NairaScribe — voice e-invoice scribe for Nigerian merchants

1. One-liner

Turns a shopkeeper’s WhatsApp voice note in Pidgin or Hausa into a FIRS-compliant e-invoice, filed in seconds.

2. Trend signal — why now?

Three things collided in the last six months, and they point at the same merchant.

The regulatory cliff is real and dated. Nigeria’s Revenue Service (NRS, formerly FIRS) e-invoicing mandate moves from large taxpayers to medium taxpayers (₦1–5bn turnover) on 1 July 2026, with emerging/small businesses (<₦1bn) following 1 July 2027. Every invoice must carry 55 mandatory fields across 8 categories, be formatted as UBL BIS 3.0 XML/JSON, and be submitted via RESTful API to the Merchant-Buyer Solution. Non-compliance penalties: ₦200,000 administrative fine, ₦10,000/day for continued failure, and a 100% surcharge on unreported transactions. The Guardian and Businessday are both running “race to meet the July deadline” stories — “compliance remains uneven, with many yet to begin full implementation.”

Most Nigerian SMEs don’t invoice — they voice-note. The actual workflow today is a WhatsApp message: “send me 20 bags cement, ₦9,500 each” answered by a typed reply, a screenshot, or a voice note. There is no structured invoice anywhere in that loop. Businessday: “Most Nigerian business owners are sending invoices on WhatsApp — a typed message, maybe a screenshot, sometimes just a voice note — and then they wait and hope the customer pays.” 55 fields of XML is a different planet from that reality.

The voice tech that bridges the two just shipped — and it’s Nigerian. In March 2026 Lagos startup Intron launched Sahara-v2: speech recognition across 57 languages including Pidgin, Hausa, Yoruba and Igbo, the first models built for Nigerian-language code-switching, claiming 68.6% better accuracy than GPT-4/Gemini on African names, organisations and locations. AethexAI raised $3M pre-seed (June 2026) offering voice at $0.03/min — a third of Western providers. AgentPesa already ships voice banking in Yoruba/Igbo/Hausa/Pidgin, proving merchants will talk to a finance app in their language. The transcription problem that made this impossible 18 months ago is now an API call.

Demand (mandate + penalties + unprepared merchants) × feasibility (cheap Nigerian-language voice) × economics (the whole accounting-software market is repricing around this mandate) = a window that opens this month.

Provenance:
  - Signal 1 (Demand/Regulatory): NRS e-invoicing — 55 mandatory fields, medium taxpayers live 1 July 2026, ₦200K + ₦10K/day penalties; SMEs "yet to begin," WhatsApp voice notes are the status quo — https://guardian.ng/news/mid-tier-businesses-in-race-to-meet-e-invoicing-july-deadline/ — 2026-06
  - Signal 2 (Feasibility): Intron Sahara-v2 ships Pidgin/Hausa/Yoruba/Igbo ASR with code-switching, 68.6% better than GPT-4 on African entities; AethexAI voice at $0.03/min — https://techcabal.com/2026/03/05/intron-expands-sahara-to-57-languages/ — 2026-03
  - Signal 3 (Economic): Nigeria's e-invoicing push is repricing the SME accounting market; accountants "at the centre," readiness gap is the worry, not resistance — https://businessday.ng/business-economy/article/nigerias-e-invoicing-push-tests-smes-readiness-for-tax-automation/ — 2026-06
  Category: Regulatory arbitrage

3. The opportunity

The incumbents (Tyms, SimpleBks, Sage, the wave of NITDA-accredited Access Point Providers) all solve the back half of the problem: given a fully-structured invoice, transmit it to NRS. They assume the merchant will sit at a screen and type 55 fields into a form. For a cement-depot owner in Aba or a fabric trader in Kano who runs the business from a phone, in Pidgin, between customers, that form is the wall they hit.

The gap is the input layer. Nobody has built the bridge from “how a Nigerian merchant actually communicates a sale” (a 12-second voice note) to “what NRS demands” (55 structured fields + buyer TIN + VAT breakdown + product codes). Kippa — once 500K merchants — imploded and pivoted out of fintech in 2024, leaving that base orphaned and now facing a mandate the dead app never supported.

