Menkan — business-tax filer for Taiwan creators

1. One-liner

Tallies every platform payout against Taiwan’s NT$50K line and drafts the 401 with audience-split evidence attached.

2. Trend signal — why now?

Taiwan gave its creators a new tax status in September 2025, gave them nine months to sort it out for free, and turned the meter on 1 July 2026. Nobody has built the monthly workflow that status demands.

The rule. On 10 September 2025 the Ministry of Finance published the 個人經常性於網路發表創作或分享資訊課徵營業稅作業規範. Any Taiwan-resident individual who regularly publishes online and whose monthly sales hit the business-tax threshold (NT$50,000 for services, NT$100,000 for goods) must register a 稅籍, and the income that counts is exactly what creators live on: 「平臺自廣告主取得勞務收入」 (ad share), 「自平臺取得分潤性質勞務收入」 (memberships and revenue share), donations, sponsorships. Below NT$200,000 a month the bureau assesses 1% on declared sales. At NT$200,000 and above the creator issues 統一發票 and files a 401 return every two months at 5%, with revenue from overseas viewers eligible for a 0% rate if it is evidenced.

The grace period is gone. The 輔導期 ran from 10 September 2025 to 30 June 2026 (pay by 15 July 2026). During it, non-registration and non-filing were 「免依營業稅法第45條、第51條、第52條及稅捐稽徵法第44條規定處罰」. After it, the accountant blogs updated in the same week: 「免罰輔導期已於 115 年 6 月 30 日截止。現在起未依規定辦理，國稅局可以直接開罰。」 (合佑會計師事務所, 10 July 2026). The bureau’s own line: 「經檢舉或查核發現違規，將依規定補稅並裁罰」.

The regulator published the count. On 19 July 2026 the Ministry told the press that 1,242 creators with monthly platform income above NT$50,000 had registered, 46 of them had since quit creating, and the cohort had paid NT$52.61 million of business tax in nine months, about NT$42,000 each including back taxes. The same Ministry, on 7 May 2026, said Taiwan has over 300,000 creators on KOL Radar’s count. Whichever slice of those 300,000 sits above NT$50,000 a month, 1,242 is not it. That gap is the market.

The audit is automated now. Deloitte Taiwan’s tax partner, quoted in 聯合新聞網 on 6 May 2026, on the bureau’s generative-AI case selection: 「未來不論是零售、電商、餐飲、網紅或代購業者，若稅負異常偏低、漏開發票或漏報收入，都可能成為高風險查核對象。」 Separately, banks now hand the bureau every personal account that receives NT$2.4 million a year with any four months of 200+ inbound transfers. A streamer’s 抖內 account looks exactly like that.

The evidence burden is monthly and nobody sells it. To claim 0% on overseas viewers the creator must, in the accountants’ words, 「報繳 0% 營業稅，但需檢附後台數據證明」 and 「把後臺報表與 Google 的外匯水單交給會計師，直接向國稅局申報適用 0% 零稅率」. That is a per-month YouTube Studio geography export, an AdSense payment receipt, a Twitch or 17LIVE payout statement, an Oen or 歐付寶 donation ledger, and a brand-deal invoice, reconciled into one return. Bookkeepers price that as hours. Donation platforms automate only their own rail.

Provenance:

• Signal 1 (demand): MOF press statement — 1,242 creators registered, 46 quit, NT$52.61M collected in nine months, grace ended 30 June 2026 — https://news.nextapple.com/finance/20260719/699940AEF4F681EF1BB5CD8D8F65C676 — 2026-07-19
• Signal 2 (regulatory): MOF 作業規範 — NT$50K/NT$100K thresholds, income types, 輔導期 2025-09-10 to 2026-06-30, statutes waived during grace — https://www.mof.gov.tw/singlehtml/384fb3077bb349ea973e7fc6f13b6974?cntId=489252e0c77d435a916dee12ce8b0450 — 2025-09-10
• Signal 3 (economic / enforcement): Deloitte on AI case selection naming 網紅 and 代購 as high-risk; MOF says 300,000 creators — https://udn.com/news/story/7243/9485812 — 2026-05-06; https://news.nextapple.com/finance/20260507/650681B39E286886F0C5166170F38AB8 — 2026-05-07
• Signal 4 (feasibility): every evidence item is an exportable CSV/PDF and frontier-model document parsing is now sub-dollar per creator-month (Gemini 3.8 Flash at $0.75/M input, 2 Sep 2026) — https://enterprisedna.co/resources/news/google-gemini-38-flash-coding-agents-enterprise-september-2026/ — 2026-09-02
• Signal 5 (incumbent shape): Oen 應援科技 auto-issues invoices only for money collected through Oen, at a 6% minimum commission — https://tech.oen.tw/posts/3ChaD9d3PUdeHma6RkN1qy6iAw0 — 2026 Category: Regulatory arbitrage

