NCMatch — cClassTrib classifier for Brazil’s contadores

1. One-liner

Upload a client catalog and get every SKU pre-tagged with the cClassTrib and CST codes Brazil’s new NF-e demands.

2. Trend signal — why now?

Brazil’s consumption-tax reform (LC 214/2025) went live 1 January 2026. From that date every NF-e and NFC-e in the Normal Regime must carry three new fields on every line item: cClassTrib (6-digit code from a national table mapping the item to one of fifteen LC 214/2025 annexes), CST-IBS/CBS (situation code), and cCredPres (presumed-credit class). Receita Federal’s Informe Técnico 2025.002 finalised the table in Q1 2026. Hard validation against rejection 1024 was originally scheduled for April 2026, was postponed under pressure, and is now reset for August 2026 mandatory-fields enforcement.

What this means in practice: every retailer, distributor and industry under Lucro Presumido or Lucro Real has to reclassify their entire product master — typically 500-5,000 SKUs each — before August. The mapping is not a one-to-one NCM → cClassTrib lookup. The same NCM splits across multiple cClassTrib codes depending on product description and intended use (the canonical example, cited by Receita-aligned blogs, is NCM 9619.00.00 covering both menstrual absorbers — cClassTrib 200013, zero rate — and disposable diapers — cClassTrib 200035, 60% reduction).

Brazilian trade press is already calling this “uma tarefa ingrata e manual” — a thankless manual task — for the country’s ~80,000 small accounting firms. The portal Contabeis.com.br and Sitecontabil.com.br are both publishing “27 most-rejected cClassTribs” lists. Enterprise tools (Avalara, Tax Radar, Taxcel, Conformidade Fácil) exist but are priced for ERP-integrated mid-market and above. The long tail of small contadores is doing this in spreadsheets, evenings and weekends.

Provenance:

• Signal 1 (demand): “reclassificação de dezenas de milhares de mercadorias cadastradas representa uma tarefa ingrata e manual” — Contabeis.com.br 2026-05 article on contador armadilhas — https://www.contabeis.com.br/artigos/76558/ — 2026-05
• Signal 2 (feasibility): Receita Federal Informe Técnico 2025.002 publishes machine-readable cClassTrib / CST / cCredPres tables tied to LC 214/2025 articles — https://www.nfe.fazenda.gov.br/portal/listaConteudo.aspx — 2026-04
• Signal 3 (economic): Taxcel, Tax Radar, Avalara, Conformidade Fácil all launched cClassTrib tooling in Q1 2026; ABES and CFC (Conselho Federal de Contabilidade) publishing guidance; GestãoClick entrepreneur survey shows 26.5% of PMEs cite invoice errors as top tax-reform concern — https://www.contabeis.com.br/artigos/76321/ — 2026-Q1 Category: Regulatory arbitrage

3. The opportunity

A new mandatory field was added to every electronic invoice in Brazil. The country’s tax authority gave companies a 7-month penalty grace period and then promised hard rejection. The classification is too context-dependent for a static lookup table, too complex for off-the-shelf ERP modules, and too tedious to do by hand at scale. Enterprise vendors are charging R$ 5,000+/month for the answer. The other 80,000 small accounting firms — each carrying 30-300 PME clients with thousands of SKUs apiece — are stuck.

The gap is shaped like a contador’s spreadsheet. They don’t have integration budgets. They don’t have RFI cycles. They have an Excel of their client’s products, two months to ship, and a panic email from the client’s COO. NCMatch ingests that spreadsheet, returns a classified spreadsheet with citations to the relevant LC 214/2025 article, and flags the ambiguous 5-10% for review.

The incumbent isn’t another SaaS — it’s the contador’s intern with a PDF of the Receita table open in another tab.