We don’t compete with the APPs — we sit in front of them. NairaScribe captures the sale by voice or WhatsApp, fills the 55 fields with AI, lets the merchant confirm in one tap, then routes the finished invoice through a partner APP’s accredited pipe to NRS. We own the merchant relationship and the moment of pain; the plumbing is somebody else’s commodity.

4. Target market

• Primary customer: Owner-operators and their counter/field staff at Nigerian trading and light-distribution SMBs — building materials, FMCG wholesale, fabrics, auto parts, agro-inputs — ₦300M–₦5bn turnover, 1–20 staff, run from WhatsApp and a smartphone, often invoicing in Pidgin/Hausa/Yoruba. The medium-taxpayer band (₦1–5bn) is the July 2026 wedge; the much larger sub-₦1bn band is the July 2027 expansion.
• Why they buy: “The mandate starts and I no sabi how to comply, my customer go reject the invoice, NRS go fine me ₦10,000 every day.” They don’t want accounting software — they want the penalty to not happen, with the least possible change to how they already work.
• Rough TAM reasoning: Nigeria has ~40M MSMEs; the formal, VAT-registered, mandate-exposed slice across the 2026–2028 phases is in the low millions. Capturing 10,000 paying merchants at ₦8,000/mo is a ~$6M/yr business — comfortably inside the sub-$5M-ARR target with room to spare even at deep discounts.
• Why now for them: The deadline is a calendar date, not a preference. Medium taxpayers are being told this quarter that unstructured WhatsApp invoicing is now a fineable offence.

5. Product sketch (MVP)

• Voice-in: Merchant sends a WhatsApp voice note (“Dangote cement, twenty bags, nine thousand five hundred each, sell give Musa Hardware”) in Pidgin/Hausa/Yoruba/Igbo/English. NairaScribe transcribes and extracts buyer, items, quantities, unit prices.
• 55-field autofill: AI maps the spoken sale to NRS’s mandatory fields, computes 7.5% VAT and totals, applies product/tax codes, and flags only the gaps it genuinely can’t infer (e.g. buyer TIN on first transaction).
• One-tap confirm: Merchant gets a clean draft invoice back in WhatsApp; taps ✅ to file or sends a correction voice note (“no, na fifteen bags”).
• Buyer directory: Remembers repeat buyers and their TINs, so the second invoice to Musa Hardware needs zero re-entry.
• Accredited filing: Routes the finished UBL invoice through a partner Access Point Provider to NRS; returns the IRN/QR-stamped invoice to share back to the buyer on WhatsApp.
• Rejection rescue: If NRS or the buyer rejects within the 72-hour review window, NairaScribe explains why in plain Pidgin and walks the fix.
• Accountant view: A simple web dashboard so the merchant’s bookkeeper/CA can see, batch-correct, and reconcile everything filed.

6. AI angle — what’s load-bearing

Remove the AI and there is no product — just another web form that already exists and that this customer cannot use. Two AI jobs are doing the lifting: (1) Nigerian-language, code-switching speech-to-structure — turning accented Pidgin/Hausa voice with mixed English numerals into clean line items, which only became commercially viable with Sahara-v2-class models; and (2) sale-to-schema mapping — inferring 55 NRS fields, VAT, and product codes from a 12-second utterance, then knowing which 3 fields it must ask about rather than guess. The defensible craft is the prompt/data layer that makes that mapping reliable enough that a merchant trusts the one-tap confirm. A generic invoicing form is not a substitute; the input modality is the product.

7. Localization angle (if any)

This is localization as the entire thesis, not a flavour. The tagged geography is Global in the taxonomy only because “Nigeria” isn’t a selectable value — the product is unapologetically Nigeria-first: Pidgin/Hausa/Yoruba/Igbo voice, ₦ pricing tiers a depot owner will actually pay (₦5,000–15,000/mo, not $49), WhatsApp as the primary surface, NRS’s specific 55-field UBL schema, and routing through NITDA-accredited APPs. The same shape ports cleanly to the next mandate cliffs — Kenya eTIMS, Egypt, Nigeria’s own 2027 small-taxpayer wave — but each is a deliberate localized re-cut, not a config flag.

