Microsteed Ration Wizard
Kentucky Equine Research
Case Study
Leveraging machine learning to automate treasury insights and reduce manual data wrangling.
CashPro Insights (now known as CashPro Data Intelligence) is an advanced decision-support capability within Bank of America’s CashPro platform, designed to help global businesses optimize their treasury operations through AI-driven insights and high-fidelity data visualizations.
Bank of America’s CashPro platform processes millions of corporate transactions every day. Each payment data point carries a stack of base‑level signals that help to identify what dimension these data points are describing and communicating.
A data point, such as a payment, is comprised of a pattern of signals, and when aggregated, exposes cost leaks, timing inefficiencies, and fraud risk. CashPro Insights was conceived to analyse both layers: signals attached to raw data and the emergent patterns those signals create in aggregate.
Our design goal was clear: transform these patterns into plain‑language recommendations that a CFO or Treasurer can absorb at a glance, yet still allow them to trace each recommendation back through insights, to the underlying signals, and down to the raw transactions if they desire full validation.
The result is a Financial Intelligence Dashboard that turns mountains of payment data into an embedded advisor, guiding corporate finance teams to optimize their payment mix, improve working capital, streamline cross‑border flows, and reduce fraud exposure—all within their existing CashPro workflow.
Additionally, a framework for scaling recommendations from the initial few to what could perhaps be thousands of ways a client could make tactical and strategic decisions that would result in operational efficiencies and cost savings.
Bank of America’s CashPro platform processes millions of corporate transactions every day. Buried in that activity are patterns—signals—that reveal cost leaks, timing issues, and fraud risk.
Bank of America recognized that while pockets of analytics (e.g. Excel macros, Tableau dashboards) existed, there was no central interface that aggregated data or provided strategic next steps. Teams were spending significant time deriving insights manually, particularly across domains like payments, working capital, FX, credit, and security. Moreover, there was an opportunity to leverage ML to surface patterns in transactional, temporal, and contextual signals. A clean, clear user interface was critical to translating these signals into strategic, user-friendly recommendations.
Clients needed a smarter, transparent, and actionable way to sift through transactional noise and drive decisions.
Large corporations process payments across various systems and countries. Important patterns – like redundant fees or suboptimal timing – were buried in spreadsheets and reports. Clients struggled to derive a clear value proposition from their raw payment data.
Companies often default to expensive payment methods (e.g. frequent wire transfers or paper checks) out of habit. Without actionable analysis, treasury teams lacked insight into how payment mix impacts cost. This led to inefficiencies and higher fees.
International payments introduced hidden FX fees, delays, and compliance hurdles. Finance managers had little visibility into how to optimize cross-border transactions or leverage local currency accounts to save money.
Timing of outbound payments and inbound receipts directly affects cash flow. However, finance teams had no easy way to pinpoint which payment practices were hurting their working capital (e.g. paying too early or not using available credit terms).
Detecting abnormal transaction patterns (potential fraud or errors) was difficult with manual monitoring. Busy operations leads could not easily spot out-of-pattern payments or potential fraud indicators among thousands of daily transactions.
Designing data experiences for corporate finance required constant calibration. We had to balance the need for abstraction—so users could grasp recommendations at a glance—with the need for traceability, allowing them to follow a clear line of reasoning back to raw data.
This tension also shaped how we handled information breadth vs. focus. Users needed a narrow entry point—typically a single metric or recommendation—yet also the ability to zoom outward into richer context. Each insight had to feel both precise and expandable, without burdening the user with unnecessary complexity.
A core challenge was crafting an information hierarchy that mirrors how treasury professionals think: moving from high‑level insight to low‑level evidence in a natural, explainable flow.
We defined and distinguished four layers of meaning—Recommendation → Insight → Signal → Raw Data—and aligned them with both users’ mental models and the platform’s data architecture. Without this structure, users would either drown in details or distrust conclusions. Our task was not merely arranging data, but designing a system that teaches users how to read it.