4. Target market

• Primary customer: Owner-operator and 2-15 staff Brazilian accounting firms (“escritórios contábeis”) serving 30-300 PME clients under Lucro Presumido or Lucro Real. Tier-1 and Tier-2 cities. Use Domínio, Alterdata, Contmatic Phoenix or SCI as the host accounting/tax system.
• Why they buy: Each of their PME clients sends a product master spreadsheet. The contador now owes correctly-coded cClassTrib + CST-IBS/CBS + cCredPres on every line. One wrong code per SKU = NF-e rejection at the cliente’s checkout once August 2026 mandatory enforcement hits = a livid phone call. Their alternative is paying R$ 5K+/month for Avalara-tier tools or hand-coding it.
• Rough TAM reasoning: CFC (Conselho Federal de Contabilidade) counts ~80,000 active accounting offices in Brazil. ~40-50% serve Normal Regime clients with goods movement. Address 30,000 firms × R$ 4,800/yr average → R$ 144M TAM. Capture 2-3% = R$ 3-4M ARR.
• Why now for them: Receita Federal moved the IBS/CBS mandatory-fields date to 1 August 2026. From that date NF-e/NFC-e without proper cClassTrib gets rejected. Every contador has 8-10 weeks of evening shifts ahead of them.

5. Product sketch (MVP)

• Catalog upload — paste a Google Sheet link or upload XLSX/CSV with the client’s products (NCM, descrição, unidade, CFOP, additional fields).
• Per-item classification — for each SKU, output cClassTrib (6-digit), CST-IBS, CST-CBS, cCredPres, plus a one-line reason citing the LC 214/2025 article and annex.
• Confidence + review queue — rows scored High / Medium / Low. Low-confidence items routed into a tight review queue with the top 3 candidate codes shown side-by-side.
• Multi-client workspaces — one workspace per PME client; product catalogs and audit logs scoped per client; “share with cliente” read-only view.
• Output formats — XLSX matching Domínio / Alterdata / Phoenix import layouts, plus a CSV → XML group for direct NF-e emitter integration.
• Change tracking — when the official cClassTrib table is updated, NCMatch flags which of a contador’s already-classified items are affected and re-runs only those.
• Audit defense pack — per item, an exportable PDF “memo” citing the legal basis used, for the inevitable fiscal challenge after the transition.
• WhatsApp alerts — table update? rejection on a previously-classified item? Contador gets a ping.

6. AI angle — what’s load-bearing

Classification is a three-input problem: NCM, product description (free text in Portuguese, often abbreviated or misspelled), and intended use. A lookup table fails because the same NCM splits across cClassTribs. A rules engine fails because LC 214/2025 has fifteen annexes with hundreds of exceptions written in legal prose. An LLM-with-retrieval over the LC + the Receita technical notes is the only tractable approach today — it can read “absorvente higiênico fem 10un” and “fralda descartável tamanho M” and route them to different annexes despite the same NCM.

Strip the AI and you have either a static lookup (wrong) or a human (slow). Both are the status quo. AI is the only reason this becomes a one-day job instead of a six-week one.

The non-trivial work: a retrieval-augmented setup grounded in the official RFB tables, the LC 214/2025 annexes and the published technical notes (which get amended monthly). Every answer has to cite the source so the contador can defend it later — no ungrounded hallucinations allowed in a fiscal context.

7. Localization angle

This is a Brazil-only play by construction — the entire product is a wrapper around one Brazilian regulation. But there are sub-localizations to exploit:

• Pricing in BRL with Pix billing. Contadores in Tier-2 cities expect to pay R$ 199-R$ 999/mo via boleto or Pix, not USD 50/mo via Stripe.
• Pt-BR voice. Product, support and onboarding in Portuguese. Audit PDFs and citations in Brazilian Portuguese legalese the auditor expects.
• Tax-software import formats. Match the column-by-column XLSX layouts that Domínio, Alterdata, Contmatic Phoenix and SCI expect — these four cover ~70% of small accounting firms.
• State-level CFOP overlays. Future expansion: state-specific cClassTrib edge cases (the IBS half splits 27 ways).

