अत ← Portfolio Compass
Compass · AI-native RM onboarding · Wholesale banking

Helping relationship managers onboard a corporate group by deciding, not typing.

The AI reads what the bank already knows. The RM confirms only what it can't settle.

Live demo at Sibos 2026 Lead UX Designer · Purple Fabric, Intellect
Compass · decoding a client (recorded from the prototype)
01

Summary

What I achieved.

I turned a months-long, form-by-form onboarding into one guided route where the AI does the reading.

7systems read before a single question
1stage open at a time, whole route always visible
~100prototypes before the final
Sibosshown live, Sept 2026
02

Problem framing

Why it mattered.

The RM was a data collector, going back and forth between bank and client.

  1. 01Gather info from many sourcesRM types it
  2. 02Verify and documentRM types it
  3. 03Check group and entity recordsRM types it
  4. 04Map the business lifecycleRM types it
  5. 05Build the money flowRM types it
  6. 06Map recommended productsRM types it
  7. 07Configure each productRM types it
  8. 08Onboard users into groupsRM types it
Accumulate → verify → documentMonths · error-prone · repeated data

Starting point: the SME's framework as detailed spreadsheets and a form-based prototype.

03

Role & team

Who did what.

Me · Lead UX Designer
  • Co-defined the overall journey
  • Owned money flow, user groups, workflow
  • Built ~100 HTML prototypes
  • Also designed the upstream PM studio and RM360
UX Architect

Co-designer. The connected canvas view.

CEO & VPs

Direct review rounds. The Google Maps idea.

SME & stakeholders

The banking framework and requirements.

04

Approach

The strategy.

Google Maps for onboarding: see the whole route, work one checkpoint at a time.

Tap a leg. The others stay visible but quiet.

Routethe whole journey
Legsbands named as the RM's questions
Checkpointsone decision each
Alerts"items need you"
Decision over data entryAlways know where you areComplexity absorbed, not exposedNo overload at any stage
05

Challenges

What changed my mind.

Every version showed less, and claimed only what it could prove.

Prototype
1 / 7

Tried, then dropped · tap a card

A call I lost: where the AI speaks

The guide speaks at the bottom of every screen
The talking guide at the bottom of the screen

OutcomeI wanted a human, talking guide. The UX Architect argued the RM was already overloaded, so the bar went quiet and the AI's voice moved into each decision. My layout stayed: bottom bar for AI, right panel for decisions.

06

Solution

The journey, recorded from the prototype.

    My call

    07

    Results & impact

    What changed.

    Shown live at Sibos 2026, and visitors loved it.

    Shipped

    Live demo

    Hosted and presented at Sibos as RM Intelligence Studio.

    Quality bar

    34 principles

    AI-native design principles used as acceptance criteria. An independent review flagged 3 heavy defects; all fixed.

    Sibos feedback

    "Intriguing"

    Visitors were amazed by the experience, found it intriguing, and loved the overall journey and the way the agents work.

    PM Intelligence StudioAuthors industry templates
    →
    RM Intelligence StudioCompass, this journey
    →
    RM360The RM's overview

    I designed across all three, so the journey and what feeds it were shaped together.

    08

    Next steps

    What I'd do next.

    • Workflow edge casesRoles that match no group, overlapping amount bands, checking a generated approval chain.
    • Near-threshold matchesA quick review lane for people just above the auto-assign bar.
    • A voice for first useBring back the talking guide for an RM's first journey only.

    Client, people and identifiers are fictional (“Meridian Group”). Recordings are from my prototypes with client data replaced.