Shawn Chan

Shawn Chan is a cross-border investor who argues that financial AI should produce defensible investment memos. At AI Engineer World’s Fair 2026, he was listed as Vice President at China Resources Holdings, leading a consumer-sector fund. In his talk, Build for the Memo, Not the Demo, he presented the investor’s perspective: deciding whether to trust a system and the founders selling it.[2][3]

Chan distinguishes a demonstration built around one clean document and a fluent answer from the memo a committee reads before capital moves. The latter draws on filings, transcripts, broker notes, spreadsheets and informal call notes, often containing conflicting information. In his view, success means producing a recommendation that survives an argument, with evidence and assumptions available for scrutiny.[3]

Source authority must travel with the claim. Chan recounts watching an expensive AI tool quote a rough estimate from a group chat six months earlier while the audited number was available in the filings. His objection is that relevance to a question does not establish evidentiary weight: an audited filing, an analyst note and an internal message should not receive interchangeable treatment.[3]

Numerical discrepancies can undermine trust in the entire review process. Chan describes a memo reporting revenue growth as 18% on its first page and 17.4% in a later table; in his account, the discrepancy raised doubts about whether harder assumptions had been checked. He also recalls an earnings-call figure conflicting with a filing, caught only because someone happened to have both documents open. His response was explicit: “Luck is not a control.” Systems should expose disagreements for human investigation.[3]

Uncertainty must survive editing. Chan recalls an expected approval becoming an assertion that approval had been received across three memo drafts. The approval did not arrive on schedule. He calls for labels that keep estimates visibly distinct from facts even when passages are copied into another person’s slides weeks later.[3]

Chan sets out five requirements he would demand before allowing a vendor’s system near a live deal:[3]

• Claim-level evidence: each sentence links directly to its source paragraph, with the source’s trust level attached.[3]
• Visible facts and estimates: the reader can distinguish established information from judgment at a glance.[3]
• Automatic numerical agreement: the system refuses to release a memo containing inconsistent figures.[3]
• Explicit contradictions: disagreements between sources are raised for review rather than silently smoothed away.[3]
• Logged human approval: an audit trail records who reviewed the memo, what changed and when the reviewer signed.[3]

For Chan, the practical verification test is whether one click reaches the exact paragraph supporting a challenged sentence. As he puts it, “the click-through is the product.” He applies the same discipline to fundraising: after a founder leaves the pitch meeting, the deck becomes an internal investment memo, and its numbers are checked against the data room.[3]

Chan’s 2026 argument draws on earlier investment work. He described fifteen years of cross-border deals and attendance at approximately 200 investment-committee meetings. His conference biography identifies experience in mergers and acquisitions, public offerings and strategic investments across Hong Kong, mainland China, the United Kingdom and the United States, and a previous role as Head of Investment at A.S. Watson Group, part of CK Hutchison.[2][3]

He recalls an early-career presentation in which a polished banker showed confidently wrong numbers. A senior participant asked, “Where does this number come from?” The banker could not provide a usable answer. Chan uses that experience to explain his standard for financial AI: trust depends on being able to substantiate the number placed before the decision-maker.[3]