In treasury operations, what is shown matters only if users trust how and why it’s shown. Even the most accurate recommendation can be dismissed if its provenance feels opaque.
Building cognitive trust meant surfacing evidence without overwhelming users, and supporting auditability without sacrificing clarity. The interface had to project integrity—making complex modeling feel honest and defensible to busy finance teams.
Throughout the lifecycle of CashPro Insights, the design process evolved in tandem with the discovery and refinement of design signals. Initially, our focus was on understanding raw payment data and its associated attributes, but as the project progressed, we reached pivotal moments where we could begin to deconstruct and categorize data into actionable signals. These signals served as the foundation for both refining existing visualizations and exploring novel data storytelling methods.
At each stage, signals acted as a cue to rethink both content and form—empowering users to engage with data intuitively and make reliable, context-aware decisions for treasury optimization.
To validate the signals chosen during our iterative design process, we conducted in-depth user walkthroughs, testing not only the visuals but the interpretability and contextual relevance of each design signal. During these sessions, we engaged users with targeted, open-ended questions to probe their understanding:
This approach helped us quickly identify designs where the signals didn’t land or didn’t support user goals effectively. We adjusted accordingly—sometimes in micro-iterations—to ensure that users could extract value effortlessly.
As a result of a series of user interviews, we also refined personas to understand how different roles responded to these design signals:
These insights about design signal perception informed how we crafted and refined visual hierarchies throughout the product.Success in Simplification
One key success was in Payment Efficiency. Initially, we visualized payment type mix as a full horizontal bar chart, grouped by paper (e.g., checks) and electronic payment types (e.g., credit, wire, ACH). Through validation, we discovered that only spatial comparison between bar lengths, essential labels, and amounts/fees were necessary. Superfluous elements—axis lines, series labels—were stripped back. This yielded a cleaner, more intuitive visualization, aligned precisely with user needs and the design signals driving their decisions.
We used the Design Signal Method to determine how to communicate and visualize data points and refine insight visualizations. This framework helped guide the visual and interactive representation of each insight, ensuring that users could understand the data clearly and act on it effectively.
This design signal methodology allowed us to create visual consistency across the platform while enabling flexibility to adapt and innovate based on user needs and data complexity. The result was a highly contextual and user-aligned visualization strategy, directly supporting the goals of treasury optimization.
Throughout the process, close collaboration between design, product, and engineering was crucial. We frequently sat with data scientists to translate complex analytics into digestible UI components. By the end, we had a validated design that was not only usable and useful, but also credible – a critical factor for adoption in the finance domain.
The presentation of insights - calculated measures derived from raw data and signals, as well as suggestions/recommendations on what actions the treasury teams can take, is an accelerator for subscribed CashPro clients to optimize their treasury operations across the provided focus areas.
Bank of America’s CashPro platform processes millions of corporate transactions every day. Each payment data point carries a stack of base‑level signals that help to identify what dimension these data points are describing and communicating.
A data point, such as a payment, is comprised of a pattern of signals, and when aggregated, exposes cost leaks, timing inefficiencies, and fraud risk. CashPro Insights was conceived to analyse both layers: signals attached to raw data and the emergent patterns those signals create in aggregate.
Our design goal was clear: transform these patterns into plain‑language recommendations that a CFO or Treasurer can absorb at a glance, yet still allow them to trace each recommendation back through insights, to the underlying signals, and down to the raw transactions if they desire full validation.
The result is a Financial Intelligence Dashboard that turns mountains of payment data into an embedded advisor, guiding corporate finance teams to optimize their payment mix, improve working capital, streamline cross‑border flows, and reduce fraud exposure—all within their existing CashPro workflow.
Additionally, a framework for scaling recommendations from the initial few to what could perhaps be thousands of ways a client could make tactical and strategic decisions that would result in operational efficiencies and cost savings.
