 포함돼 이해할 수 없는 상황” — a pizza owner who had been hoping to land in the bottom 10%, then checked and found himself top 35%. (아주경제, 2025-02-21)
• “매출이 얼마 안되는데도 상위 35%에 포함된 점을 이해할 수 없다. 이같은 상황에서는 가게를 접거나 메뉴 가격 인상을 고려할 수밖에 없다” — a 분식집 owner: close the shop or raise prices. (same)
• “한 치킨집의 매출이 3개월간 800만원 가량이었는데, 상위 35%가 적용됐다. 하루 매출 9만원만 넘어도 상위 35%가 되는 꼴” — 김준형, 공정한플랫폼을위한사장협회 공동의장. (경향신문, 2025-02-20)

The associations have gone straight at the denominator. The franchise owners’ chair: “3년 전 폐점한 가게도 배민에 입점돼있다고 하던데, 그런 가게들까지 포함해 수수료를 책정하는 듯하다. 수수료 책정에 관한 정보를 투명하게 공개해야 한다” — stores that closed three years ago are still listed and appear to be in the calculation (same). And the structural objection, still live in 2026: “플랫폼이 제시한 매출 구간은 국세청 기준이 아닌 배달앱 자체 내부 매출 기준… 이 구조에서는 배달 의존도가 높은 업종일수록 하위 구간에 포함될 수 없다” (머니투데이, 2026-04-28).

Seoul City’s 2025 배달플랫폼 상생지수, built from real merchant settlement data, found total platform fees running 16.9%–29.3% of revenue and singled out “입점업체의 매출 정산내역이 지나치게 복잡해 비용 구조를 파악하기 어려운 점” — the settlement statement is too complex for owners to work out their own cost structure (헤럴드경제 via 네이트, 2025-12-18). That is a government finding that the document this product parses is unreadable by the person who receives it.

The dispute has already reached court on the adjacent issue: 366 franchisees (BBQ, Baskin-Robbins and others) filed suit on 2026-01-12 over commission charged on pre-discount prices — “실제로 받지 않은 2000원에 대해서도 수수료를 냈다”, we paid commission even on the ₩2,000 we never received (조선일보, 2026-01-14).

Baemin’s merchant console, self.baemin.com, shows the owner the band they are in. It does not show the band they are heading into, and it never will — publishing a live distance-to-threshold would invite every merchant to manage revenue down to the cutoff, a direct attack on take rate. The trade-secret position makes that refusal explicit. This is a forecast the incumbent is structurally barred from shipping.

Nor has anyone else shipped it. Checking the obvious candidates — 토스플레이스, 페이히어, 캐시노트, 오터, 배달핏, plus merchant apps 오늘얼마 and 내일장사 — every one stops at aggregating and displaying sales. None recomputes commission, audits a settlement line, or forecasts a band. What exists instead is a layer of blog explainers telling owners to check their band and reconcile their statements by hand: the classic sign of a job that is real, recurring, and unproductised.

That is the gap. Not “delivery fees are too high” — that is a lobbying problem, not a product. The product is: for the next 90 days, here is the cutoff we estimate, here is your current pace against it, and here is what the gap is worth in won.

4. Target market

• Primary customer: Owner-operator of a single independent restaurant in Korea running 배민1플러스 and/or 쿠팡이츠, doing roughly ₩80,000–400,000 a day in delivery revenue — the band that actually sits near a cutoff. Chicken, pizza, bunsik, cafe, late-night. One to five staff. The owner does the settlement reconciliation personally, at night, in Excel.
• Why they buy: The rate spread alone is 1.0pp between the two upper bands and 4.8pp between middle and bottom. On ₩6M of monthly delivery revenue, a 6.8%→7.8% slip is ₩60,000 a month, ₩180,000 a quarter, and the owner had no warning it was coming. On Coupang Eats, sliding out of the bottom band also flips packaging orders from free to 6.8%. Meanwhile 26.2% of revenue is already gone to the platform and rising 5.2pp a year.
• Rough TAM reasoning: Baemin alone carries hundreds of thousands of merchant storefronts; Korea has well over 500,000 F&B businesses with delivery exposure. The addressable slice is the cohort near a cutoff and technically willing to upload a spreadsheet — realistically 80,000–150,000 stores. At 1% penetration and ₩19,900/mo that is ₩190M–360M MRR. Capture 3,000 stores and the business clears ₩700M ARR (~$500K); 12,000 stores puts it near $2M ARR.
• Why now for them: The April 2026 packaging-fee rollout converted the band from a rate question into a channel question, and the September 2026 KBIZ numbers gave every owner a headline proving the burden is accelerating, not easing.

