DesencalhaNF — NF-e rejection desk for Brazilian accountants

1. One-liner

Reads a rejected NF-e, names the cClassTrib/CST error, fixes the SKU catalog, and re-validates before SEFAZ blocks the sale.

2. Trend signal — why now?

Brazil’s consumption tax reform (LC 214/2025) put two brand-new taxes — IBS and CBS — onto every electronic invoice starting January 2026. The kicker isn’t “fill two more fields.” It’s that SEFAZ began validating internal consistency on 2 February 2026: if the Grupo UB block is present, the CST, cClassTrib, alíquotas and the math must all cohere — or the note is rejected and the sale is legally blocked until fixed. For normal-regime issuers (Lucro Real/Presumido) the hard, rejection-as-blocker deadline is 3 August 2026. The classification tables (cClassTrib, CST, cCredPres) only landed publicly in Nota Técnica RT 2025.002 in mid-2025, and the official “Conformidade Fácil” cClassTrib assistant didn’t ship until 29 January 2026.

The signal is loud and live, not speculative. Accountant and tributarista WhatsApp groups have been “flooded” since the start of the year with screenshots of rejections and “cadastro não encontrado” messages even from pros who followed the protocol (FENACON). The IBPT estimates 95% of Brazilian companies already make tax-calculation errors — a rate everyone expects to worsen during the transition. Tax-reform tooling is being called “the greatest revenue-scalability opportunity for accounting firms in decades,” and the fiscal-SaaS incumbents (Tecnospeed, Taxcel, Omie, Jettax, Fiscal.io) are all shipping reform features right now.

Direct quote from a fiscal-software CEO (Taís Baruchi), describing the trap plainly:

“Se essa combinação estiver incorreta, o destaque dos tributos também estará — e… a nota fiscal será rejeitada.”

Provenance:

• Signal 1 (demand): Accountant WhatsApp groups flooded with NF-e/NFS-e rejection screenshots; SEFAZ internal-consistency validation live since 2 Feb 2026; rejected note = blocked sale — https://fenacon.org.br/reforma-tributaria/contadores-relatam-instabilidade-para-emissao-de-nfs-e-nacional-receita-orienta-buscar-os-municipios/ and https://www.contabeis.com.br/noticias/74328/ibs-e-cbs-2026-erros-de-classificacao-que-podem-causar-rejeicao-de-notas-fiscais/ — 2026
• Signal 2 (feasibility): Full cClassTrib/CST/reject-code tables published in NT RT 2025.002; reject codes 1021/1024/1026 documented; LLM can map SKU→classification and decode reject XML cheaply — https://blog.tecnospeed.com.br/principais-rejeicoes-da-nf-e-com-ibs-e-cbs/ — 2025/2026
• Signal 3 (economic): Tax reform framed as decade-defining revenue opportunity for accounting firms; fiscal-SaaS incumbents actively shipping reform tooling — https://www.socialhub.pro/blog/reforma-tributaria-2026-escritorio-contabil-oportunidade-receita/ — 2026 Category: Regulatory arbitrage

3. The opportunity

The reform created a new, recurring failure mode that didn’t exist a year ago: a note that won’t transmit because the IBS/CBS classification is internally inconsistent. Free emitters (Sebrae, NFe+, Click NF-e) emit fine in the happy path — they do not diagnose why SEFAZ rejected reject-code 1024 (cClassTrib×CST incompatibility) on item 37 of a 400-SKU catalog, nor fix the underlying cadastro so it stops recurring. The official Conformidade Fácil assistant is a static lookup table; it tells you what a code means, not what’s wrong with your note.

That gap is the wedge. The incumbents sell you a better emitter or a tax calculator. Nobody sells the fix loop: ingest the rejected XML + reject code, diagnose root cause across the whole product registry, propose the corrected cClassTrib/CST per SKU, re-validate against SEFAZ rules, and remember the mapping so the same SKU never bounces again. A focused AI-first team can collapse what is currently a 30-minute manual hunt-through-tables (per rejection, per client, dozens of times a day across a firm) into a 30-second paste-and-fix.

4. Target market

• Primary customer: Small-to-mid escritórios contábeis (accounting firms) in Brazil — 5 to 40 staff, serving 50–400 SMB clients each. Secondary: in-house fiscal analysts at Lucro Presumido SMBs (retail, distribution, light manufacturing) that issue NF-e directly.
• Why they buy: In their words — rejections are now blocking client sales and the firm is the one on the phone at 6pm. “A nota foi rejeitada e o cliente não consegue faturar” is a billing-stopping emergency. The firm carries the reputational and SLA risk for every client’s blocked invoice.
• Rough TAM reasoning: Brazil has tens of thousands of registered accounting firms and millions of NF-e-issuing businesses. Even a narrow beachhead — the ~80,000 firms actively managing Lucro Presumido clients — at a few hundred reais/month is a multi-hundred-million-real category. We need a few thousand firms, not all of them.
• Why now for them: The Aug 3 2026 hard deadline plus the already-live Feb 2026 consistency checks mean the pain is present-tense and escalating through the year. The transition runs to 2033, so this isn’t a one-quarter spike — the classification surface keeps changing.

