KhaiSẵn — quarterly filing desk for Vietnamese shop owners

1. One-liner

Reconciles a Vietnamese shop’s sales against its e-invoices on Zalo and hands back a ready-to-file quarterly tax declaration.

2. Trend signal — why now?

On 1 January 2026 Vietnam abolished the lump-sum (thuế khoán) tax regime for all business households. For decades ~2 million of the country’s 5.2 million registered households paid a flat, negotiated annual figure and kept no books. As of this year every household in the 500M–3B VND revenue band must self-declare actual revenue quarterly, keep simple accounting books, and issue e-invoices — and a brand-new penalty framework (Decree 310/2025, effective 16 Jan 2026) fines invoice/declaration violations VND 10–20M with suspension in severe cases.

The fear is documented and large:

• A VCCI survey (Feb–Apr 2026) found 73% of households cite lack of tech/accounting skills, 53% say the declaration procedures are too complicated, 49% worry about the time cost of managing invoices/documents, and 32% fear data-security/mistake risk. 81% reported falling 2025 revenue — so nobody has cash to hire an accountant.
• Traders are literally closing stalls in HCMC over “digital tax fears,” and many are insisting on cash to stay off the radar.
• The government is scrambling: a pilot online declaration portal, Hanoi tax offices running three-tier Zalo networks to hand-hold owners, HCMC segmenting its 363,000 households into support tiers.
• Big vendors (MISA, Sapo 6870, KiotViet, Viettel Tendoo, VNPT) are giving away POS + e-invoice + auto-ledger + digital signature to win logos.

What changed: the compliance obligation is new (12 months old), Vietnamese-capable LLM inference is now cheap enough to reconcile and explain filings at ₫-scale per shop, and the tax authority already holds each household’s e-invoice data — so a reconcile-before-you-file layer is finally buildable.

Provenance:

• Signal 1 (demand): VCCI 2026 survey — 73% lack tech/accounting skills, 53% find declaration too complex, 49% fear document-management time; traders closing HCMC stalls over tax fears — https://en.sggp.org.vn/many-market-traders-in-hcmc-close-stalls-amid-new-digital-tax-fears-post118073.html + https://vietnamnet.vn/en/from-manual-to-digital-how-small-businesses-will-survive-the-2026-tax-shift-2473996.html — 2026
• Signal 2 (feasibility): Cheap Vietnamese-capable LLM inference (e.g. DeepSeek-V4 at ~$0.09/M input tokens) makes per-shop reconciliation + plain-language explanation economical; gov e-invoice system already holds each household’s invoice data to reconcile against — https://inference.net/content/llm-api-pricing-comparison/ — 2026
• Signal 3 (economic): Money moving hard — MISA giving free eShop to 2M households, Sapo 6870, Viettel Tendoo, KiotViet all funding this transition; incumbent accounting service fees run 500K–3M VND/mo, too pricey for a corner shop — https://www.vietnam.vn/en/misa-tang-mien-phi-phan-mem-cho-2-trieu-ho-kinh-doanh + https://vietanlaw.com/accounting-services-in-vietnam/ — 2026 Category: Regulatory arbitrage

3. The opportunity

The incumbents are fighting over the wrong layer. MISA, Sapo, KiotViet, Viettel and VNPT are all giving away the commodity — a POS that rings sales, issues e-invoices, and auto-builds ledgers. That’s a solved, subsidized race a solo builder cannot win.

But a subsidized POS assumes the owner rings every sale through the POS. The 73%-can’t-do-tech / elderly-trader / cash-insisting majority won’t. And even the ones who adopt a POS hit the actual moment of terror the incumbents don’t own: “It’s the end of the quarter. Does what I’m about to declare match what the tax authority already sees from my e-invoices? Did I miss a cash sale, a purchase invoice, a threshold? If I’m wrong, I get fined.”