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

• Pricing tiers:
  • Solo — R$ 199/mo, up to 5 client workspaces, 5,000 items/mo classified.
  • Escritório — R$ 499/mo, up to 40 workspaces, 50,000 items/mo.
  • Bureau — R$ 1,499/mo, up to 200 workspaces, 250,000 items/mo, dedicated support.
• ACV blended: ~R$ 7,200/yr (USD ~1,440).
• R$ 1M ARR path: ~140 paying firms — achievable in 6-9 months given the time pressure to August 2026.
• R$ 5M ARR path: ~700 paying firms — ~1% of the addressable 70K. Plausible by end-2027 if NCMatch becomes the de-facto small-contador tool, expanded to handle the 2027 split-payment phase and 2028 ICMS/ISS phase-out.
• Expansion path: monthly table-change re-validation (sticky), per-state CFOP add-ons, audit-defense memo generation (per-PDF micro-charge), then upmarket into the split-payment reconciliation 2027 wave.

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

• Facebook Groups for contadores — there are 5-10 large pt-BR Facebook groups (10K-80K members each: “Contadores do Brasil”, “Profissionais da Contabilidade”, “Reforma Tributária Brasil”). Post weekly “27 most-rejected cClassTrib codes” content + free table-update bulletins; convert to trial via DM.
• CRC (Conselho Regional de Contabilidade) state-chapter newsletters — paid placements run R$ 800-2,000 per regional newsletter, hitting tens of thousands of registered accountants per state. Run in São Paulo, Minas Gerais, Paraná, Rio first.
• YouTube partners — 5-10 contador YouTubers have audiences of 20K-100K each (“Contabilizando”, “Tributo Aprendido”, “Reforma na Veia”). Sponsored deep-dives with a coupon code; expect 1-2% conversion.
• Cold outreach to top-50 small-firm ERPs — Domínio, Alterdata, Phoenix, SCI, ContaAzul, Onvio. Pitch a no-cost CSV import partnership. Even one named integration lifts trust hard.
• WhatsApp distribution lists — contadores share tools in WhatsApp groups. Build a one-link “update bulletin” subscriber (free) that drops twice a month: 30-50% of subscribers convert to paid over 6 months once August hits.

The pitch line is short: “August 1st your clients start getting NF-e rejections. Upload your worst spreadsheet here and we’ll classify it free.”

10. Build complexity — justification

Medium. Stack is mostly off-the-shelf: Next.js / Node / Postgres / S3 + Anthropic or OpenAI for the classification with retrieval over the public LC 214/2025 annexes and RFB tables. The classification quality bar is high (and requires real domain work — building the eval set, fine-tuning prompts, handling edge cases per annex) but the surface is small. Estimated 10-14 weeks for a 2-person team to a credible v1; the hard work is the evals and the iterative prompt + retrieval tuning against a held-out gold-standard set of 5K manually-classified items.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Tool advises on public regulation; no enforcement role.
Ethical — no harm / dark patterns	✅	Helps small businesses comply with mandatory regulation.
Market exists (evidence above)	✅	Multiple incumbents launched, regulation goes mandatory Aug 2026.
1–5 person team can build this	✅	2 engineers + 1 fiscal-domain advisor for 10-14 weeks.
Launchable with <$50K / ₹40L	✅	Inference + infra ≤ $1K/mo at first 100 customers; gold-standard eval-set sourcing ~$5K.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Hard rejection coming Aug 2026; contador loses billable hours to manual work; client churn risk if NF-e starts bouncing. Hair-on-fire by July.
Demand evidence	15	13/15	Multiple competing tools already launching; published RFB tables; specific complaint quotes in trade press; GestãoClick survey numbers. A skeptic would nod.
Build feasibility	15	11/15	Standard web stack + LLM + retrieval. Real work is the eval-set construction and accuracy tuning — non-trivial but bounded.
Distribution clarity	15	11/15	Concrete channels (CRC newsletters, Facebook groups, YouTube contadores). Conversion math reasonable, not guaranteed.
Revenue mechanics	15	12/15	Pricing benchmarked vs Avalara-tier and small-firm ERPs. R$ 1M ARR needs ~140 firms — credible given panic-cycle.
Time to first revenue	10	8/10	Pre-sale offer to 50 contadores via Facebook groups can produce paying pilots in <6 weeks of beta.
Defensibility	10	6/10	Workflow lock-in (audit logs, change tracking) + brand in the contador niche. The eval-set is the soft moat. Copyable by a funded competitor in 9-12 months.
Total	100	78/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

A Portuguese-speaking technical founder paired with a Brazilian fiscal-domain advisor (a senior contador or tax lawyer who’s read LC 214/2025 cover to cover) is the ideal pair. The advisor isn’t optional — they author the eval set.