5. Product sketch (MVP)

• Band pace meter. Connect via 배민비즈회원 consent login, or upload the settlement file, and Gugan plots your trailing daily-average delivery revenue against the estimated cutoff, with days remaining in the current measurement window.
• Won-value verdict. “You are ₩7,400/day above the 6.8% cutoff with 26 days left. Staying above costs an estimated ₩180,000 next quarter.” Never a percentile alone — always the cash.
• Coupang Eats rebate audit. Recompute the 7.8%-then-refund cycle against your own order data and flag months where the rebate that landed within five business days does not match what the band implies.
• Packaging-fee cliff alert. For merchants near the bottom-20% line on Coupang Eats, show what crossing it does to pickup orders, which flip from free to 6.8%.
• Line-item settlement diff. Commission charged on pre-discount versus post-discount price, delivery fee per order versus the published band schedule, ad spend against the weekly benchmark Baemin now publishes.
• Quarterly band letter. A dated PDF of your own computed figures — the artefact to attach when disputing a band assignment or filing with a merchant association.
• KakaoTalk alerts at 14 days, 7 days and 3 days before the measurement window closes.

6. AI angle — what’s load-bearing

Two jobs, and the product does not exist without either.

Cutoff estimation. The platform never publishes the won amount, so it must be inferred — from the distribution of uploaded member data, cross-checked against published aggregate statistics and against the quoted anecdotes that anchor the range (the ₩90,000/day figure above). This is a live statistical estimate that sharpens with every upload and must be re-fit every quarter as the merchant population shifts. Early on it is a wide band with an honest confidence interval, not a fake point estimate.

Document extraction. Settlement files differ by platform, change format without notice, and arrive as Excel or PDF in Korean with inconsistent column naming. Parsing them into a comparable order-level ledger — then reconciling one platform’s monthly rebate cycle against another’s quarterly percentile lock — is exactly the messy-document work that got cheap in the last year. Strip the AI out and you have a spreadsheet template that breaks the first time Baemin renames a column.

7. Localization angle

This is Korea-only by construction and that is the entire point. The percentile-band mechanic exists because of a specific Korean political settlement — the 상생요금제 negotiated under government pressure through the 배달앱 상생협의체. No other market prices delivery commission by rank-against-peers, so no global incumbent has a reason to build this.

Everything downstream is local: Korean-language settlement documents, KakaoTalk as the alert channel, ₩19,900/mo pricing that fits a Korean SMB wallet where $49/mo would not, and distribution through 자영업 communities (아프니까 사장이다, 디시인사이드 자영업 갤러리) and merchant associations. A foreign team cannot parse the documents or reach the owners.

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

• Pricing: ₩19,900/mo single store; ₩14,900/mo per store for 3+ (franchise owners running multiple units); one-time ₩49,000 “band dispute pack” for the quarterly letter plus a reconciliation of the last four quarters.
• ACV: ₩239,000 ($170) single store; ~₩540,000 for a 3-store operator.
• Rough math to $1M ARR: 6,000 single stores × ₩19,900 × 12 ≈ ₩1.43B ≈ $1.03M. In a merchant base of hundreds of thousands, that is well under 5% of the near-cutoff cohort.
• Rough math to $5M ARR: ~25,000 paying stores, which requires the multi-store tier to carry real weight plus a second revenue line — most plausibly a per-recovery fee on Coupang Eats rebate shortfalls found, or a paid data product (anonymised band-cutoff benchmarks) sold to merchant associations and the press, who currently have no independent source for these numbers.
• Expansion path: single store → multi-store → franchise HQ monitoring all its units’ bands → the benchmark dataset itself, which gets more valuable and harder to replicate with every quarter of uploads.

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

1. Publish the number nobody else has. Collect 200–300 volunteer settlement uploads in exchange for a free quarterly band letter, then publish “the estimated won cutoff for the top-35% band this quarter” as a dated public post. This figure is currently unavailable anywhere, merchants argue about it constantly, and Korean business press actively covers this beat — 경향신문, 뉴스핌 and 전자신문 have all run merchant-side pieces. One pickup puts the tool in front of the exact cohort.
2. Go where the arguing already happens. 아프니까 사장이다 (Korea’s largest self-employed community) and the 디시인사이드 자영업 갤러리 carry recurring band-dispute threads. Answer them with a free calculation for that owner’s numbers, not a pitch. Target 30–40 conversions from the first 300 replies.
3. Merchant associations as a channel, not a customer. 공정한플랫폼을위한사장협회 and 전국가맹점주협의회 are on record demanding the calculation basis be disclosed. Give them the aggregate dataset free for advocacy; ask for a member mailing in return. These bodies have thousands of members and an active grievance.
4. Franchise HQ back door. A single franchise head office with 40–200 outlets has every unit sitting on the same band ladder. Ten HQ deals at the multi-store tier is ~800 stores without retail acquisition cost.
5. Ride the KFTC news cycle. With fines of ₩239–510B pending against Baemin and the consent settlement rejected, commission-fairness coverage will keep running. Have the calculator ready each time a story lands.