That reconcile-and-confidence step is not a POS feature. It’s a focused product: pull the household’s e-invoice data + their messy sales/expense records (photos, Zalo forwards, a POS export if they have one), reconcile the two, compute the exact VAT+PIT owed for their band, flag every gap before filing, and produce a ready-to-submit quarterly declaration with a one-paragraph plain-Vietnamese explanation of what they owe and why. A safety layer sitting on top of whatever free POS the owner already got.

4. Target market

• Primary customer: A Vietnamese business household (hộ kinh doanh) in the 500M–3B VND/year band — the grocery, phở shop, hardware store, salon, market stall now forced onto quarterly declaration + simple books + mandatory e-invoice. Owner is 35–60, one to a few staff, keeps records in a notebook or Zalo, is not an accountant and is scared of Decree 310 fines. Secondary buyer: the local bookkeeper / small tax-agent who suddenly inherited 30–80 of these households and needs to file each one’s quarterly declaration without drowning.
• Why they buy (their words): “business is slow, revenue is down, now we have to raise prices because taxes are slowing sales” (Ms. Ngan, shop owner); 53% say the procedures are too complicated; 32% fear making a mistake. They don’t want to become accountants — they want to not get fined and not overpay.
• Rough TAM reasoning: ~2M households moving off lump-sum; the mandatory-quarterly 500M–3B band is a large fraction. Even 1% of 1M reachable households at a modest annual fee is a healthy bootstrap business; the tax-agent segment multiplies each sale.
• Why now for them: First quarterly declarations under the new regime fall due in 2026. The fear is acute this year, the penalty regime is brand-new, and the government’s own Zalo hand-holding proves the channel and the desperation.

5. Product sketch (MVP)

• Zalo-native intake: owner forwards a photo of the day’s sales notebook, a receipt, or a voice note (“bán 3 bao gạo, 450 nghìn”) — no app to learn.
• E-invoice pull + reconcile: ingests the household’s issued e-invoices (upload/export) and matches them against recorded sales to surface unrecorded cash sales, missing purchase invoices, and duplicates.
• Band-correct tax math: computes VAT + PIT for the shop’s sector and revenue band (e.g. grocery = 1% VAT + 0.5% PIT on revenue), tracks the 500M exemption and 3B threshold, warns before a band change bites.
• Quarterly declaration, ready to file: produces the filled declaration values matching the tax portal’s form, with a plain-Vietnamese “you owe ₫X because…” summary the owner actually understands.
• Pre-file gap alerts: a “before you submit” checklist — “3 cash sales this quarter have no matching record; the portal already sees ₫Y in e-invoices you haven’t booked.”
• Simple compliant book: maintains the Finance-Ministry-standard simple ledger passively from what’s forwarded, so the mandatory book exists without the owner “doing accounting.”
• Deadline nudges: Zalo reminders before each quarterly filing date.
• Agent console (v1.1): the bookkeeper view — a queue of all their household clients, each with a red/green “ready to file / needs a fix” status.

6. AI angle — what’s load-bearing

AI does three things a rules engine alone can’t: (1) reads the mess — Vietnamese handwritten notebooks, receipt photos, voice notes, informal Zalo shorthand — into structured line items; (2) reconciles ambiguous records against e-invoices (is “gạo 450” the same sale as the ₫450,000 invoice?) and reasons about what’s missing; (3) explains the declaration in plain Vietnamese to a scared non-accountant. Remove the AI and you’re left with a form the incumbents already give away free — the whole product is turning informal human records into a defensible, portal-matching filing. AI is the product, not a sticker.

7. Localization angle

This is a Vietnam-only play by construction — it exists because of a specific 2026 Vietnamese regulation. Localization is total:

• Language/script: Vietnamese OCR + voice, informal trader shorthand, plain-Vietnamese explanations.
• Channel: Zalo-first (the de facto business-comms OS in Vietnam; the tax offices themselves use it), not WhatsApp or email.
• Payment rails: priced in ₫, collectible via VietQR/bank transfer / Zalo.
• Regulatory: the exact Decree 70/123/310 bands, rates, thresholds, forms, and the pilot declaration portal. A generic global bookkeeping tool cannot do any of this; the regulation is the moat’s raw material.