3. The opportunity

Three parties touch this money and none of them owns the monthly file.

Bookkeepers (記帳士) sell hours, not the artefact. Taiwan’s going rate for a business under NT$1M annual revenue is NT$1,500–2,000 a month; 記帳士 quote NT$2,000–4,000. Every firm that wrote a “網紅稅” guide in 2026 (合佑, 聖彥, 合承, 潤元, 勤億磐石, 森左, 長奕, 專精) is selling the same thing: you send us the exports, we file. The accountant’s own FAQ shows what they get sent: 「我 YouTube 每個月分潤大概六、七萬，需要辦營業登記嗎？」 They answer that by email, monthly, per client. The geography report, the 外匯水單, the 二聯 vs 三聯 question (「發票到底該開二聯還是三聯？」), the 抖內 question (「粉絲熱情斗內到公司戶頭，這筆心意也要開發票嗎？」) are all handled by hand.

Donation platforms automate their own rail only. Oen 應援科技 will issue an e-invoice 「粉絲一付款自動寄送發票」 and warn you near NT$180,000, for a 6% cut of donations routed through Oen. It cannot see AdSense, Twitch, 17LIVE, Facebook Stars, or the brand deal paid to your bank. The MOF’s own first principle is that the NT$50K test is the sum of all streams, so a per-rail tool is structurally unable to answer the only question that matters.

The bureau gives a simplification for income tax and none for business tax. For 綜所稅 the ministry allows a flat 50% domestic attribution if you can show overseas viewership. For 營業稅 the 0% rate needs the actual monthly split and the actual payment record. That is the artefact: a dated, per-period evidence file that a 401 return points at.

Two regimes, one decision nobody computes. A 查定 creator pays 1% on everything with no 0% available; a 統一發票 creator pays 5% on the domestic share and 0% on the overseas share. Which is cheaper depends on the audience mix and the run-rate, and it flips around 80% overseas share. Accountants answer it once at onboarding. The product answers it every period.

The gap is the monthly loop: tally, threshold, split, draft, evidence. The 1,242 who registered need it now. Everyone who crosses NT$50K from here on has 15 days to register and no grace.

4. Target market

• Primary customer: Taiwan-resident individual creators paid directly (not through an agency) by two or more of: YouTube AdSense, YouTube memberships/Super Chat, Twitch, 17LIVE, Facebook, Oen/歐付寶/綠界 donations, Podcast sponsors, brand deals. Combined services income NT$50,000–500,000 a month. Sole proprietors, no staff, no accountant or a 記帳士 they resent paying.
• Why they buy: Registration is a 15-day clock they did not know was running. The 0% rate is worth 5% of their overseas share every period and requires evidence they have never exported. Fines are NT$3,000–30,000 per notice, repeatable, plus up to 5× the tax evaded and a possible business suspension, and the bureau’s bank feed and AI case selection now find them.
• Rough TAM reasoning: 1,242 registered by end-June 2026 is the floor and it was collected under an amnesty. KOL Radar’s count is 300,000 creators with 68.5% under 10,000 followers, so the sub-1-million-follower band is roughly 95,000 accounts. If 3–5% of those clear NT$50K a month across streams, the pool is 3,000–5,000 creators, of whom a quarter have registered. Add 代購 and 團購 hosts on the NT$100K goods line as a second cohort of unknown size but named in the same enforcement memo. At NT$12,000–24,000 a year this is a NT$50–100M (US$1.5–3M) ceiling on creators alone. Honest business, not a venture.
• Why now for them: From 1 July 2026 every month above the line without a 稅籍 is a fineable month. The July 2026 press cycle put the number in front of them. The next 401 period closes 15 November 2026.

