OrderPabili — voice-note order desk for Philippine sari-sari distributors

1. One-liner

Turns the Viber voice notes and handwritten-list photos your clerks re-key all day into a clean, priced order sheet.

2. Trend signal — why now?

Three things landed in the same 12 months, and they point at the same clerk drowning in Viber messages.

First, the retailer-side gold rush is over and it skipped the distributor entirely. Growsari has wired up 50,000+ sari-sari stores; Packworks reaches ~23% of the estimated 1.3M stores across 96% of provinces. Every peso of venture money went into apps that ask the store owner to order through an app. Nobody built for the distributor who still receives those orders the old way — as voice notes and photos of handwritten lists in Viber — and pays a clerk to re-type them.

Second, the store owners refuse to switch. SimpliDOTS — a 400-distributor Indonesian incumbent — says it plainly in its own marketing: “many retailers continue relying on WhatsApp instead of adopting dedicated sales order applications.” The order-app model has been fighting the messaging channel for a decade and losing. The orders live in chat and they are staying in chat.

Third, the tech to read that chat got cheap and good in the last year. Soniox transcribes Tagalog — including Manila, Batangas and Bulacan dialects — at ~1.25% word error rate for $0.10/hour. Meta launched its Business Agent Platform on WhatsApp on 1 July 2026 with published token pricing, so a small team can run a compliant automated agent inside the channel the orders already flow through.

The right move isn’t to drag the sari-sari owner into an app. It’s to stop fighting Viber and read it.

Provenance:

• Signal 1 (demand): Distributors receive orders as voice notes, photos of handwritten lists and casual chat; a clerk manually re-keys them, with 1-2% transcription error even from good staff — https://bizowie.com/how-distributors-can-automate-80-of-manual-order-entry & https://sunraydatalinks.com/blog/whatsapp-order-management-system-for-distributors-wholesalers/ — 2026-07-15
• Signal 2 (feasibility): Soniox transcribes Tagalog dialects at ~1.25% WER for $0.10/hr; Meta Business Agent Platform launched on WhatsApp 1 July 2026 with token pricing — https://soniox.com/pricing & https://techcrunch.com/2026/06/03/metas-ai-agent-for-whatsapp-business-is-now-available-globally/ — 2026-07-15
• Signal 3 (economic): Sari-sari = ~28% of PH FMCG revenue across 1.3M stores; Growsari (50K stores, Temasek/Tencent-backed) and Packworks (300K stores) funded heavily — all on the retailer side, none on the distributor’s inbox — https://packworks.io/ & https://techcrunch.com/2021/06/21/growsari-a-b2b-platform-for-small-stores-in-the-philippines-adds-investors-like-temaseks-pavilion-capital-and-tencent/ — 2026-07-15 Category: Geographic arbitrage

3. The opportunity

The whole sari-sari digitization wave attacked the wrong end of the pipe. It assumed the way to clean up ordering was to get the store owner to type into an app. The store owner won’t. She sends a 40-second Tagalog voice note — “pahingi ng dalawang case ng Kopiko, isang dosenang Lucky Me, tapos yung Palmolive na sachet, yung dati” — because that’s how she talks to her suko (regular supplier) and always has.

So the pain didn’t disappear. It moved downstream to the distributor, where a clerk sits with headphones on, scrubbing voice notes and squinting at photos of handwritten lists, retyping each one into an Excel order sheet or a DMS. That clerk makes errors on 1-2% of lines even on a good day — wrong SKU, wrong quantity, missed the “yung dati” (the usual) shorthand — and every error is a wrong delivery, a return, or a stock-out the salesman gets blamed for.

The incumbents can’t fix this because their whole thesis is “get off Viber.” OrderPabili’s thesis is the opposite: the order is already in Viber, so read it there. Connect to the distributor’s existing Viber/WhatsApp number, transcribe every voice note and OCR every handwritten photo, match the free-text to the distributor’s own SKU list and price file, and drop a clean, priced, confirmed order sheet in front of the clerk — who now reviews instead of transcribes. Same channel the store owner loves, 80% less re-keying for the distributor.