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

• Pricing: household tier ₫99,000/month (~$3.80) — deliberately below a ₫500K–3M/mo accountant, framed as “fine insurance.” Tax-agent tier ₫1.5–3M/month per firm for the batch console (covers 30–80 clients).
• ACV: ₫1.19M/year ($45) per household; ₫30M/year ($1,150) per tax-agent firm.
• Rough math to $1M ARR: ~22,000 households at $45/yr = ~$1M. Or a blend: 8,000 households ($360K) + 550 tax-agent firms at $1,150 ($630K) ≈ $1M. Against a ~1M-household addressable band, ~2% penetration.
• Rough math to $5M ARR: ~100K households, or a heavier tax-agent mix (2,500 firms × $1,150 = $2.9M + 60K households = $2.7M). Requires the agent channel to compound — each firm onboards its whole book.
• Expansion path: add loan-readiness exports (banks want the clean books), annual settlement filing, digital-signature resale, and a “growing past 3B → become an enterprise” upgrade path as households scale.

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

• Ride the tax offices’ own Zalo networks. Hanoi and HCMC tax departments already run three-tier household Zalo groups. Show up where the panic is: offer a free “will my Q1 declaration get flagged?” reconciliation check in those groups and local ward (phường) trader groups. The pain is acute and communal — one saved fine sells the group.
• Recruit tax-agents / bookkeepers as the wedge. Each newly-overloaded đại lý thuế brings 30–80 households. Cold-DM and visit the tax-agent listings and accounting-service pages in the 3–4 biggest cities; sell the batch console; they become resellers with skin in the game.
• Market-and-ward canvassing. Traditional markets (chợ) are dense clusters of the exact customer. A ₫20K reconciliation demo done live on one stall’s records, in Vietnamese, converts the neighbors. Operations-heavy but cheap and fast.
• Piggyback the free-POS crowd. MISA/Sapo/KiotViet gave a shop a POS but not confidence it filed correctly — target their existing users with “reconcile what your POS reports before you file.”

10. Build complexity — justification

Medium. Off-the-shelf: Vietnamese OCR/voice + LLM, Zalo Official Account API, standard web/backend. Custom work: the regulation engine (band logic, sector rates, thresholds, the exact quarterly form fields) and the reconciliation logic matching informal records to e-invoices — plus keeping current as the Finance Ministry issues guidance through 2026. A small team ships a usable Zalo reconciliation + declaration for one or two sectors (grocery, food) in ~3–4 months, then widens sector coverage. Not Low — the domain rules and Vietnamese-language mess make it real work — but nothing research-grade.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Filing-assist tool; gov actively invites tech help. Not itself a licensed tax-agent — must not misrepresent as one.
Ethical — no harm / dark patterns	✅	Helps owners comply and not overpay; opposite of a dark pattern.
Market exists (evidence above)	✅	2M households forced onto declaration; VCCI survey quantifies the fear.
1–5 person team can build this	✅	Medium build, ~3–4 months to v1.
Launchable with <$50K / ₹40L	✅	Off-the-shelf APIs; ops-heavy canvassing is cheap in-country.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Hair-on-fire: legal obligation + new fines + first filings due this year. Traders closing stalls over it.
Demand evidence	15	14/15	Quantified VCCI survey, named owner quotes, mass vendor money, gov Zalo hand-holding. Multiple independent 2026 signals.
Build feasibility	15	11/15	Off-the-shelf AI, but the regulation engine + Vietnamese-mess reconciliation + moving guidance is real work.
Distribution clarity	15	11/15	Zalo groups, tax-agent resellers, market canvassing are concrete; but reaching non-tech elderly owners at scale is slow and ops-heavy.
Revenue mechanics	15	11/15	Low ₫ price fits thin wallets and undercuts accountants; needs volume, and volume against subsidized free POS is the risk.
Time to first revenue	10	8/10	Pain is now; a paid reconciliation check can sell within weeks of a working Zalo bot.
Defensibility	10	4/10	Regulation knowledge + agent relationships are soft moats; well-funded incumbents (MISA/Viettel) could bolt this on. Speed and focus only.
Total	100	76/100

