HoKhai — quarterly tax filer for Vietnamese household shops

1. One-liner

Zalo-bot that turns till-tape photos and VietQR inflows into quarterly VAT+PIT filings for Vietnam’s 5M household shops.

2. Trend signal — why now?

Vietnam just blew up the way 5 million micro-shops pay tax. Effective 1 January 2026, the lump-sum tax regime (thuế khoán) — the rough estimate the local tax officer wrote on a slip every year — is abolished. Every household business (hộ kinh doanh) above the new 500M VND/yr threshold now has to self-declare actual revenue, calculate VAT (1%/2%/3% by activity code) and personal income tax, and file quarterly through the eTax Mobile app. First filing cycle was Q1/2026, deadline 30 April 2026. Next: 31 July 2026.

The VCCI survey is the killer: 49% of households worry about cost/time of invoices, 73% lack the tech skills, 53% find the new procedures complicated. One Đà Nẵng grocer quoted in VietnamNet hires an accountant for ₫4M/month — almost 4× their old fixed tax bill. Cục Thuế leadership publicly apologized for the rollout chaos. Only 72,000 of ~5M households are actively using eTax Mobile — 1.4% adoption.

Decree 70/2025 also already forced households >1B VND in F&B, hotels, retail, transport, beauty, entertainment to switch to cash-register-generated e-invoices from June 1 2025. Q3/2026 deadline lands while the gov is still firefighting Q1 complaints. The pain is fresh, recurring, and quarterly.

Provenance:

• Signal 1: “Bỏ thuế khoán từ 2026: Gần 5 triệu hộ kinh doanh lúng túng” — https://vietnamnet.vn/bo-thue-khoan-tu-2026-gan-5-trieu-ho-kinh-doanh-lung-tung-truoc-cuoc-choi-moi-2473219.html — 2026-03
• Signal 2: VCCI survey — 49% cost/time worry, 73% lack tech skills, via VietnamNet “From manual to digital: how small businesses will survive the 2026 tax shift” — https://vietnamnet.vn/en/from-manual-to-digital-how-small-businesses-will-survive-the-2026-tax-shift-2473996.html — 2026-04
• Signal 3: Cục Thuế leadership apology re: Decree 70 rollout — https://vietnamnet.vn/bo-thue-khoan-gay-vuong-mac-lanh-dao-cuc-thue-xin-loi-ho-kinh-doanh-2507612.html — 2026-05 Category: Regulatory arbitrage

3. The opportunity

The Vietnamese government just created the largest mandatory bookkeeping migration in SEA — 5M owner-operators forced into quarterly self-declaration overnight — and the existing players are aimed at the wrong end of the problem.

• MISA meInvoice / EasyInvoice / VIN-HOADON all sell invoice issuance (the seller-side compliance need). They charge per invoice and ship a desktop or per-seat app. They do not interpret a shoebox of receipts and a bank statement into a filed VAT declaration.
• eTax Mobile (the official Cục Thuế app) is bare-bones. It accepts a declaration; it doesn’t prepare it. The user still has to know their activity code, the 1% vs 3% bucket, what counts as deductible, and which day a missing receipt landed in.
• Kế toán dạo (roaming bookkeepers) charge ₫1.5–4M/mo per shop. They don’t scale; there aren’t enough of them; they hate the work.

The 10× wedge: a Zalo Official Account that the shop owner already lives in. They snap a photo of the till tape at end-of-day, the bot OCRs it, dedupes against the VietQR inflows pulled from their bank, infers the activity bucket from prior tagging, and drafts a one-tap quarterly declaration that the owner literally just confirms in eTax Mobile. The “no app to install” Zalo-first surface beats every desktop incumbent for the elderly phở-shop / nail-salon owner who is the actual ICP.