4. Target market

• Primary customer: Owner or operations manager of a regional/provincial FMCG distributor or wholesaler in the Philippines — the businesses in Divisoria, Carbon Market (Cebu), Bankerohan (Davao) and provincial hubs that supply 50-500 sari-sari stores each. Typically 3-30 staff, 1-3 of them full-time “encoders” whose entire job is turning chat into order sheets.
• Why they buy: “My two encoders can’t keep up on the 15th and 30th when everyone reorders. We ship wrong SKUs, we miss lines, the store calls angry, and I’m paying two salaries to retype voice notes.” The pain is a recurring, dated, staff-cost pain — not an abstraction.
• Rough TAM reasoning: Sari-sari stores are ~28% of PH FMCG revenue across ~1.3M stores. Each store is served by a distributor; distributors number in the low tens of thousands (thousands of regional players plus each major brand’s appointed distributors per territory). Even 8,000 addressable distributors at ₱6,000/mo is a ₱576M ($10M) ARR ceiling in the Philippines alone — before Indonesia’s warung grosir and Vietnam.
• Why now for them: Labor is getting more expensive and encoders are hard to retain; order volume through chat keeps rising as more store owners get comfortable ordering by voice note; and for the first time the transcription is accurate and cheap enough in Tagalog/Bisaya that “let the machine do the first pass” is actually credible.

5. Product sketch (MVP)

• Connect the distributor’s existing Viber/WhatsApp Business number — no new number, no asking store owners to change anything.
• Auto-transcribe every incoming voice note (Tagalog, Bisaya, Taglish) and OCR every photo of a handwritten order list.
• Match the messy free text to the distributor’s uploaded SKU catalog + price list — resolve “yung Kopiko brown, dalawang case” to the exact SKU, pack size and current price.
• Produce a clean, priced draft order per store, with low-confidence lines flagged for the clerk to eyeball (not retype).
• One-tap confirm-back to the store owner in the same chat: “Order mo: 2 case Kopiko, 1 dosena Lucky Me… total ₱X. Tama ba?” — closing the loop before dispatch.
• Flag out-of-stock and suggest the substitute the distributor actually carries, right in the draft.
• Daily export/push into whatever the distributor already runs (Excel order sheet, or a DMS via CSV) — OrderPabili is the front door, not a rip-and-replace.
• Simple dashboard: orders processed, lines auto-matched vs. flagged, encoder hours saved.

6. AI angle — what’s load-bearing

Remove the AI and there is no product — it’s just a shared inbox. The AI does three jobs a human currently does by hand: (1) multilingual, dialect-tolerant speech-to-text on noisy phone voice notes, (2) OCR + handwriting recognition on photographed lists, and (3) the hard part — fuzzy entity resolution from casual vernacular (“yung dati,” brand-nicknames, quantities in cases/dozens/tingi) to a specific distributor’s SKU and price row. That matching layer, tuned per distributor’s own catalog and each store’s ordering history, is where the accuracy — and the moat — lives. This is a task that was genuinely not automatable at acceptable cost or accuracy 18 months ago.

7. Localization angle (if any)

This is the localization play — it’s the entire wedge. A generic US “voice order capture” tool (VoiceOrder Solutions et al.) assumes English voicemail and phone; it cannot read a Taglish Viber voice note or a photo of a handwritten sari-sari list. The Philippine cut needs: Tagalog + Bisaya + Taglish transcription; Viber as first-class (Viber, not just WhatsApp, dominates PH sari-sari ordering); local units (case, dosena, tingi/sachet); GCash-era price sensitivity so the tool has to sell at a ₱-native price a provincial distributor will actually pay (~₱6,000/mo, not $200). The same shape ports next to Indonesia (warung, Bahasa, WhatsApp) and Vietnam, each a fresh localized build rather than a translation.