13. Qualitative modifiers

Founder-fit tags

domain-expertise-required (Vietnamese tax/regulatory fluency is non-negotiable) · operations-heavy (Zalo groups, market canvassing, agent recruiting are ground-game).

Key assumptions to validate (3–5)

1. Assumption: 500M–3B households will pay ~₫99K/mo for filing confidence rather than rely on the free POS + gov portal. How to test: Offer a paid reconciliation check to 50 households in two ward Zalo groups; measure paid conversion.
2. Assumption: Tax-agents will adopt a batch console and bring their whole book. How to test: Pitch 15 đại lý thuế / bookkeepers in HCMC + Hanoi; get 3 to run a real quarter of clients through it.
3. Assumption: Reconciliation is accurate enough on real Vietnamese notebooks/receipts to be trusted near a fineable filing. How to test: Run 100 real records; measure precision/recall on catching missing sales vs a human accountant.
4. Assumption: The gov portal’s form fields are stable and machine-fillable enough to auto-produce a submittable declaration. How to test: File a real quarterly declaration for 5 friendly households end-to-end.

Risk flags

1. Incumbent bolt-on: MISA/Viettel/Sapo already own distribution and could add reconciliation to their free tier. Mitigation: go deep on the fearful non-POS segment + tax-agent channel they underserve; move fast.
2. Regulatory churn: Finance Ministry guidance is still landing through 2026; forms/rates/bands can shift and break the engine. Mitigation: treat the rules as a maintained data layer, not hard-coded logic.
3. Willingness-to-pay in a shrinking economy: 81% of households report falling revenue; ₫99K/mo is still a real ask. Mitigation: price as fine-avoidance, sell via agents who bundle it.
4. Reachability: the most-fearful owners are the least digital — canvassing cost per customer could stay high. Mitigation: lean on the tax-agent multiplier, not direct-to-elderly-owner.

14. Structured verdict

Score:                  76/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Vietnamese founder (or Vietnam-based) with tax/accounting domain fluency + local ground game
Time to revenue:        6–10 weeks after a working Zalo reconciliation bot
Capital to launch:      ₫250–600M ($10–24K)
Top 3 assumptions to validate first:
  1. Households pay ~₫99K/mo for filing confidence — 50-household paid reconciliation test in ward Zalo groups
  2. Tax-agents adopt the batch console and bring their book — 3 firms run a real quarter
  3. Reconciliation is accurate on real Vietnamese records near a fineable filing — precision/recall vs a human on 100 records
Kill criteria:
  - Abandon if <10% of 50 households pay after a free reconciliation check
  - Abandon if MISA/Viettel ship equivalent reconciliation in their free tier before v1
  - Abandon if reconciliation precision on missing-sale detection stays below a level a tax-agent will stake a filing on

15. Next step — 1-week validation sprint

• Day 1–2: Sit in 3–4 ward/market trader Zalo groups and 2 tax-agent groups. Collect 20 real sets of household records (notebook photos, receipts, e-invoice exports). Confirm the exact quarterly form fields from the pilot portal.
• Day 3–4: Hand-run reconciliation + declaration for 10 of those households (concierge, no product yet). Show each owner their “you owe ₫X, and here are 2 sales you forgot” result on Zalo. Ask for ₫99K to do it next quarter.
• Day 5: Decide. Go if ≥6 of 10 owners say yes to paying and ≥1 tax-agent commits to piloting their book. Falsifiable: real money committed for a real filing, or not.