ClearPresent — LC discrepancy catcher for Bangladesh exporters

1. One-liner

Reads your LC and shipping docs, flags every discrepancy, and rebuilds a clean set before the bank rejects it.

2. Trend signal — why now?

Bangladesh runs on the documentary letter of credit. LCs settle more than 65% of the country’s export payments and 97% of imports — there is no alternative payment rail at scale. And the document set behind those LCs is broken on arrival: industry commentary holds that over 95% of the import LCs Bangladeshi exporters rely on are discrepant, and global first-presentation discrepancy rates under UCP 600 sit at 60–80%. The discrepancy is almost never the goods — it’s “the cost of a comma”: a date format, a missing phrase in the bill of lading, a 47A special-condition the commercial officer didn’t parse. The bank can’t release payment until it’s fixed, and each LC already carries roughly US$2,500 in issuance, risk, and discrepancy-processing fees before the delay even starts.

What changed in the last 12 months: 2026 multimodal LLMs can finally read the messy real inputs — a scanned LC with a dense Field 47A block, a freight forwarder’s bill of lading, a commercial invoice and packing list — and cross-check every field against the LC terms and the implied UCP 600 / ISBP 821 rules in seconds. The category is being proven right now by SmartLC (UK), which sells exactly this “four-layer” check to “exporters, freight forwarders, and trade finance teams” — but it ships English-first, carries jurisdiction rules for China, UAE, and India, and has no Bangladesh ruleset, no Bangla, and enterprise-trade-finance positioning. The single most LC-dependent export economy on earth is unserved by the one tool that does this.

The money is visibly moving and visibly stuck: BTMA members reported Tk420 crore (~US$36M) in LC payments delayed past maturity across ~52 textile mills, with delays exceeding six months tying up working capital.

Provenance:

• Signal 1 (demand): “Over 95% of import LCs Bangladeshi exporters rely on are discrepant”; documentation errors called the “hidden tax” on Bangladesh’s exports; ~US$2,500 fee load per LC — The Daily Star — 2026 — https://www.thedailystar.net/opinion/views/news/the-cost-comma-documentation-errors-are-the-hidden-tax-bangladeshs-exports-4120986
• Signal 2 (feasibility): SmartLC ships AI four-layer LC checking (LC terms, UCP 600 implied, 47A conditions, cross-document consistency) for exporters/forwarders — built on UCP 600/ISBP 821/eUCP 2.1, jurisdiction rules for China/UAE/India only, no Bangladesh — https://smartlc.ai/ — 2026
• Signal 3 (economic): Tk420 crore (~US$36M) in LC payments delayed past maturity across ~52 Bangladesh textile mills; first-presentation discrepancy rates 60–80% under UCP 600 — Apparel Views / Trade Finance Global — 2026 — https://www.apparelviews.com/bangladesh-textile-sector-faces-crisis-over-delayed-lc-payments/ Category: Geographic arbitrage

3. The opportunity

The gap is a clean overlay of two facts: the document discrepancy problem is universal and brutal in Bangladesh, and the one tool built to solve it doesn’t speak Bangladesh.

SmartLC validated that exporters and forwarders will buy an AI LC-checker. But it was built for the London/Dubai/Mumbai trade-finance desk: English UI, generic global rules, demo-led enterprise motion, no pricing that survives contact with a Dhaka commercial department. Bangladesh’s exporters need the same engine tuned to their reality: the specific 47A boilerplate that Bangladeshi issuing and advising banks layer on, the back-to-back LC chain (an export LC drives a web of local fabric/accessory LCs, each a fresh discrepancy surface), Bangla-English document handling, and a price a US$5–50M-revenue factory’s commercial team can expense without a board meeting.

