Pitch & evidence pack
Traction evidence
Evidence vs claim
LANDED is pre-traction. 6 claims are backed by demonstrable product evidence; 4 are open items shown as action items rather than filled with invented numbers.
Demonstrate it live
The strongest evidence we have is the working product itself. Both of these are one click away during a pitch.
Claim-by-claim
A working product is deployed and publicly reachable.
Evidence verifiedEvidence: Live custom domain smartimport.trendflux.digital; / and /analyze return HTTP 200; production build passes.
How to show it: Open the live domain, or run the analyze flow in front of the judge.
A full landed-cost report can be produced end to end without an account.
Evidence verifiedEvidence: Demo report renders instantly from the landing page; live analysis path runs from /analyze.
How to show it: Click 'See the Demo' on the landing page — no AI call, no network dependency.
Supplier documents can be ingested from real-world file formats.
Evidence verifiedEvidence: Browser-side extraction for PDF, DOCX, image, TXT and CSV, with 12 MB / 18,000-character caps and typed errors for password-protected or corrupt PDFs.
How to show it: Drop a real proforma invoice into /analyze and show the extracted text.
Outputs are labelled as estimates and routed to human verification.
Evidence verifiedEvidence: Per-field provenance labels (from document / AI estimate / not in input), verification checklist, and a global verification disclaimer.
How to show it: Scroll the report to the provenance labels and verification checklist.
Documents are not stored on LANDED servers.
Evidence verifiedEvidence: Extraction happens in the browser; the server forwards extracted text and page images to a third-party AI model provider and persists nothing. Analysis history is device-local.
How to show it: Point to the data-handling disclosure on /analyze.
The codebase passes security and dependency scanning.
Evidence verifiedEvidence: Server-only API key, Zod validation on input and model output, no client-side secrets, clean security and dependency scans.
How to show it: Describe the architecture; do not claim a formal external audit — there has not been one.
Real importers have used LANDED.
No evidence yetEvidence: None. Zero users, zero pilots, zero letters of intent, zero revenue.
How to close it: Run 3–5 real importer documents with named importers and capture written feedback.
LANDED's estimates are accurate.
No evidence yetEvidence: None. No accuracy benchmark exists. No accuracy percentage is claimed anywhere in the product or this deck.
How to close it: Run 5 real invoices through LANDED and place a licensed broker's actual duty/landed-cost numbers beside them. This is the single highest-value missing artifact.
Product usage analytics show engagement.
No evidence yetEvidence: Client-side event instrumentation exists in code, but no production analytics provider is connected, so there is no usage data to show.
How to close it: Connect an analytics provider, then present real numbers — or say nothing about usage.
Customer demand is validated by interviews.
No evidence yetEvidence: None recorded in the project. The problem framing is founder-observed.
How to close it: Conduct and document 5 named importer interviews with quotes and consent.
What we know / assume / still need to validate
What we know
- The product works end to end and a judge can drive it live.
- Bangladesh imported roughly USD 67bn of merchandise in FY2024-25 (derived from Bangladesh Bank country-share data).
- Supplier documentation in this market arrives as PDFs, scans and photos — which is exactly the input LANDED accepts.
- We have no users, no revenue and no accuracy benchmark.
What we assume
- SME importers will pay a monthly fee to reduce uncertainty at the moment of commitment.
- Indenting houses and C&F agents are a viable distribution channel rather than a competitive threat.
- AI classification suggestions are good enough to save a broker consultation, not replace it.
- Inference cost per analysis stays low enough for SME-level pricing.
What we still need to validate
- Broker-verified accuracy comparison on at least 5 real invoices.
- Willingness to pay: a signed pilot or a paid trial from a real importer.
- Registry-grounded importer counts for SAM/SOM instead of estimated ranges.
- Whether the buyer is the owner or the procurement lead, and how the purchase is approved.
This pitch pack contains no customer, revenue, partnership or accuracy claims. Where a number is not directly sourced, it is labelled as derived or as an assumption with its method shown. Prepared by TrendFlux Digital for Startup World Cup Bangladesh 2026. Last reviewed 27 August 2026.