5. Product sketch (MVP)

• Statement intake. Drop in YouTube Studio geography and revenue exports, AdSense payment receipts, Twitch/17LIVE/Facebook payout PDFs, Oen/歐付寶/綠界 ledgers, brand-deal invoices. Traditional Chinese and English formats both parse.
• Threshold tally. One number per month: combined services income versus NT$50,000 (and goods versus NT$100,000). The month you cross, a 15-day registration countdown starts and the 稅籍登記 application packet is pre-filled.
• Audience split. Per period, domestic versus overseas revenue share per platform, tied to the geography export it came from, so the 0% claim has a document behind it.
• Regime pick. 1% assessed versus 5%/0% invoiced, computed on your actual mix, with the crossover shown. Re-run every period, not once.
• 401 draft. For invoiced creators: the bimonthly return figures with the 0% schedule, the 外匯水單 references, and the invoice list, ready to hand to a 記帳士 or key into the e-filing portal.
• Invoice guidance. Which counterparties need an invoice, which copy (二聯 or 三聯), and which donation platform already issued one so you do not double-issue.
• Evidence binder. Every period’s exports, receipts and drafts sealed with dates, exportable as one folder when the bureau letter comes.
• Bookkeeper seat. A read-only view for the creator’s 記帳士, so the firm stops asking for exports by LINE.

6. AI angle — what’s load-bearing

Parsing is the work. Twelve platforms, each with its own statement layout, two languages, currencies in USD and NT$, geography tables at country level that have to be mapped onto payment periods that do not line up with calendar months. A model reads each document, extracts period, gross, platform fee, net, currency, and the per-country revenue table, then classifies the line as ad share, revenue share, donation, sponsorship or goods, because the MOF taxes each differently. The regime comparison and the 401 arithmetic are deterministic once that is done. Take the model out and you are back to a bookkeeper retyping AdSense PDFs, which is the incumbent.

7. Localization angle (if any)

This is a Taiwan-only product by construction. Traditional Chinese interface, NT$ thresholds, the 401 form, the 統一發票 system, 15-day registration clock, and the exact statutes cited on the fine letter. Payment by 綠界 or credit card. Support over LINE, because that is where creators and their 記帳士 already talk. The identical shape exists in Japan (消費税 for creators past ¥10M) and Korea, but each needs its own rules engine and its own bookkeeper channel; treat them as year-three options, not launch scope.

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

• Pricing: NT$899/month (≈US$28) for the 查定 tier (tally, threshold, registration packet, evidence binder). NT$1,899/month (≈US$60) for the 統一發票 tier (adds split, 401 draft, invoice guidance). NT$399/creator/month white-label seat for 記帳士 firms managing 10+ creators.
• ACV: Blended NT$14,000 (≈US$440) per creator-year.
• Rough math to $100K ARR: 230 creators, which is under a fifth of the cohort the Ministry has already named.
• Rough math to $1M ARR: 2,300 paying creators, or 1,200 direct plus 1,100 through bookkeeper seats. That is roughly the whole registered cohort plus half the unregistered pool, so it needs the 代購/團購 cohort on the NT$100K goods line to be real. Eighteen to twenty-four months.
• Rough math to $5M ARR: Not from Taiwan creators alone. Requires the Japan or Korea rules engine or the broader Taiwan online-seller population (the same bureau data pull covers Shopee and PChome sellers above NT$200K a month). A year-two decision.
• Expansion path: 查定 tier upgrades to 統一發票 tier as the creator grows past NT$200K. Annual 綜所稅 pack as a paid add-on each May. Bookkeeper seats compound as firms move their creator books onto it.

