Portfolio brief

Commercial Judgment Infrastructure

Yawen Gao
May 2026
High-Touch Sales / Luxury E-commerce
Series thesis
The institutional asset is execution-linked expert judgment: a structured record of what a professional chose, for whom, in what context, how it was framed, how it was adapted, and what happened after. This data cannot be bought as a dataset; it only exists when real experts make real decisions inside real workflows.
Problem

Expert judgment is exercised daily, then disappears.

High-touch commercial work depends on expert judgment, yet most systems preserve catalog facts, customer behavior, CRM fragments, and final outcomes while losing the professional decision that connected them.

The gap appears when a human chooses among plausible actions: send or hold, push or soften, approve or deny, assign or monitor, escalate or resolve.

Design constraint

The capture mechanism must originate from real work.

The system should not ask experts to explain themselves after the fact. It should give them a better workflow surface, then emit an event record as a byproduct of the action.

The tool captures the trace of judgment. It does not claim to capture the full internal state of the expert.

Proposed architecture

One shared execution spine connects capture, operating intelligence, and Product / Client learning loops.

SYSTEM VIEW WP01 Real workflow capture CEE from client-facing work WP02 Operating layer CEE + selected ODE Expert judgment events captured from real workflows CEE + selected product/client ODE WP03 Shared execution spine common event structure Learning Product Client Future workflow context / guidance surfaces in the next judgment moment common structure
Core data objects
  • Commercial Execution Event / CEE: client-facing commercial action with product set, recipient logic, message frame, timing, adaptation, tracking, and attributed outcome.
  • Operating Decision Event / ODE: workflow judgment such as hold, watch, avoid, prioritize, readiness, approval, benefit, policy, or review decision. Only selected product/client execution-related ODE enters the first Dual Intelligence MVP.
  • Expert Judgment Event: umbrella term for captured expert judgment in workflow, including CEE and selected product/client execution-related ODE.
Learning signals
  • Parent-Child Edit Diff: what changed when a human-created parent edit became a human-adapted child edit.
  • Copilot diff: the future comparison between a system proposal and the advisor's final edit or decision.
  • Outcome evidence: open, click, purchase, silence, return, delayed conversion, unsubscribe, revenue, and other attributed outcomes.
First MVP boundary
Primary workflow userHigh-touch sales produces the CEE stream and selected product/client execution-related ODE context.
First learning layerDual Intelligence is the first company-learning slice: product learning and client learning connected through expert judgment events.
Reuse boundaryOther functions first consume validated, aggregated, or read-only insights from Product Loop and Client Loop outputs.
Product Loop

Learns how products function in expert-led selling: anchor role, complement role, risk role, message sensitivity, client-state fit, pairing behavior, and commercial context sensitivity.

Client Loop

Learns context-specific client and segment state: category readiness, timing sensitivity, brand fatigue, purchase saturation, tone sensitivity, relationship risk, and service sensitivity.