4. Target market

• Primary customer: Tier-2 household businesses earning 200M–3B VND/yr (~$8K–$115K) — phở shops, cafés, nail salons, motorbike-repair shops, mini-marts, family-run guesthouses, tiffin services. Owner-operator, 1–3 employees, no in-house accountant. Concentrated in Hà Nội, HCMC, Đà Nẵng, Hải Phòng, Cần Thơ first.
• Why they buy: Q1 2026 was their first cycle. They missed deadlines, screamed at the local tax officer, then panic-hired a ₫4M/mo bookkeeper for Q2. They want a Q3-or-Q4-of-2026 solution that costs an order of magnitude less and doesn’t require them to “open the computer.”
• Rough TAM reasoning: ~5M household businesses. ~1.5–2M in the must-file 500M–3B VND tier. 500K already on VietQR/POS and reachable via Zalo OA day one. Capture 5% of that reachable cohort = 25K customers @ ₫149K/mo ($6) = ~$1.8M ARR.
• Why now for them: First quarterly cycle is over; they have lived the pain once and will pay to not live it again next quarter. The 30 Jul 2026 deadline is the natural pre-sale moment.

5. Product sketch (MVP)

• Add the HoKhai Zalo Official Account; complete a 3-question setup (activity code, address, registration number) over chat
• Snap-and-send end-of-day till-tape photos to the bot — OCR + tagging, dedupe vs. VietQR inflows
• Auto-import VietQR/bank-account inflows by scraping the daily bank-SMS forward or via VietQR API where supported
• A daily Zalo reminder (“you closed yesterday — did you forget the cash box?”) that catches missed days
• End-of-quarter draft: ready-to-submit VAT (1% / 2% / 3% / 5% by activity) + PIT calculation + line-by-line backup
• Generates the eTax Mobile-compatible XML / form so the owner taps “submit” inside the official app
• Vietnamese-only UI; voice-note Q&A for elderly owners (“bạn ơi, hôm qua bán bao nhiêu?”)
• One-tap export packet for the local tax officer if challenged

6. AI angle — what’s load-bearing

Three places AI does the actual work, not the decoration:

1. Receipt OCR + activity classification — Vietnamese till-tape photos are low-resolution and use weird character sets. A small fine-tuned VL model maps each line to the correct VAT activity bucket. Hand-coded rules can’t do this — there are thousands of permutations of “phở bò tái” written shorthand.
2. Bank-inflow ↔ till-tape reconciliation — matching a VietQR inflow to a day’s receipts is fuzzy (timing, batching, gratuities). LLM with retrieval over prior matches gets it right with one-shot owner correction.
3. Conversational Vietnamese tax interview — the owner messages “tôi mua thêm 2 thùng nước ngọt hôm qua” and the bot infers a deductible, asks for the supplier invoice, files it. Replaces the chat-with-an-accountant function entirely.

Strip the AI and you have a glorified spreadsheet — and the customer would already use MISA. The AI is the only reason this is a Zalo chat at all.

7. Localization angle

This is a 100% Vietnam play. Local-only mechanics:

• Zalo is the messenger (76M+ users; tax depts already organize household-business communication in Zalo groups). Building on Zalo OA + ZNS is non-negotiable; WhatsApp doesn’t matter here.
• VietQR / NAPAS is the payment rail. 85% YoY street-vendor cashless growth means inflow data is now reliably digital.
• Vietnamese language, tone, activity codes — must speak the language of an elderly market trader, not a Hanoi accountant.
• eTax Mobile compatibility — the only legal submission surface for the household-business segment.
• Price floor — ₫149K/mo ($6) maps to roughly a day’s revenue at the smallest covered shops. $49/mo SaaS is dead on arrival.

A foreign player would need a Vietnamese co-founder, a Zalo OA approval, and tax-officer relationships. That’s a 12-month moat for a local team that already has them.

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

• Pricing: ₫149K/mo ($6) self-serve via Zalo. ₫299K/mo ($12) with priority human escalation when a Cục Thuế audit notice arrives.
• ACV: ~$72–$144/yr per household; weighted blend ~$90.
• Path to $1M ARR: ~11,000 paying households on the $90 blended ACV. From a reachable pool of ~500K Zalo-OA-friendly households, that’s 2.2% conversion.
• Path to $5M ARR: ~55,000 households. At that scale, add (a) white-label seat for the kế toán dạo network (each bookkeeper services 50–100 shops on top of HoKhai), (b) lender referral (lender wants 90 days of clean revenue for an SME loan — HoKhai is the cleanest source), (c) e-invoice issuance bundle to keep MISA from upselling around us.
• Expansion: When the shop crosses 3B VND and is forced to incorporate into a real enterprise, the same data set drives a paid annual-settlement bundle (~$200/yr one-time at year-end) and an optional bookkeeper-marketplace match.