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

• Pricing: ₱6,000/mo ($105) base per distributor for up to 1,500 orders/mo, then usage tiers up to ₱15,000/mo ($265) for high-volume hubs. Priced against the loaded cost of one encoder (₱18,000-25,000/mo).
• ACV: ₱90,000 ($1,600) blended.
• Rough math to $1M ARR: ~625 distributors × ~$1,600 ACV ≈ $1M. That’s a small slice of the addressable base.
• Rough math to $5M ARR: ~3,000-3,500 paying distributors across PH + Indonesia, plus an upsell tier (auto-confirm agent, credit/collections nudges, brand-funded promo insertion). Requires proving accuracy holds across dialects and that expansion into Indonesia converts.
• Expansion path: land on order-transcription, expand to (a) automated confirm-back replies as a paid Business Agent add-on, (b) analytics distributors sell back to brands (“what’s actually being reordered in Bicol”), (c) collections chase on unpaid consignment orders — the same voice/chat rails, higher ACV.

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

• The wholesale physical clusters are a list you can walk. Divisoria (Manila), Carbon Market (Cebu), Bankerohan (Davao) concentrate hundreds of distributors in a few blocks. Send a two-person team, demo live by feeding a real Viber voice note through the tool on a phone, and close on the spot. This is a stall-to-stall sales motion, not a webinar.
• Encoder-pain referral loop. The buyer is the owner; the champion is the exhausted encoder. Give the encoder a visibly lighter 15th/30th and they’ll name the three other distributors their cousin/friend works at. Warm intros in a tight-knit trade community.
• Ride a brand’s appointed-distributor network. URC, Nestlé PH, Alaska each appoint distributors per territory and want cleaner order data. Pilot with one brand’s distributor list as a co-selling wedge — the brand gets reorder analytics, the distributors get the tool. One brand relationship = dozens of pre-qualified distributors.
• Viber/Facebook wholesaler groups. The same “PUSAT DISTRIBUTOR & GROSIR” and sari-sari supplier groups where orders happen are where distributors hang out. Seed a before/after demo video (“2 encoders → 1 reviewer”) into 5-10 of these groups.

10. Build complexity — justification

Medium. Transcription (Soniox), OCR/handwriting (off-the-shelf vision) and Viber/WhatsApp connectivity (Business API + Meta Business Agent Platform) are all bought, not built. The genuine engineering is the per-distributor SKU/price matching layer and a fast human-review UI for flagged lines — plus the unglamorous work of ingesting each distributor’s messy catalog. A pair can ship a credible v1 in ~10-14 weeks; the risk is accuracy tuning per dialect and per catalog, which is iteration, not research.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Processing a business’s own inbound orders on its own number; standard messaging-API terms.
Ethical — no harm / dark patterns	✅	Confirm-back loop reduces wrong deliveries; transparent to store owners.
Market exists (evidence above)	✅	1.3M stores, thousands of distributors, incumbents admit the WhatsApp habit persists.
1–5 person team can build this	✅	Bought AI + one hard matching layer; pair in ~3 months.
Launchable with <$50K / ₹40L	✅	Off-the-shelf APIs, field sales is two people and shoe leather.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Recurring, dated, staff-cost pain felt hard every 15th/30th; wrong deliveries and stock-outs. Not hair-on-fire daily, but sharp and expensive.
Demand evidence	15	12/15	Multiple independent sources describe the voice-note/handwritten re-keying pain; incumbents confirm the WhatsApp habit won’t die. Direct PH-distributor quotes are thin — hence not higher.
Build feasibility	15	11/15	Bought transcription/OCR/messaging; real work is per-catalog matching + review UI. Pair in ~3 months.
Distribution clarity	15	12/15	Physical wholesale clusters are a walkable list; brand-distributor networks give pre-qualified leads. Field sales is proven but labor-real.
Revenue mechanics	15	11/15	Priced below one encoder’s salary — easy ROI story. ACV modest; needs volume, and provincial ₱ pricing caps per-seat upside.
Time to first revenue	10	8/10	Live demo closes on-site; pilot-to-paid in weeks once matching accuracy is credible.
Defensibility	10	4/10	Execution + per-distributor catalog/history lock-in and local trade relationships. Copyable; a funded incumbent (Packworks) could bolt this on.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (dialect ASR + fuzzy SKU matching is the whole product) · operations-heavy (field sales in wholesale markets, per-distributor catalog onboarding).