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

• Pricing: ₦8,000/mo ($5) per merchant for a standard tier (up to N invoices/mo), ₦15,000/mo for high-volume; thin per-invoice voice cost (₦5–10 at $0.03/min) baked in. Accountant/CA multi-client seat at ₦25,000/mo managing 20+ merchants.
• ACV: ~₦96,000–180,000/merchant/yr ($60–115). CA seats push blended ACV higher.
• Rough math to $1M ARR: ~14,000 merchants × ₦8,000/mo × 12 ≈ ₦1.34bn ≈ $0.9M. Layer in CA seats and high-volume tiers and $1M lands at ~12,000 paying accounts.
• Rough math to $5M ARR: ~55,000–65,000 merchants, which requires riding the July 2027 sub-₦1bn wave (a far larger pool) plus the CA/bookkeeper channel doing the selling. Plausible but assumes the 2027 phase lands on schedule.
• Expansion path: Start at compliance (file the invoice). Expand to debt-chasing (the Kippa-proven killer feature — “Musa never pay, remind am”), then VAT-return prefill, then working-capital/credit referrals off the now-structured transaction history. Each step raises ACV without changing the merchant relationship.

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

• Ride the accountants, not the merchants. Businessday is explicit: “accountants and advisors will be at the centre of this transition.” Nigeria has thousands of small CA/bookkeeping firms each shepherding 20–50 SME clients into July compliance and dreading the data-entry load. Sign 5–10 such firms; each onboards their panicked client book. That alone clears 100+ merchants.
• Target the deadline-panic directly. Scrape NRS/NITDA lists and trade-association directories (cement dealers, FMCG distributors, fabric markets) of medium-taxpayer firms exposed on 1 July; send a WhatsApp demo where their own voice note becomes a filed invoice in 20 seconds. The demo is the pitch.
• Market-association sweeps. Nigerian trade is organised by physical markets with leadership (e.g. building-material and fabric market unions). One association endorsement + a WhatsApp broadcast reaches thousands of mandate-exposed merchants who all face the same deadline.
• Kill-the-fine ad angle. Pidgin short-form video (“NRS fit fine you ₦10,000 every day — talk your sale, we go file am”) on TikTok/WhatsApp Status where this audience actually is.

10. Build complexity — justification

Medium. The voice-to-text and TTS are off-the-shelf (Intron/AethexAI APIs already tuned for Nigerian languages — no model training). The genuine work is the sale→55-field mapping reliability, WhatsApp Business flow, and a clean integration with one accredited APP partner for the actual NRS submission (a commercial/API deal, not an accreditation we must win ourselves). A focused pair can ship a credible v1 in ~3–4 months: weeks on the voice→invoice loop, weeks on the APP integration and NRS round-trip, weeks hardening the confirm/rejection UX in real markets.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	We enable compliance; filing goes through an accredited APP, so no NITDA accreditation needed to launch.
Ethical — no harm / dark patterns	✅	Reduces penalty exposure for small traders; merchant confirms every invoice before filing.
Market exists (evidence above)	✅	Dated mandate + penalties + unprepared, voice-noting merchants.
1–5 person team can build this	✅	Off-the-shelf voice APIs + one APP integration.
Launchable with <$50K / ₹40L	✅	API-metered voice, standard stack, no hardware.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Dated, fineable mandate vs. merchants who literally cannot use a 55-field form. Hair-on-fire for the July cohort; slightly less so for those who’ll just hire a CA.
Demand evidence	15	12/20→/15	Strong, sourced: mandate, penalties, “race to deadline” press, WhatsApp-voice status quo. No direct “I’d pay for a voice filer” quote yet — that’s the gap to validate.
Build feasibility	15	11/15	Voice is API-grade; the APP integration and mapping reliability are real but bounded engineering.
Distribution clarity	15	11/15	The accountant channel and market-association sweeps are concrete; conversion math is estimated, not proven.
Revenue mechanics	15	11/15	₦ pricing fits wallets; $1M needs ~12–14K accounts, $5M leans on the 2027 wave landing on time.
Time to first revenue	10	7/10	Deadline-driven urgency compresses the funnel, but v1 needs the APP round-trip working before anyone pays — call it 8–12 weeks post-build.
Defensibility	10	6/10	Soft moats: Nigerian-language voice tuning, merchant + buyer-TIN graph that compounds, APP relationship, CA-channel lock-in. Copyable by a funded local incumbent in ~12 months.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