10. Build complexity — justification

Medium, and the data path is the risky part rather than the modelling. There is no self-serve merchant API: the 2023 배달 플랫폼 표준 API is an order-relay standard, contract-gated to 13 named firms, with settlement data out of scope. The live route is 배민비즈회원 merchant-consent login — the channel Baemin built to replace ID/PW and the one Cashnote runs on in production today — with manual file upload as the fallback, as 정산플러스 does for dispatch agencies.

That path carries real friction the product must absorb: Baemin delivers per-order settlement detail by email or fax rather than in-browser download, and Coupang Eats does not expose the current month until the 5th of the following month (“당월 자료는 다음 달 5일부터 조회할 수 있어요”). The custom work is the cutoff estimator plus parsers against two platforms with different billing clocks and unannounced format changes. A pair with Korean-language capability ships v1 in 10–14 weeks; the estimator only earns trust after one full quarterly cycle.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Runs on the merchant’s own data with explicit consent, via the 배민비즈회원 channel Baemin itself built to replace ID/PW scraping, or by direct upload. Conditional on 배민비즈 terms permitting third-party integration — verify before building.
Ethical — no harm / dark patterns	✅	Gives the weaker party its own numbers. Must be disciplined about presenting cutoffs as estimates with confidence intervals, never as platform-confirmed fact.
Market exists (evidence above)	✅	KBIZ 1,250-merchant survey, named association officials quoted on the record, funded adjacent fintech.
1–5 person team can build this	✅	Two people, 10–14 weeks.
Launchable with <$50K / ₩40L	✅	Parsing and inference costs are modest; the real spend is Korean-language community presence.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Real, recurring, quantified money — 26.2% of revenue and rising 5.2pp a year. Docked because the pain is quarterly-lumpy rather than daily, and 84.8% telling KBIZ the tiered scheme “didn’t help” cuts both ways: it proves the anger and hints owners may see the whole band game as rigged beyond rescue.
Demand evidence	15	14/15	Unusually well-evidenced: a 1,250-merchant KBIZ survey, Seoul City’s settlement-data index finding the statement unreadable, a 366-plaintiff lawsuit over commission basis, named association officials quoted disputing bands, pending KFTC fines of ₩239–510B, and funded adjacent fintech. Docked one point only because no one is yet paying for band forecasting specifically.
Build feasibility	15	9/15	Docked two points after verification: there is no self-serve merchant API, the 2023 표준 API is contract-gated to 13 firms with settlement out of scope, and the live route (배민비즈회원 consent access) may be bilateral. Baemin’s email/fax delivery of per-order detail and Coupang’s 5th-of-next-month lag add real friction on top of the estimator and dual-platform parsers.
Distribution clarity	15	11/15	Named communities, named associations, a press beat that reliably covers this, and a publishable hook no competitor owns. Docked because the largest community is a gated Naver cafe and Korean forums are hostile to outside vendors.
Revenue mechanics	15	10/15	₩19,900 is well inside a Korean SMB wallet against ₩180,000-a-quarter stakes. Docked because $1M ARR needs ~6,000 paying stores of low-ACV SMBs — a churn-exposed base — and the $5M path depends on a second revenue line that is not yet proven.
Time to first revenue	10	6/10	Needs one full measurement cycle before the forecast is credible enough to charge for. The dispute-pack one-off can pull revenue forward, but honest first paid customers are 3–4 months out.
Defensibility	10	4/10	The accumulating upload corpus is the only real moat and it compounds slowly. Cashnote or Toss Place could add this feature; the platform could blunt it by publishing cutoffs. Execution-and-focus moat, not a structural one.
Total	100	70/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · content-heavy — and Korean-language fluency is non-negotiable, not a nice-to-have. This is not a market a non-Korean team can serve.