The incumbent the AI-first team is actually beating is not SmartLC — it’s the bank’s own document checker plus the factory’s overworked commercial officer. Today the workflow is: commercial officer assembles the set, the bank’s checker finds the discrepancy days later, the set bounces, the buyer learns the shipment is “discrepant” and uses it as leverage to demand a discount or delay 120 days. ClearPresent moves the catch from “after the bank, in front of the buyer” to “at the exporter’s desk, before anyone sees it.” Same outcome SmartLC promises — localized to the market that needs it most and can’t buy the English version.

4. Target market

• Primary customer: The commercial officer / commercial manager inside a Bangladeshi RMG or textile export factory (US$5–80M annual export revenue, Dhaka/Chattogram/Gazipur/Narayanganj). This person personally assembles and presents the LC document set and personally eats the blame when it bounces. Secondary buyer: independent C&F / freight-forwarding agents who prepare documents on exporters’ behalf.
• Why they buy (in their words): Discrepancies “arise not from negligence but from misunderstanding the detailed requirements embedded in letters of credit” — exactly the 47A wording trap. As one MD put it: “LCs is supposed to mitigate risks, yet banks charge commissions without ensuring timely payments.” They’re paying for a rail that bounces their docs and still bills them.
• Rough TAM reasoning: Bangladesh has ~3,500–4,000 active RMG export factories plus thousands of textile mills and a deep C&F-agent layer. Even a serviceable core of ~3,000 mid-size exporters and ~1,000 forwarders that present LC sets regularly is a real wallet. Each presents dozens of LC sets a year.
• Why now for them: Post-LDC-graduation (2026) Bangladesh loses preferential market access — margins compress, and every avoidable discrepancy fee + working-capital delay matters more than it did two years ago. Buyers are already using discrepancies as price-renegotiation leverage.

5. Product sketch (MVP)

• Upload the LC (with Field 47A block) plus the draft document set — commercial invoice, packing list, bill of lading/AWB, certificate of origin, beneficiary’s certificate, insurance doc.
• Discrepancy report in minutes: every mismatch rated Critical / Warning / Info, each tied to the specific LC clause or UCP 600 / ISBP 821 rule it breaks, in plain Bangla-English.
• 47A condition parser: extracts each special condition into a plain-language checklist (“invoice must state contract no. + ‘Made in Bangladesh’ + exact buyer style code”) and verifies the docs satisfy it.
• Cross-document consistency check: quantities, amounts, descriptions, dates, port names, and consignee details must agree across all docs and the LC — the single most common bounce.
• Fix-it drafts: for each fixable discrepancy, a corrected wording block the officer can paste into the invoice/certificate before re-export.
• Pre-presentation “clean / not clean” verdict with a one-page summary the officer can show their manager and the negotiating bank.
• Back-to-back LC linker: flags where an export LC’s terms aren’t satisfiable by the local fabric/accessory LCs feeding it (the BD-specific chain risk).

6. AI angle — what’s load-bearing

Remove the AI and there is no product — just a checklist nobody fills in correctly, which is the status quo that produces a 60–95% discrepancy rate. The work AI does: (1) read unstructured, scanned, inconsistently formatted documents and a dense free-text 47A block; (2) interpret what each LC condition implies under UCP 600 / ISBP 821 (the implied-rule layer is exactly where human checkers and officers miss things); (3) reconcile dozens of fields across six-plus documents for exact-match consistency; (4) draft corrected wording. This is precisely the multimodal-reasoning workload that only became reliable in 2026 — and the reason SmartLC exists at all. The moat question isn’t whether AI can do it; it’s whether someone localizes it for Bangladesh before SmartLC does (see §13).

7. Localization angle

This is the wedge — it’s a geographic-arbitrage play end to end. Bangla-English document handling; the specific 47A boilerplate Bangladeshi issuing/advising banks attach; the back-to-back LC chain that’s structurally central in BD and largely absent from the generic global tool; pricing in BDT at a level a mid-size factory’s commercial budget absorbs; and distribution through BGMEA/BKMEA/BTMA channels and the C&F-agent network rather than a London “book a demo” funnel. A generic global product (SmartLC) cannot match the ruleset, the language, the price, or the channel simultaneously — that’s the whole bet.