Key assumptions to validate (3–5)

1. Assumption: A meaningful share of a distributor’s inbound orders genuinely arrive as voice notes / handwritten photos (not already-typed text). How to test: Sit in 10 distributors’ Viber inboxes for a week and count the format mix.
2. Assumption: The matching layer can hit >90% line accuracy on real Taglish/Bisaya voice notes against a real catalog, good enough that clerks trust the review flow. How to test: Run 500 real voice notes from 3 distributors through a prototype; measure line-level precision/recall vs. the encoder’s manual entry.
3. Assumption: Owners will pay ~₱6,000/mo — i.e., they perceive it as cheaper than the encoder headcount, not as “another app.” How to test: 30 in-person price interviews across Divisoria/Carbon/Bankerohan.
4. Assumption: Store owners tolerate the auto confirm-back message and it reduces (not increases) friction. How to test: A/B the confirm-back on one distributor’s live orders; track complaint and correction rates.

Risk flags

1. Platform dependency: Built on Viber + WhatsApp/Meta Business Agent Platform. Terms, pricing (token charges began Aug 2026) or access can shift under you. Mitigate by supporting both channels and keeping the review UI channel-agnostic.
2. Incumbent bolt-on: Packworks/Growsari have the distributor relationships and could add “read the voice note” as a feature. Your only lead is focus and speed to per-catalog accuracy.
3. Accuracy trust cliff: If early matching is sloppy, encoders revert to manual and never come back. First 10 accounts must be babysat to high accuracy before scaling sales.
4. Market timing: Voice-note-to-order is being actively built for India FMCG right now; the idea’s freshness is the geography, and that window is not permanent.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (dialect ASR + fuzzy matching) with a PH ops/sales partner who knows the wholesale trade
Time to revenue:        8–12 weeks (live-demo close in wholesale clusters)
Capital to launch:      ₱1.5–3M ($26–52K) — mostly two-person field sales + onboarding labor
Top 3 assumptions to validate first:
  1. Order-format mix is voice/handwritten-heavy — audit 10 distributor inboxes for a week
  2. >90% line accuracy on real Taglish/Bisaya voice notes vs. real catalog — 500-note prototype test
  3. Owners pay ~₱6,000/mo as encoder-cost substitution — 30 in-person price interviews
Kill criteria:
  - Abandon if <40% of a typical distributor's inbound orders arrive as voice notes / handwritten photos (then it's just a text-parser, no wedge)
  - Abandon if line-level matching accuracy stalls below ~85% after tuning on 3 real catalogs (clerks won't trust it)
  - Abandon if a funded incumbent ships an equivalent distributor-side voice reader before your 10th paying account

15. Next step — 1-week validation sprint

• Day 1–2: Recruit 5 provincial/regional distributors (via a Divisoria contact + one brand’s distributor list). Get read-access to a week of their Viber/WhatsApp order inboxes.
• Day 3–4: Hand-tally the order-format mix (voice / handwritten photo / typed text). Pull ~500 real voice notes and run them through a Soniox + off-the-shelf-matching prototype against those distributors’ actual SKU/price files. Score line-level accuracy vs. what the encoder actually entered.
• Day 5: Decide go / no-go on a falsifiable bar: ≥40% of orders are voice/handwritten AND prototype line accuracy ≥85% on ≥2 of the 3 catalogs. Miss either and the wedge isn’t there — go back to signals.