← back
Build for the Memo, Not the Demo — Shawn Chan, China Resources Holdings
Takeaway
Financial AI earns trust through traceable evidence, consistent numbers, and visible uncertainty.
Summary
- Investment memos must withstand scrutiny across conflicting filings, transcripts, and informal notes, whereas polished demonstrations can conceal unsupported claims.
- Financial AI should distinguish authoritative audited sources from analyst opinions and informal guesses instead of relying solely on semantic similarity.
- Numeric consistency across a document is a trust signal, and contradictory sources should be surfaced for human examination.
- The proposed trust checklist includes separating facts from guesses, quickly proving claims, and identifying the person accountable for the decision.
financial-aiprovenancetrust
Original description
When a company is about to spend $100 million, Shawn Chan is the person in the room deciding whether it happens, and after fifteen years and roughly 200 investment committees he can spot the flaw fast. Now some of the decks in front of him are AI generated: cleaner formatting, never defensive, and confidently wrong. His frame is the difference between a demo and a memo. A demo is a beautiful confident slice built to impress; a memo is what survives a room whose entire job is to find what breaks. One marketing demo with a single wrong sentence, he reminds you, once erased enormous value the moment the market noticed. So he lays out what an AI product needs to pass the memo test. Not every source deserves equal trust, since an audited filing is not a number from a group chat. Figures have to reconcile, and a contradiction between two numbers is not a bug but the most important signal in the room. Facts and guesses must stay labeled, because across three drafts an estimate quietly hardens into a fact nobody remembers assuming. Attach provenance rather than a citation tab, lock who claimed what, and the product that wins is the one that lets someone approve a deal without opening seven tabs at midnight. Speaker info: / shawn-chan-2b58a9129 Timestamps: 0:00 - A fifteen year confession 3:27 - Two hundred investment committees 4:22 - Money follows trust 5:46 - Two machines: the demo and the memo 7:31 - When one wrong sentence costs billions 9:45 - Not every source deserves equal trust 11:22 - The numbers have to agree 14:59 - Facts and guesses stay separate 17:06 - When fake citations reach a court 19:21 - The airline chatbot that made a promise 21:30 - Attach provenance, not a citation tab 22:52 - The products that will win