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

• Pricing: Hybrid SaaS tier + per-set check. Starter BDT 4,000/mo (~$33) for solo C&F agents / small exporters (limited checks); Pro BDT 12,000/mo (~$100) for a factory commercial team (higher volume, back-to-back linker, multi-user); Enterprise BDT 30,000/mo (~$250) for groups with many LCs/month. Overage per extra LC set checked.
• ACV: Blended ~$1,000/year (mix skews toward Pro for factories).
• Math to $1M ARR: ~1,000 paying accounts × ~$1,000 = $1M. ~1,000 of ~4,000 mid-size exporters/forwarders is a credible 18–24 month target in one country.
• Math to $5M ARR: ~4,000–5,000 accounts across Bangladesh plus the obvious expansion: same engine, new geos (Vietnam, Pakistan, Sri Lanka — all high-LC RMG exporters), and an import-side check for the back-to-back chain.
• Expansion path: per-seat growth inside larger commercial departments; per-set volume; add a “submit-ready packet export” and bank-specific profiles; later, a paid analytics layer (“your top 5 recurring discrepancy types this quarter”).

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

• Scrape the BGMEA + BKMEA + BTMA member directories (public, thousands of factories with contact + commercial-dept details). Email/WhatsApp the commercial manager a free one-LC discrepancy report: “Send us your last bounced LC set; we’ll show you in 10 minutes exactly what the bank flagged and how to fix it.” A bounced set is a wound that’s still open — high reply rate.
• C&F / freight-forwarder agent network: these agents prepare docs for many exporters; sign 10 agents and you reach hundreds of factories. Offer them a per-set Starter plan and a referral cut.
• Run a “discrepancy clinic” at a Dhaka/Chattogram apparel-tech event (or a sponsored BGMEA webinar): live-screen 5 anonymized real LC sets, catch the discrepancies on stage, collect emails. The 95%-discrepant statistic sells itself when the room watches it happen to their own format.
• LinkedIn + textilemerchandising/garments-merchandising communities where commercial officers and merchandisers already swap LC horror stories — seed the free-report offer there.

10. Build complexity — justification

Medium. The document ingestion + multimodal extraction + cross-doc reconciliation is off-the-shelf 2026 LLM tooling — no custom model. The genuinely hard, defensible work is encoding the UCP 600 / ISBP 821 rule layer plus the Bangladesh-specific 47A and back-to-back patterns into a reliable, auditable check (false positives erode trust fast, false negatives are catastrophic). That’s domain engineering, not research. A 2–3 person team with one trade-finance domain expert ships a credible v1 in ~3–4 months.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Advisory/preflight tool; doesn’t issue/negotiate LCs or replace the bank’s examination.
Ethical — no harm / dark patterns	✅	Helps exporters get paid faster on legitimate trade; no dark patterns.
Market exists (evidence above)	✅	65%+ of BD export payments via LC; 60–95% discrepancy rates; SmartLC proves WTP.
1–5 person team can build this	✅	2–3 people + trade-finance advisor.
Launchable with <$50K / ₹40L	✅	LLM API + standard web stack + domain advisor time.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Hair-on-fire: every bounced set = discrepancy fee + working-capital delay + buyer price leverage. Felt on most shipments.
Demand evidence	15	13/15	Multiple independent signals: 95% discrepant claim, 60–80% UCP rates, Tk420cr delayed, and SmartLC charging money for the exact check.
Build feasibility	15	11/15	Off-the-shelf multimodal LLM; the rule-layer + BD-specific encoding is the real work. ~3–4 months.
Distribution clarity	15	11/15	Named directories (BGMEA/BKMEA/BTMA), C&F-agent multiplier, “bounced set” free-report hook. Conversion still unproven.
Revenue mechanics	15	11/15	Clear hybrid pricing benchmarked to SmartLC + bank fees; BDT tiers fit the wallet; ACV blend is an assumption.
Time to first revenue	10	7/10	Free-report → paid in weeks once v1 checks accurately; needs trust-building on accuracy first.
Defensibility	10	5/10	Execution + localized ruleset + channel relationships. SmartLC could add a BD module — moat is speed + local depth, not structural.
Total	100	75/100