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

• The registered cohort is a public list. The Ministry’s 全國營業(稅籍)登記資料集 on data.gov.tw is a daily CSV with 設立日期, 組織別, 使用統一發票 and 行業代號. Filter sole proprietorships set up since September 2025 under advertising-services and related codes, cross-match business names against channel names, and you have most of the 1,242. They already pay a bookkeeper or file themselves. A LINE message with their own audience-split estimate from their public channel gets a reply. Expect 300 matches, 10% conversation rate, 30 customers.
• The accountants who wrote the guides. Eight firms published “網紅稅” explainers in 2026 and each carries a book of creator clients whose exports arrive by LINE. Offer the white-label seat free for their first ten creators. Two firms adopting at 15 creators each is 30 customers and a reference.
• Threads and YouTube accountants with audiences. 聖彥 posts creator-tax threads to a creator following (「一旦你的頻道做大，月收穩定超過 5 萬…必須去辦理「稅籍登記」」). Sponsor one explainer video per month with a free threshold check as the call to action. Expect 1,000 checks a month, 3% convert, 30 customers a quarter.
• The agencies’ overflow. CAPSULE, PressPlay and VS Media manage creators whose agency-paid deals do not need a personal 稅籍, but whose direct AdSense and donation streams do. The agency wants those creators registered so its own invoices are clean. One agency pilot is 10–20 creators.
• Unregistered but public. Streamers who show their 抖內 totals on screen and YouTubers whose Social Blade estimates exceed NT$50K a month. Cold outreach with the estimate and the statute. Low yield, but the leads are enumerable.

10. Build complexity — justification

Medium. No API access is required: every input is an export the creator already has. The custom work is the statement parser across a dozen formats, the period-alignment logic, the regime comparison, and the 401 draft with its 0% schedule. The registration packet and invoice-copy rules are a small rules table maintained against MOF notices. A pair with Traditional Chinese fluency ships v1 in 10–12 weeks. The moat is the rules table and the parser corpus, both of which need a 記帳士 advisor on retainer.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Drafting returns and evidence for the creator to file, or hand to a licensed 記帳士, is bookkeeping software. Signing and filing on the creator’s behalf stays with the licensed firm.
Ethical — no harm / dark patterns	✅	Gets creators registered and paying the right rate, including the 0% they are entitled to.
Market exists (evidence above)	✅	1,242 registered, NT$52.61M paid, eight accounting firms publishing guides, one donation platform selling a partial fix.
1–5 person team can build this	✅	Two people, one of them Chinese-fluent with a tax advisor.
Launchable with <$50K / ₹40L	✅	Model inference is cents per creator-month. Main cost is the advisor retainer and the first agency pilot.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	13/20	Real, monthly, fineable since 1 July 2026, and the 0% rate is money left on the table each period. But the fine for non-registration is NT$3,000–30,000 and waived if you comply on first notice, so the hair is smouldering, not on fire, until a 5× evasion penalty lands.
Demand evidence	15	10/15	Primary numbers from the Ministry, a Big Four warning, eight firms competing on SEO, a platform already selling a partial fix. Missing: direct creator quotes (Dcard is behind a captcha) and a published post-grace penalty case.
Build feasibility	15	12/15	Exports only, no API gating. Parser breadth is the work. 10–12 weeks for a pair.
Distribution clarity	15	9/15	The registered cohort is enumerable from an open dataset and the bookkeeper channel is named. Conversion from a LINE cold message is a guess, and the unregistered majority has to be reached through accountants’ audiences.
Revenue mechanics	15	8/15	Pricing sits under bookkeeper rates and above Oen’s effective take. But $1M ARR needs most of the identifiable cohort or a second population, and the 查定 tier customer can plausibly do the tally in a spreadsheet.
Time to first revenue	10	7/10	The 401 period closing 15 November 2026 is the first natural close. Six to eight weeks from launch to first paid bookkeeper seat.
Defensibility	10	4/10	Parser corpus and rules table are copyable by any 記帳士 firm with a developer. Workflow lock-in through the evidence binder is the only moat.
Total	100	63/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

Traditional Chinese fluency is not optional. A 記帳士 advisor on retainer from week one.