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

The first 100 are concrete, named, and reachable in <8 weeks:

• Hà Nội traditional-market trader Zalo groups — Cục Thuế Hà Nội already structured a 3-tier Zalo communication network connecting tax officers to households. Buy a list of ~200 of those traders from a market-association contact, message each personally in Vietnamese, offer free Q3/2026 filing as the trial. Target 30 conversions.
• HCMC nail-salon and phở-shop Facebook groups (e.g. “Hội Chủ Quán Phở”, several hundred-thousand-strong groups) — one-line post offering “miễn phí kê khai Q3 cho 100 hộ đầu tiên” + a Zalo OA link. Target 40 conversions.
• Kế toán dạo affiliate — pay 30% rev-share for the first year to the freelance bookkeepers who already service 10–30 shops each. Recruit 10 bookkeepers via Facebook bookkeeper groups → carry 30+ shops as charter customers.
• Tax-office partnership pilot — Đà Nẵng tax dept is already doing in-person visits to traditional markets to drive Q3 compliance. Offer them HoKhai as the “tool we hand out at the doorstep” with co-branding. Conversion: 100% of in-person handouts that go home with the QR.

Marketing copy stays in Vietnamese: “Đừng để kế toán dạo lấy 4 triệu/tháng nữa. Chụp ảnh – chốt sổ – nộp thuế. Hết Q3 trong 3 phút.”

10. Build complexity — justification

Medium. The hard pieces are real but boring: a Zalo OA bot, a Vietnamese VL model fine-tune for till-tape OCR, an eTax Mobile XML adapter, a bank-SMS scraper or a permissioned VietQR API integration per bank. No novel ML; cheap inference (Vietnamese-tuned Gemini or local LLMs at $0.05/MTok-ish). Ship v1 in 10–14 weeks with 2 engineers + 1 Vietnamese tax-domain advisor. The wedge is operational, not algorithmic.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	We are a 3rd-party preparer, not a signing TaxApp service provider — explicitly allowed under Law on Tax Admin 2025
Ethical — no harm / dark patterns	✅	Saving micro-shops from a bookkeeper bill they can’t afford
Market exists (evidence above)	✅	5M households, 49% in pain per VCCI
1–5 person team can build this	✅	2 engineers + 1 domain advisor for v1
Launchable with <$50K / ₹40L	✅	~$25K for v1 (Zalo OA setup, OCR fine-tune, infra, one tax advisor 3-month contract)

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	18/20	Quarterly recurring pain, gov-imposed, hard deadline, public apology from regulator. Hair-on-fire.
Demand evidence	15	13/15	VCCI 49%, kế toán dạo at ₫4M/mo, MISA chasing 2M households free. Hard signals on all three legs.
Build feasibility	15	12/15	10–14 weeks v1. OCR + Vietnamese fine-tune + eTax Mobile adapter is fiddly but not novel.
Distribution clarity	15	12/15	Zalo OA + named Facebook groups + kế toán dạo affiliate + tax-office partnership. Concrete first 100.
Revenue mechanics	15	12/15	₫149K floor maps to ~1 day of small-shop revenue. 11K customers to $1M ARR from a 500K reachable pool.
Time to first revenue	10	8/10	First paid Q3/2026 cycle (Jul-Sep) means revenue in 6–10 weeks of launch. Pre-sale possible.
Defensibility	10	5/10	Workflow lock-in (own the chat history & tagging memory) and Zalo OA approval friction. Copyable by a well-funded local in 6–9 months — speed-and-focus moat only.
Total	100	80/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (OCR + LLM + Zalo OA + e-tax integration) · operations-heavy (Cục Thuế relationships, bookkeeper recruitment, in-person market activation) · domain-expertise-required (Vietnamese tax code on hộ kinh doanh activity buckets — non-Vietnamese founder will get this wrong)

Key assumptions to validate (3–5)