domain-expertise-required · technical-heavy — needs someone who genuinely understands NRS compliance and the Nigerian merchant context, paired with an engineer comfortable with voice pipelines and WhatsApp/APP integrations. A purely foreign team will mis-build the voice and the trust.

Key assumptions to validate (3–5)

1. Assumption: Merchants trust a one-tap confirm enough to file AI-generated invoices to a tax authority. How to test: Wizard-of-Oz with 25 merchants — they voice-note real sales, you hand-produce the invoice, measure whether they’d file it unedited.
2. Assumption: Sahara-v2-class ASR hits usable accuracy on real, noisy, code-switched market Pidgin/Hausa with numerals. How to test: Run 200 real merchant voice notes through the API; measure field-level extraction error, especially quantities and prices.
3. Assumption: An accredited APP will partner (reseller/API) rather than build the voice layer themselves. How to test: Pitch 3 APPs; get one signed LOI before building the integration.
4. Assumption: CA/bookkeeping firms will channel-sell to their client book at ₦25K/mo. How to test: 15 CA-firm interviews; pre-sell 3 firm seats.

Risk flags

1. Platform dependency: Two single points of failure — WhatsApp Business API terms and the one APP partner. Mitigate by lining up a second APP and a fallback web/USSD entry.
2. Market timing (knife-edge): If NRS slips the July 2026 medium-taxpayer date (African mandates slip often), the urgency that drives the funnel softens. The 2027 wave is the bigger prize but further out.
3. Incumbent fast-follow: A funded local accounting player (or an APP) bolts a voice front-end on within 12 months. Defence is speed + owning the CA channel + the buyer-TIN graph.
4. Accuracy liability: A mis-extracted price files a wrong invoice to the tax authority. The confirm step is the safety valve and must be unskippable.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Nigerian (or Nigeria-deep) operator with NRS/tax-compliance domain knowledge + a voice/integration engineer
Time to revenue:        8–12 weeks after a ~3–4 month build (deadline-compressed)
Capital to launch:      $15–30K (≈₦25–45M) — mostly API metering + a few months of two-person runway
Top 3 assumptions to validate first:
  1. Merchants will file AI-drafted invoices on one-tap confirm — Wizard-of-Oz with 25 merchants
  2. Real-market code-switched ASR is accurate enough on prices/quantities — 200 live voice notes through the API
  3. An accredited APP will partner rather than build it themselves — signed LOI from 1 of 3 APPs before integration
Kill criteria:
  - Abandon if field-level extraction error on prices/quantities exceeds ~5% on real market voice notes and can't be driven down
  - Abandon if no accredited APP will partner within 60 days (becoming an APP ourselves blows the capital/regulatory gate)
  - Abandon if <3 of 15 CA firms pre-commit a channel seat after demo

15. Next step — 1-week validation sprint

• Day 1–2: Record 200 real sale voice notes from 25 merchants across two physical markets (one Pidgin-heavy, one Hausa-heavy). Push them through Intron/AethexAI; hand-score field-level extraction accuracy on buyer, items, quantity, price.
• Day 3–4: Wizard-of-Oz the full loop for those 25 merchants — they voice-note, you manually produce the NRS-shaped invoice and send it back on WhatsApp; record how many would file it with zero edits. In parallel, pitch 3 accredited APPs for a partnership LOI and interview 15 CA firms on channel willingness at ₦25K/mo.
• Day 5: Decide. Go only if (a) extraction accuracy on price/quantity ≥95%, (b) ≥60% of merchants would file unedited, and (c) ≥1 APP says yes in principle. Anything less → fix the weakest input and re-test, or kill.

A falsifiable bar: 95% field accuracy AND 60% file-unedited AND one APP yes. Miss any of the three and this is not yet real.