Deep Dive · Outcome

FXpress · AI Co-Pilot for Cross-Border Remittance

Property	Value
🏷️ Status	Complete · Live
🔗 Live Build	fxpress.netlify.app
🔖 Phase	Outcome
🎯 Purpose	The closing argument — the research-to-design-to-build trace end to end, what each major decision cost, and the bet at the center of the product

Why This Page Exists

The Outcome parent shows what shipped. Final Design shows the states and the live build at a glance. This page does something different: it makes the case for FXpress as a design artifact.

Read this if you want the strongest answer to two questions a hiring manager actually asks — why is this work serious? and what is the claim this product makes?

The Research-to-Design Trace

Every consequential decision in FXpress traces to a specific research finding. The chain is documented, not asserted — each row can be defended with a participant quote in the research artifacts, and the right-hand column can be verified by anyone who opens the URL.

Finding (Discover)	Design decision	Visible in the live build
F1 — Senders try to optimize timing and can't	Fexi's rate read sits on the canvas and surfaces without any user action	Yes — the widget's primary value proposition
F2 — Crypto is accepted only if invisible	Zero crypto vocabulary in the default flow; routing available on request only	Yes — no "USDC," "wallet," or "gas" anywhere in the widget
F3 — Recipient-amount certainty is non-negotiable	The recipient amount at the current rate, in the largest type on the canvas, before the send button is reachable	Yes — and restated in the hand-off modal's pre-fill chip
F4 — Automation tolerance is a spectrum	Two control paths (Send Now / Set Trigger); default is Send Now, automation is opt-in	Yes — both paths persist on the canvas
F5 — Trust is built through proof, not promises	Fexi describes and never forecasts; session-based delta and best-rate-seen flags	Yes — every read is grounded in observed history

Two structural conclusions from the same synthesis shaped the product's shape rather than a single screen:

Recipient familiarity constrains the sender's product. The recipient can never be asked to learn anything new — which is why FXpress hands off to Wise, Xoom, or the user's bank rather than moving money itself.
Two personas, one adaptive interface. The spectrum runs from Perry's refusal of any automation to Sadie's programmable rules, with Jenni and Kang as the two design anchors. A fixed default would have served one of them.

The Three Trust-Anchoring Decisions — With Their Costs

These three decisions make an AI-assisted money product defensible. They are decisions, which means each one cost something. The cost is part of the judgment.

1 · Fexi as co-pilot, not autopilot

What it costs. The product can't deliver the cleanest version of its own value proposition — set-and-forget automation that quietly saves money in the background. Every send requires explicit authorization. Sadie, the ceiling of the spectrum, would prefer more autonomy than she gets.