In data science, what are often referred to as “dimensions” or “attributes” can be understood here as signals. Signals describe the specific qualities or dimensions a data point conveys. These qualities can be leveraged to explore anomalies or identify patterns across data sets. In our design context, we treat signals not just as descriptors but as the basis for design decisions—translating them into effective visualization patterns that help communicate the underlying record or transaction with clarity.
Signals as descriptors of a single data point
Every payment embodies at least five descriptive layers: it is transactional (an exchange of value), temporal(happens at a precise moment), contextual (ACH vs. wire, domestic vs. cross‑border, etc.), inferred (risk score or forecast added by modelling), and behavioral (the pattern of how often that payment type occurs). Labelling those layers up front tells us which story the data point can tell and informs the visual grammar—bar, line, heat‑map, gauge—that will communicate it best.
Signals as patterns across many data points
A single payment’s signals matter, but clusters of similar signals expose cost leaks, timing issues, or fraud. We therefore differentiate base‑level signals (attached to raw records) from pattern‑level signals (aggregations and correlations). This distinction underpins our four‑layer hierarchy—Recommendation → Insight (pattern) → Signal (data point) → Raw data—so users can trace every conclusion back to first principles.
In data science, signals refer to meaningful information extracted from raw data. In the context of UX, we use these extracted signals to extrapolate meaning, inform user understanding, and guide design decisions. Thus, for our purposes, signals are both data descriptors and tools that enable effective communication through design.
Signals steer us toward well‑known visualization patterns—bars for magnitude, lines for trends—and inspire new visual forms when conventional charts fall short. Having a shared signal vocabulary lets designers and data scientists jointly decide when to reuse a tried‑and‑true chart and when to invent a bespoke graphic that better reveals a relationship.
Signal Type | Example in CashPro | Typical / Novel Visual |
|---|---|---|
Transactional | Fee per wire transfer | Baseline bars or lollipop chart |
Temporal | 6‑month ACH‑volume trend | Area chart with moving‑average overlay |
Contextual | ACH cost vs. industry avg | Bullet chart or slope graph |
Behavioral | User approval timing patterns | Heat‑map calendar |
Inferred | Predicted fraud probability | Gauge, radial band, or risk badge |
When stakeholders asked “How should we show FX corridor efficiency?” we replied “Which signals tell that story?” By mapping signals first, the team chose visuals with intentionality. Every pixel now earns its place—simplifying complexity while preserving traceability.
50%
Reduction in Manual Work
30%
Cost Savings
95%
User Satisfaction
Designing CashPro Insights was a masterclass in blending data science and UX design. One major takeaway was the importance of trust and transparency when introducing AI-driven insights in a conservative domain like corporate banking. We learned to never present a “black box” recommendation – users like Dave, the data analyst, quickly challenged any insight that wasn’t clearly backed by data. By implementing layered drill-downs and explaining the “why” behind each suggestion, we gained user trust. As a result, even skeptical users began to rely on the dashboard, knowing they could always peek under the hood if needed.
Another takeaway was the value of focusing on the user’s context and workflow. Initially, we weren’t sure if busy finance folks would log into a separate dashboard regularly. By integrating insights into the existing CashPro portal (and planning for features like email alerts or in-app notifications for new recommendations), we fit into the users’ natural routine. This taught us that the best insights are useless if they don’t reach users at the right time and place. Our design had to meet users where they already are – in this case, right when they log in to manage their daily cash operations.
Collaboration was also key. We navigated challenges around data availability and accuracy by maintaining a tight feedback loop with engineering. When the data scientists informed us that some data would only update monthly, we adjusted our designs to set the right expectation (e.g. labeling insights as “This month’s analysis” and avoiding real-time language). This project reinforced that designing with real data early prevents creating pie-in-the-sky solutions that can’t be implemented. By prototyping with actual (anonymized) client data from the start, we uncovered technical constraints early and designed around them.