Pitch & evidence pack
Demo narrative
The 2–3 minutes judge walkthrough
Problem → input → AI analysis → landed-cost decision intelligence → verification and confidence → business action. Every beat has a line to say and a thing to show.
Before you present
- Open the live domain with /analyze already loaded in a second tab.
- Have one real supplier PDF on the desktop, plus the demo report as a zero-network fallback.
- If the AI path fails or stalls, the client times out visibly and you switch to 'See the Demo' — say out loud that it is a pre-computed illustrative report.
Beat by beat
The problem, in one sentence
"An importer here places an order knowing the unit price — and not knowing the real landed cost, the right HS code, or what compliance will demand. That gets discovered after the money is committed."
Show: Landing hero.
The input a real importer actually has
"This is what they get from the supplier — a PDF proforma. No structured data, no ERP feed."
Show: Drop the supplier PDF into /analyze; extraction preview appears.
The analysis
"LANDED reads it in the browser, sends only the extracted text to the model, and returns one structured analysis."
Show: Run the analysis; the report reveals and focus scrolls to it.
The decision intelligence
"Line items, ranked HS candidates with reasoning, duty and tax estimates, landed cost per unit, compliance checklist, risk signals — and a proceed, renegotiate or hold call."
Show: Scroll the report: HS candidates → landed cost table → risks → recommendation.
Confidence and verification — the credibility beat
"Every figure is labelled: from the document, AI estimate, or not in the input. Duty is an estimate, not a live tariff feed. This is decision support that routes to a broker for confirmation, and we say so in the product."
Show: Provenance labels and verification checklist.
The business action
"They export it as CSV or a branded PDF, send it to their broker or partner, and decide before the money moves. That is the whole product: know before you trade."
Show: Export menu, then the printed report.
Prepared answers to the hard questions
How accurate is it?
We do not claim an accuracy figure because we have not benchmarked one. Every output is labelled as an estimate requiring broker verification. Building that benchmark against real broker numbers is the first item in our 90-day plan.
How many customers do you have?
None. We are pre-pilot. What we have is a working product a judge can drive today, and a specific plan to convert that into pilot evidence.
What stops a broker or an LLM doing this?
Nothing stops either from answering — but neither produces a structured, provenance-labelled, exportable landed-cost analysis at the decision point. Brokers are also our distribution channel, not just a substitute.
Where does the data go?
Documents are read in the browser. Only extracted text and page images go to our server, which passes them to a third-party AI model provider. We store nothing server-side; history is on the user's device.
Source / note: These answers deliberately concede what we cannot prove. Do not improvise an accuracy figure or a customer count on stage.
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.