5. Product sketch (MVP)

• Paste-a-rejection: drop the rejected NF-e XML (or just the reject code + item), get a plain-Portuguese diagnosis: which field, which SKU, why SEFAZ bounced it.
• One-click fix proposal: the corrected cClassTrib/CST/alíquota for the offending item(s), with the LC 214/2025 article cited so the accountant can defend it.
• Catalog sweep: upload the client’s full product registry once; flag every SKU whose current classification will reject before it ever hits SEFAZ.
• SEFAZ-rule re-validation: runs the corrected note through the published NT 2025.002 consistency rules and shows green/red before re-transmit.
• Per-client SKU memory: the fixed mapping is saved per client, so the same item never bounces twice and new notes inherit the correct classification.
• Firm dashboard: which clients have open rejections, which catalogs are still un-swept, daily rejection count — the partner’s morning triage view.
• Change-watch: when a new technical note revises the tables, flag which saved SKUs are now affected.

6. AI angle — what’s load-bearing

Remove the AI and this is a spreadsheet of reject codes — i.e. what already exists and doesn’t solve the problem. The load-bearing work is (a) reading a messy rejected XML + free-text SEFAZ reason and localizing the actual cause to a specific item and field, and (b) classifying an arbitrary product description into the correct cClassTrib/CST given the LC 214/2025 treatment — a fuzzy mapping from “Parafuso sextavado M6 inox” to a tax classification that a lookup table can’t do because the catalog descriptions are non-standard. The model also drafts the citation-backed justification. The static tables and validators are inputs; the judgment that turns a rejection into a fix is the product.

7. Localization angle (if any)

This is the localization play — it cannot exist outside Brazil. Portuguese-first; built entirely around LC 214/2025, NT RT 2025.002, the cClassTrib/CST tables and SEFAZ reject codes. Distribution is WhatsApp-native (the channel where the pain is already being screamed about) and pricing in reais at a tier a small firm clears without procurement. A generic “tax AI” from outside Brazil has zero chance here — the regulatory surface is the entire product.

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

• Pricing: R$249/mo (solo/in-house issuer, single CNPJ) → R$699/mo (firm, up to ~50 client CNPJs) → R$1,499/mo (firm, up to ~200 CNPJs). Usage-metered overage on catalog SKUs swept.
• ACV: ~R$8,400 blended (firm tier weighting).
• Rough math to $1M ARR (~R$5.5M): ~650 firms on the R$699 tier ≈ R$5.4M ARR. Very reachable against tens of thousands of firms in acute pain.
• Rough math to $5M ARR (~R$27M): ~2,200 firms blended across tiers, plus the in-house SMB tier — requires becoming the default “rejection desk” referenced in accountant WhatsApp groups, and surviving the Aug 2026 deadline as the tool people already trust.
• Expansion path: ACV grows by CNPJs-under-management (firms add clients), by SKU volume swept, and by adjacent reform modules (split-payment reconciliation in 2027, NFS-e national-standard fixes, MEI/Simples cohort when their rejection deadline lands in 2027).

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

• Infiltrate the WhatsApp/Telegram accountant groups where rejection screenshots already circulate. Answer real rejection screenshots with a free correct diagnosis; drop the tool link. This is where the pain is being posted today.
• Reject-code SEO + a free public diagnoser: rank for “rejeição 1024 NF-e”, “rejeição 1026 IBS”, “cClassTrib errado” — high-intent queries a panicking analyst types at the moment of failure. Free single-note diagnosis, paywall the catalog sweep + memory.
• Conselho Regional de Contabilidade (CRC) and contábil-influencer channels: Brazil has a large, active accounting-creator scene on YouTube/Instagram already producing reform content. Sponsor 5–10 to demo a live rejection fix.
• Firm-level land-and-expand: one analyst at a firm fixes one rejection, shows the partner; partner rolls it across the client book. Seed via direct outreach to firms posting reform-prep content on LinkedIn.

10. Build complexity — justification

Medium. Off-the-shelf: LLM for diagnosis/classification, standard web stack, the published NT 2025.002 tables and SEFAZ validation rules as ground truth. Custom work: a reliable XML-rejection parser, the SKU→classification mapping engine with per-client memory, and keeping the rule set current as technical notes revise. No certificate-emission integration needed for v1 (we validate and hand back the corrected classification; the firm re-transmits in their existing emitter). A pair could ship a credible v1 in 10–14 weeks.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Helping correct classification; we advise, the firm transmits. No regulated activity.
Ethical — no harm / dark patterns	✅	Reduces erroneous filings; cites the law for every fix.
Market exists (evidence above)	✅	Live rejections, WhatsApp-group panic, hard Aug 2026 deadline.
1–5 person team can build this	✅	Medium build, 10–14 weeks to v1.
Launchable with <$50K / ₹40L	✅	Inference + web stack; main cost is domain expertise (a fiscal advisor).