Key assumptions to validate (3–5)

1. Assumption: At least 30% of creators above NT$50K a month are paid directly by two or more platforms, so the cross-stream tally is a real problem and not an Oen feature. How to test: Ask 40 registered creators from the open dataset which platforms pay them directly. If most are single-stream or agency-paid, the product collapses to a threshold calculator.
2. Assumption: Creators at NT$200K+ (the 統一發票 tier) exist in the hundreds, not dozens, and value the 0% split at NT$1,899 a month. How to test: From the same dataset, count 使用統一發票 = yes among post-September-2025 sole proprietorships in advertising-services codes. Under 150 means the top tier is a feature, not a plan.
3. Assumption: 記帳士 firms will put creator clients on a white-label seat rather than build a spreadsheet. How to test: Pitch the eight firms that published guides. Two pilots by week four or the channel is dead.
4. Assumption: The bureau accepts YouTube Studio geography exports as the 0% evidence without a further platform certificate. How to test: Ask the advisor to confirm against two filed cases; if a platform letter is required, the evidence binder needs a different anchor.
5. Assumption: The 代購/團購 cohort on the NT$100K goods line is reachable through the same channels. How to test: Not before month six. Park it.

Risk flags

1. Regulatory risk: The MOF could publish a business-tax simplification (a flat domestic share like the income-tax 50%) and remove the evidence burden that the top tier sells. It could also raise the threshold. Either shrinks the product to a tally.
2. Platform dependency: If YouTube, Twitch or Oen start issuing Taiwan-compliant invoices and geography certificates natively, the per-platform parsing loses value. Oen already does this for its own rail.
3. Market size: 1,242 registered is a floor with an unknown multiple above it. If the true above-threshold population is under 3,000, this is a NT$30M business and a lifestyle product.
4. Customer voice gap: Every quote in this file is a regulator, a Big Four partner, or an accountant relaying client questions. Direct creator testimony was not obtainable from Dcard or Threads through search. Collect it in week one.

14. Structured verdict

Score:                  63/100
Verdict:                VALIDATE
Confidence:             Medium
Best-fit builder:       Traditional-Chinese-fluent technical founder with a 記帳士 advisor; ideally someone who has filed a 401 for a creator client
Time to revenue:        6–10 weeks (first bookkeeper seat or first 統一發票-tier creator before the 15 Nov 2026 filing)
Capital to launch:      NT$600K–1M (US$20–30K): advisor retainer, parser build, one agency pilot
Top 3 assumptions to validate first:
  1. Multi-stream direct payment is the norm above NT$50K — interview 40 creators from the open 稅籍 dataset
  2. Hundreds, not dozens, of creators are on 統一發票 — count 使用統一發票 among post-2025-09 sole proprietorships in the dataset
  3. Bookkeeper firms will adopt a white-label seat — pitch the eight firms that published 2026 guides, need two pilots in four weeks
Kill criteria:
  - Abandon if fewer than 12 of 40 interviewed creators are paid directly by two or more platforms
  - Abandon if the dataset shows under 150 invoicing creator sole proprietorships and no bookkeeper firm pilots by week six
  - Abandon if the MOF publishes a flat domestic-share rule for business tax before launch

15. Next step — 1-week validation sprint

• Day 1–2: Pull the 全國營業(稅籍)登記資料集 CSV. Filter sole proprietorships with 設立日期 after 10 September 2025 in advertising and media services codes. Count them, split by 使用統一發票. Cross-match 60 names to channels. That is the market size and the lead list in one afternoon.
• Day 3–4: LINE or email 40 of them with a one-line offer: send me your last AdSense receipt and geography export and I will tell you your 0% share and your regime crossover for free. Log who replies, how many platforms pay them directly, who has a 記帳士 and what they pay. Simultaneously send the white-label pitch to the eight accounting firms.
• Day 5: Go if 12+ creators replied with exports, 8+ are multi-stream, and at least one firm asked for a pilot call. No-go if the dataset shows under 150 invoicing creators and the firms are silent.