13. Qualitative modifiers

Founder-fit tags

domain-expertise-required · technical-heavy

Key assumptions to validate (3–5)

1. Assumption: A Bangladeshi commercial officer will trust an AI “clean/not-clean” verdict enough to act on it before bank presentation. How to test: Run 30 real bounced sets through the v0 engine; measure whether it catches what the bank actually flagged (precision/recall vs the bank’s real discrepancy notice).
2. Assumption: Factories will pay ~$100/mo when the bank “already checks” the docs. How to test: 20 paid pilots at BDT 12,000/mo after a free report; measure conversion and willingness to pre-pay a quarter.
3. Assumption: The C&F-agent channel multiplies reach (one agent → many factories). How to test: Sign 5 agents, measure how many distinct exporters each brings in 60 days.
4. Assumption: The BD-specific 47A + back-to-back patterns are encodable to >90% catch rate. How to test: Build the rule layer against 100 historical LC sets with known outcomes; measure catch rate.

Risk flags

1. Incumbent fast-follow: SmartLC (or a bank’s own vendor) ships a Bangladesh ruleset + Bangla UI before you reach scale. Moat is speed and local channel depth, not structure.
2. Accuracy/trust risk: A single high-profile false-negative (tool said “clean,” bank bounced it, buyer took a discount) can kill word-of-mouth in a tight industry. Accuracy is existential, not a feature.
3. Buyer concentration: RMG is consolidating into larger groups; if the mid-size segment shrinks, the core wallet thins. Mitigate with the C&F-agent layer and multi-geo expansion.
4. Platform dependency: Reliant on third-party multimodal LLM APIs for extraction quality and cost.

14. Structured verdict

Score:                  75/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder + trade-finance/LC domain expert (ex-bank trade desk or RMG commercial manager), Bangladesh-based or with deep local channel access
Time to revenue:        8–12 weeks (free report → paid pilot)
Capital to launch:      $8–15K (LLM API + web stack + domain-advisor time)
Top 3 assumptions to validate first:
  1. Catch rate vs real bank discrepancy notices — run 30 bounced sets, compare to the bank's actual flags
  2. WTP at ~$100/mo despite "the bank already checks" — 20 paid pilots post-free-report
  3. C&F-agent channel multiplier — sign 5 agents, count distinct exporters reached in 60 days
Kill criteria:
  - Abandon if the engine catches <80% of what banks actually flag on the 30-set test
  - Abandon if <15% of free-report recipients convert to a paid pilot after 60 days
  - Abandon if SmartLC or a local bank-vendor ships a full Bangla 47A/back-to-back module before your v1

15. Next step — 1-week validation sprint

• Day 1–2: Collect 20–30 real, recently bounced LC document sets with the bank’s actual discrepancy notice — via 3–4 friendly commercial managers and 2 C&F agents. The notice is the ground truth.
• Day 3–4: Run all sets through a thin multimodal-LLM prototype (no UI) that checks against LC terms + a hand-coded UCP/ISBP/47A rule subset. Record catch rate against each bank’s real flags.
• Day 5: Decide. Go if the prototype catches ≥80% of the discrepancies the banks actually flagged AND ≥3 of the commercial managers say “I’d pay BDT 12,000/mo for this.” Anything less = the rule layer isn’t there yet; iterate or no-go.

The result is falsifiable: it’s a measured catch-rate number against real bank notices, not “people seemed interested.”