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Rejection blocks the sale — billing-stopping, felt daily during transition. Hair-on-fire.
Demand evidence	15	12/15	Live rejections, WhatsApp-group reportage, IBPT 95%-error stat, incumbents shipping. Strong but indirect (no public revenue figure for this exact product).
Build feasibility	15	11/15	Tables/rules published; XML parsing + classification engine is real but not research-grade. 10–14 wks.
Distribution clarity	15	12/15	WhatsApp groups + reject-code SEO are named, high-intent, cheap. Conversion math credible.
Revenue mechanics	15	11/15	Pricing benchmarked vs R$99/mo emitters; firm tier reasonable; depends on firm-CNPJ expansion holding.
Time to first revenue	10	8/10	Pain is present-tense; free diagnoser → paid sweep funnel can convert in weeks.
Defensibility	10	5/10	Soft moat: per-client SKU memory compounds + firm workflow lock-in. But classification tables are public and incumbents are nearby. Execution + speed race.
Total	100	76/100

13. Qualitative modifiers

Founder-fit tags

domain-expertise-required · technical-heavy — needs a Brazilian fiscal/tax advisor co-founder or close advisor; the classification logic must be right or trust evaporates instantly.

Key assumptions to validate (3–5)

1. Assumption: Firms will pay R$699/mo for a fix-loop layer alongside their existing emitter (not demand it be free like the Conformidade Fácil assistant). How to test: 30 firm interviews + a pre-sale landing page with the price; measure deposit/intent conversion.
2. Assumption: AI classification of arbitrary catalog descriptions to cClassTrib/CST is accurate enough (>90%) that accountants trust it rather than re-checking everything. How to test: blind-test against a sample of 200 real SKUs classified by a senior fiscal analyst.
3. Assumption: The fix-loop is materially better than the free static assistant — i.e. accountants experience the diagnosis as faster, not just prettier. How to test: time-to-fix A/B with 10 accountants on live rejections.
4. Assumption: WhatsApp-group distribution converts at a workable rate without being flagged as spam. How to test: seed 5 groups with genuine free help, measure click-through and signups over 2 weeks.

Risk flags

1. Platform/regulatory dependency: the entire product is pinned to SEFAZ rules and NTs that revise frequently; a major table change is constant maintenance, and an official tool that adds a fix-loop could compress the moat.
2. Commoditization: classification tables are public; the defensibility rests on SKU-memory and firm lock-in, which must be built fast before a fiscal-SaaS incumbent bolts on the same feature.
3. Trust/liability: a wrong classification that gets a client fined kills word-of-mouth; positioning must stay “advisory, firm confirms,” and accuracy must be visibly high from day one.
4. Market timing: Simples/MEI deferred to 2027 — the 2026 beachhead is normal-regime only, narrower than the headline “millions of issuers.”

14. Structured verdict

Score:                  76/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder + Brazilian fiscal/tax advisor co-founder
Time to revenue:        6–10 weeks (free diagnoser → paid catalog sweep)
Capital to launch:      R$60–120K ($11–22K) — mostly fiscal-advisor time + inference
Top 3 assumptions to validate first:
  1. Firms pay R$699/mo for the fix-loop vs the free official assistant — 30 firm interviews + pre-sale page
  2. AI cClassTrib/CST classification >90% accurate on real catalogs — blind test vs senior analyst on 200 SKUs
  3. WhatsApp-group + reject-code SEO converts panicking analysts — seed 5 groups, measure 2-week signups
Kill criteria:
  - Abandon if <8% of firms in 50 cold WhatsApp/outreach conversations show pre-pay intent
  - Abandon if classification accuracy stays <85% after tuning (trust collapses below this)
  - Abandon if an official/incumbent ships a free fix-loop (not just a lookup) before v1 ships

15. Next step — 1-week validation sprint

• Day 1–2: Build the free public single-note diagnoser for the three top reject codes (1021/1024/1026). Stand up a R$699 pre-sale landing page.
• Day 3–4: Join 5 accountant WhatsApp/Telegram groups; answer real rejection screenshots with correct free diagnoses and the link. DM 40 firms posting reform-prep content on LinkedIn.
• Day 5: Decide go/no-go on a falsifiable bar — ≥8% of engaged firms (≥4 of ~50) leave a pre-pay deposit or book a paid pilot, AND the diagnoser correctly diagnoses ≥17 of 20 real rejections collected during the week. Miss either → back to Stage 2.