CASE STUDY
Machine Learning Data Modeling & Forecasting for Foreign Exchange Risk Exposure
Executive Summary
When working as a Principal Product Design Consultant at Kyriba, I led the end-to-end redesign of the FX Exposure capability—transforming it from a basic data-store into an intelligent, ML-driven forecasting and approval workflow. By centralizing historical and projected data, embedding variance tolerance alerts, and automating multi-level approvals, we reduced forecasting cycles from months to days and drove widespread feature adoption. This project earned CEO endorsement, accelerated treasury decision-making, and was a key differentiator in Kyriba’s acquisition.
Project Context
Company / Product
Kyriba is a leading cloud-based treasury and finance management platform. I worked on the FX Exposure capability, which identifies and manages foreign-exchange risk.Timeline & Team
This project started in 2019 and ran for 6-months.
Cross-functional team of ~20: product managers, business-development managers, engineers (front-end, back-end, DevOps, QA), data scientists, & sales engineers.My Role
Principal Product Designer embedded in the R&D mission team. Responsibilities spanned UX research, ML/AI product strategy, prototyping, hi-fi design, and stakeholder alignment (3 PMs, 2 BD leaders, Sales Director, CEO, CTO, & VP of Engineering).
Problem & Goals
Business Problem
Existing FX Exposure feature had 35% adoption; analysts defaulted to Excel because the in-app worksheet only stored numbers and required manual approval outside of the platform.User Problem
Financial analysts juggled ≥5 tools and windows to model exposures, manually emailed spreadsheets for sign-off, leading to errors, delays, and poor visibility.Success Metrics
Adoption → majority of treasury team within two quarters
Forecast cycle time ↓ from 3-4 months to 1 month
Manual‐entry errors ↓ significantly
Stakeholder satisfaction ↑ (CFO endorsement)
Existing Kyriba FX Cash Position Worksheet
An example of a user using Excel to model FX Data externally
Users & Research
User Profiles / Personas
Financial Analyst – Builds exposure models; needs accuracy and speed.
Treasury Manager – Reviews forecasts; needs consolidated dashboards and governance.
CFO – Monitors net exposure; needs audit trails and strategic insights.
Methods & Key Insights
Onsite follow-me-home visits with 35 customers revealed 5+ open spreadsheets/windows per task.
Pendo, Heap, & Qliksense analytics showed a sharp drop-off at the manual approval step.
Primary insight: a centralized workflow engine and intelligent forecasting would eliminate context-switching and manual hand-offs.
Design Process
Ideation & Sketches
Facilitated workshops with PMs, BD, and sales to define ML forecasting scope and approval-flow requirements.Wireframes & Low-Fi Prototypes
Rapid mid-week sprints in Sketch → InVision; feedback loops with 3 PMs, 2 BD leads, and Director of Sales.Iterations & Trade-Offs
Balanced full-featured algorithm controls vs. cognitive load; opted for a simplified “variance tolerance” slider.
End to End Proposed Sitemap
Onboarding Experience Wireframe
Subsidiary Staffer Workspace
Treasury Analyst Workspace
Solution & Design
High-Fi Mockups
Data-modeling worksheet with integrated historical/projection panels
Variance alerts dashboard highlighting out-of-tolerance forecasts
Surface hedge recommendations based on real-time FX analytics
Automated approval workflow UI for analyst → manager → CFO
Role-based views and permissions
Interaction Highlights
Real-time variance coloring, one-click approve/reject, drill-down to historical drivers.
Outcome & Metrics
Quantitative Results
Forecasting cycle time shrank from ~3.5 months to <1.5 weeks
Feature adoption rose from <35% to ~83% of target users in Q1 post-launch
Qualitative Feedback
“Finally, one platform—I can see all the historical forecasting data in one place and this cuts my analysis work down almost completely. This trend insight is incredibly helpful and reduces my calls to my treasury manager for their opinions…”
– Senior Treasury Analyst, Customer“The forecasting review and approval process used to take us months of meetings and refinements, your new features have helped us to cut down on meetings and has saved us so much time because we can now easily do the review process through your platform…”
– Chief Financial Officer, Customer“This is a game-changer for our platform.”
– Jean-Luc Robert, CEO of Kyriba
Reflection & Next Steps
What Went Well
Rapid alignment across Product, Engineering, and Executive Leadership
Quick ML concept validation via BI-tool proof-of-concept
What I’d Do Differently
Conduct A/B tests on algorithm-threshold settings
Integrate accessibility audits earlier
Thanks for Reading
Check out other case studies below or visit the Case Studies page.