Key assumptions to validate (3–5)

1. Assumption: The cutoff can be estimated usefully from a few hundred uploads rather than needing platform-scale data. How to test: Collect 200 real settlement files with known assigned bands; check whether the inferred cutoff separates known-top-35% from known-middle stores with a tight enough interval to act on. If the interval is wider than the behaviour change it would prompt, the product has no spine.
2. Assumption: Owners will change behaviour on a forecast — shift orders to direct or pickup channels, pace Baemin1Plus days — rather than shrug. How to test: Interview 30 owners near a cutoff with a concrete won figure and ask what they would actually do. The 84.8% “didn’t help” response is the specific risk here.
3. Assumption: Coupang Eats rebate errors exist at a findable rate. How to test: Recompute 50 merchant-months against the published band schedule and count mismatches. A material hit rate turns the audit into the lead product and materially improves time-to-revenue.
4. Assumption: Merchants will hand over settlement data to an unknown vendor. How to test: Run the free-band-letter offer cold in two communities and measure conversion.
5. Assumption: 배민비즈회원 consent access is open to third parties rather than a bilateral Cashnote deal. How to test: Register a merchant account, read the 배민비즈 terms and 정보제공 동의 scope directly, and confirm whether an independent developer can connect. This is the first thing to check, before any modelling work.

Risk flags

1. Platform dependency and countermeasure risk. Baemin could publish cutoffs and vaporise the core value, or change the mechanic entirely — plausible given it is under active KFTC pressure with fines of ₩239–510B pending. Mitigation: the settlement-audit and rebate-verification lines survive any band change.
2. Estimate liability. Selling an inferred number as if it were confirmed invites both merchant anger when it is wrong and platform legal attention. The confidence interval must be visible on every screen, and the product must never claim platform endorsement.
3. Fatalism in the base. 84.8% told KBIZ the tiered scheme did not help them. A demoralised market is harder to sell agency to than an angry one.
4. Data-access dependency — the sharpest risk. Everything rests on 배민비즈회원 consent access continuing to permit third-party integration. Cashnote operates on it, but it is unclear from public sources whether that is an open channel or a bilateral arrangement, and Woowa publishes no developer documentation for it. If it is bilateral, the product falls back to manual upload with materially worse retention. Verify this before writing code. Do not count on the 본인전송요구권 (MyData) regime as a rescue: the effective dates are news-sourced rather than confirmed against the enacted 시행령, and more fundamentally it is a right of a natural person over personal data — whether a store’s settlement ledger is even in scope is unresolved, and for a 법인 it very likely is not.
5. Cold-start on the estimator. The forecast is weak until the corpus exists, which is precisely when the product must convince its first customers. The free-letter-for-upload trade is the intended bridge and it may not clear.

14. Structured verdict

Score:                  70/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Korean-fluent technical founder, ideally with F&B or POS-adjacent
                        background, paired with someone who can hold a presence in
                        자영업 communities without being ejected as a vendor
Time to revenue:        3–4 months (one measurement cycle); dispute-pack one-offs sooner
Capital to launch:      ₩30–50M ($22–36K)
Top 3 assumptions to validate first:
  1. 배민비즈회원 consent access permits independent third-party integration — read the
     terms from a real merchant account before anything else is built
  2. Cutoff is inferable from ~200 connected stores to an interval tight enough to act on
     — back-test against stores whose assigned band is already known
  3. Owners act on a forecast rather than shrug — 30 interviews with a real won figure
Kill criteria:
  - Abandon if 배민비즈 access proves bilateral-only AND manual upload retention in the
    pilot falls below 50% at month two — without a data path there is no product
  - Abandon if the inferred cutoff interval stays wider than ±15% after 200 stores —
    the forecast cannot drive a decision it cannot resolve
  - Abandon if fewer than 10 of 30 interviewed owners name a specific action they would
    have taken on a concrete won-denominated warning
  - Abandon if Baemin publishes official band cutoffs or ships a forward-looking band
    indicator in self.baemin.com

15. Next step — 1-week validation sprint

• Day 1: Register a merchant account and read the 배민비즈회원 terms and 정보제공 동의 scope first-hand. If third-party consent access is closed, the entire product shape changes and the rest of the week should test manual-upload willingness instead.
• Day 2: Recruit 40 owners from two 자영업 communities and the two merchant associations with a free offer: send your settlement data, get a written band analysis back. Measure what fraction actually follow through — that alone tests the data path.
• Day 3–4: Back-test. For every uploaded file where the owner knows their assigned band, fit the cutoff and check whether the estimate separates known-top-35% from known-middle stores. Simultaneously recompute the Coupang Eats rebate for every merchant-month in the sample and count mismatches.
• Day 5: Call 30 of those owners with their specific number — “you were ₩6,200/day over the line last quarter, it cost you roughly ₩170,000” — and ask one question: what would you have done differently if you had known in week two?

Falsifiable outcome: go only if (a) the fitted cutoff interval is tighter than ±15%, and (b) at least 10 of 30 owners name a specific action they would have taken. Anything softer than a named action is politeness, not demand.