1. Assumption: Tier-2 households (200M–3B VND) will pay ₫149K/mo for a Zalo-first filer. How to test: 50 owner WhatsApp/Zalo interviews in 3 Hà Nội traditional markets and 2 HCMC nail-salon clusters. Quote price, take ₫50K deposits.
2. Assumption: The till-tape OCR works on real-world Vietnamese receipts at >85% line accuracy. How to test: Collect 500 real photos from interviewees, run candidate models (Gemini, Qwen2-VL, fine-tuned local), measure activity-bucket classification recall.
3. Assumption: eTax Mobile accepts a draft prepared off-platform that the owner just confirms. How to test: Build a stub flow, file one real Q2 2026 declaration end-to-end for a friendly household, confirm the receipt.
4. Assumption: Cục Thuế will tolerate a 3rd-party prep layer — not treat it as an unauthorized TaxApp provider. How to test: Direct meeting with Hà Nội Cục Thuế in week 2; get a written non-objection or registration path.
5. Assumption: Kế toán dạo will affiliate-resell instead of compete. How to test: 20 bookkeeper Zalo-DM cycles offering 30% rev-share for the first year; measure conversion.

Risk flags

1. Regulatory risk: Vietnam could formalize a “TaxApp Service Provider” license requiring us to be locally licensed. Mitigate by registering as a Vietnamese LLC from day one and applying for the bookkeeping-services license track if/when defined.
2. Platform dependency: Heavy Zalo-OA + ZNS reliance; rate limits or pricing changes from VNG (Zalo’s parent) could squeeze unit economics. Mitigate by maintaining a Facebook-Messenger and SMS fallback channel.
3. Government competition: Cục Thuế itself could ship a smarter eTax Mobile in 12–18 months — they have political motive to fix the optics. Window is real but bounded. Mitigate by becoming the unofficial “ground-truth dataset” for the gov’s own efforts (offer aggregate, anonymized analytics).
4. Bank-statement access: VietQR per-bank API access varies; some banks require BIN-level partnership. Worst case fall back to user-forwarded SMS, which is messier but legal.

14. Structured verdict

Score:                  80/100
Verdict:                STRONG GO
Confidence:             Medium
Best-fit builder:       Vietnamese technical co-founder + Vietnamese tax-domain advisor; 2 engineers + 1 ops/partnerships hire in Hà Nội or HCMC
Time to revenue:        6–10 weeks (Q3 2026 cycle is the pre-sale)
Capital to launch:      $25–35K (Zalo OA, OCR fine-tune compute, infra, 3-mo advisor retainer)
Top 3 assumptions to validate first:
  1. 50 owner interviews → ≥15 ₫50K deposits = real WTP
  2. 500-receipt OCR test → ≥85% activity-bucket recall on real Vietnamese till tapes
  3. Hà Nội Cục Thuế meeting → written non-objection on 3rd-party prep layer
Kill criteria:
  - Abandon if <10% of 50 interviewed households put down a ₫50K deposit
  - Abandon if Cục Thuế signals an exclusive-licensing track for prep tools within Q3 2026
  - Abandon if MISA or VIN-HOADON ships a near-identical Zalo-first filer before our v1 (track quarterly)

15. Next step — 1-week validation sprint

• Day 1–2: Pull 200 hộ kinh doanh contacts from Hà Nội Đồng Xuân market + HCMC Bến Thành nail-salon ring via a local market-association contact; line up 30 Zalo voice calls.
• Day 3–4: Run the 30 calls in Vietnamese. Show a 90-sec Loom: “snap till tape → bot drafts Q3 declaration → tap-submit in eTax Mobile.” Quote ₫149K/mo. Take ₫50K refundable pre-orders.
• Day 5: Meet Hà Nội Cục Thuế officer who runs the local Zalo support tier; ask the explicit licensing question and offer to share aggregate anonymized Q3 friction data.
• Day 6: Collect 500 till-tape photos from the 30 interviewees; benchmark Gemini-Vietnamese and Qwen2-VL on activity-bucket recall.
• Day 7: Go / no-go on: (a) ≥15 pre-orders, (b) Cục Thuế signals “no objection,” (c) OCR ≥85%. If 2 of 3 pass, build v1 toward 31 July 2026 Q2 deadline shadow-launch, real revenue Q3 deadline (31 Oct 2026).

Falsifiable outcome: ≥15 of 30 pre-orders + Cục Thuế non-objection + ≥85% OCR. Anything less, drop or rescope.