Key assumptions to validate (3–5)

1. Assumption: Small contadores will pay R$ 499/mo for a tool that classifies their clients’ SKU catalogs. How to test: Run a paid pilot landing page targeting 5 specific Facebook groups; collect 30 R$ 100 deposits-toward-pilot in 14 days.
2. Assumption: A retrieval-augmented LLM on LC 214/2025 + RFB tables can hit ≥92% precision on a held-out gold-standard set of 5,000 contador-classified items. How to test: Build the gold set with a paid advisor (2 weeks), build the retrieval prompt (2 weeks), measure.
3. Assumption: Receita Federal does not hold the August 2026 mandatory date again (or if they do, only by 3-4 months — not 18). How to test: Track CG-IBS / RFB releases monthly; watch for political pushback from CNC and Fecomercio.
4. Assumption: The four major small-firm ERPs (Domínio, Alterdata, Phoenix, SCI) will not bundle a free equivalent in time. How to test: Read each vendor’s product roadmap; talk to 10 customers per ERP about what they’re being told.

Risk flags

1. Regulatory risk: The cClassTrib table is amended monthly during 2026. A v1 that doesn’t auto-track table changes ships stale answers within weeks. Must build the table-change pipeline from day one.
2. Platform dependency: Heavy reliance on a single LLM provider’s quality on Portuguese fiscal prose. Mitigate by benchmarking 2-3 providers and being able to swap.
3. Market timing: If Receita postpones mandatory enforcement past 2026 again, the urgency drops and the price tolerance with it. The Aug 2026 date is the wedge — if it slips to 2027, this still works but valuation/urgency falls.
4. Incumbent bundling: Avalara, TOTVS, Conta Azul could bundle this for free to defend ERP seats. Defence: speed, focus, contador-shaped UX, and pricing the long tail nobody else wants.

14. Structured verdict

Score:                  78/100
Verdict:                GO
Confidence:             High
Best-fit builder:       Portuguese-speaking technical founder + senior Brazilian contador or fiscal lawyer as co-founder/advisor
Time to revenue:        6-8 weeks from beta to first 20 paying firms; 4-6 months to R$ 1M ARR
Capital to launch:      R$ 80K-150K ($16K-$30K) — eval-set sourcing + inference credits + light marketing
Top 3 assumptions to validate first:
  1. ≥30 pre-pay R$ 100 deposit signups in 14 days from Facebook groups (demand)
  2. ≥92% precision on 5K-item gold standard set (accuracy)
  3. Domínio / Alterdata / Phoenix do not announce a free bundled equivalent before launch (competitive)
Kill criteria:
  - Abandon if <8% conversion from free trial to paid after 60 days on 100+ trial firms.
  - Abandon if accuracy stays under 88% on the gold-standard set after 6 weeks of tuning.
  - Abandon if Receita Federal pushes mandatory enforcement past Q2 2027 (urgency window collapses).

15. Next step — 1-week validation sprint

• Day 1–2: Build a 2-page landing site in pt-BR: “Classificador cClassTrib para escritórios contábeis — uploads grátis até 31 de julho”. Post into 5 large contador Facebook groups + LinkedIn pt-BR with a personal note from the founder.
• Day 3–4: Run a free classification of 3 contador’s product catalogs by hand (or with a thin Sonnet wrapper) — measure: (a) how long it takes them to send the file, (b) how many items they have, (c) what they currently pay for the alternative, (d) whether they would pay R$ 499/mo.
• Day 5–6: Build a 200-item gold-standard set with a paid fiscal advisor; run a basic retrieval+LLM classifier; measure precision and recall.
• Day 7: Decide go / no-go based on: (a) ≥10 contadores submit catalogs in week (demand), (b) ≥3 verbally commit to R$ 499/mo pricing (willingness), (c) prototype hits ≥85% precision on the 200-item set (feasibility).

Falsifiable outcome: under 10 catalog submissions in 7 days, or under 3 verbal commitments at R$ 499/mo, or under 85% precision on a small gold set → pause and reconsider GTM angle, do not build.