In enterprise product design, few environments carry as much cognitive pressure as financial fraud detection.
At Amazon’s Transaction Risk Management Systems (TRMS), manual fraud investigators stand on the front lines protecting millions of customers and global marketplace transactions. Their daily job is intense: they evaluate accounts flagged by risk algorithms, spot subtle patterns of fraudulent behavior (such as account takeovers, unauthorized chargebacks, or coordinated syndicate rings), and decide whether to clear an account or restrict it.
When I set out to lead the foundational UX research for our next-generation investigation platform—Nautilus BRI (Buyer Risk Investigations) 2.0—the challenge wasn't just aesthetic.
Investigators were managing high-volume queues while toggling between fragmented legacy tools, external databases, and internal spreadsheets. Every unnecessary click, every buried data point, and every disjointed transition didn’t just slow investigators down—it introduced cognitive fatigue that could lead to costly false positives or missed fraud.
To redesign the core investigation experience, I conducted an in-depth global research study with 30 investigators.
Here is how we structured the research, what we uncovered, and how we translated qualitative behavioral insights into an enterprise platform that reduced time-at-task while boosting decision accuracy.
The Research Strategy: Studying the Full Spectrum of Expertise
To design a tool that works at enterprise scale, you cannot rely on a homogeneous user sample. A seasoned tenured investigator navigates risk patterns through intuition and muscle memory, while a newly onboarded investigator relies strictly on standardized documentation and step-by-step SOPs.
I structured a multi-phase qualitative study across 30 participants worldwide:
Tenured Senior Investigators (US): Power users handling complex, multi-layered fraud queues.
New-Hire Investigators (US & International): Freshly trained investigators navigating the system for the first time, exposing friction points in onboarding and information architecture.
Global / Non-US Investigators: Specialists operating in regional international marketplaces with unique fraud trends and localization challenges.
Investigator Advisors & Leads: Senior mentors who audit decisions, track operational metrics, and manage escalation queues.
Methodology
Rather than relying on self-reported surveys, I used a hybrid methodology:
Contextual Inquiry & Shadowing: Observing investigators live as they processed real-time queues—capturing their unspoken workarounds, dual-monitor setups, and offline notes.
Task & Cognitive Walkthroughs: Mapping every micro-task across the investigation lifecycle: from initial queue sign-in, case profiling, and linked-account discovery, to final decision logging.
Iterative Design Focus Groups: Bringing mid-fidelity prototypes back to subsets of the cohort over several months to validate interaction patterns before writing production code.
Key Insights: Where the Cognitive Friction Lived
Watching 30 investigators work through hundreds of fraud cases revealed three critical systemic bottlenecks:
Insight 1: The "Swivel-Chair" Tax
Investigators were constantly jumping between disconnected legacy applications (such as Jupiter and IW) and manual spreadsheets just to assemble a single customer story. Gathering order histories, comparing billing addresses, and verifying linked compromised accounts required dozens of open browser tabs. This context switching was the single largest contributor to latency and cognitive fatigue.
Insight 2: Data Fragmentation vs. Decision Velocity
Investigators didn’t suffer from a lack of data; they suffered from unstructured data abundance. Critical risk signals (e.g., rapid device ID shifts or mismatched IP geolocations) were displayed with the same visual hierarchy as routine, low-risk account metadata. Investigators were forced to visually scan walls of text to find the needles in the haystack.
Insight 3: The Documentation Bottleneck
After making a decision, investigators spent up to 30% of their case time manually annotating their rationale, copying ticket IDs, and updating external queue trackers. This administrative overhead drained energy that should have been spent on high-risk analysis.
"Great enterprise UX doesn't mean simplifying complex work until it's superficial. It means structuring complexity so the user's brain can focus entirely on high-judgment decisions."
The Design Framework: Transforming Insights into Nautilus BRI 2.0
With our engineering, product, and operations partners aligned around the research findings, we translated these insights into four core architectural improvements:
A Unified Investigation Canvas: We eliminated the multi-tool shuffle by integrating core risk mining tools into a single, modular interface. Everything needed to resolve a case was aggregated into one structured view.
Context-Aware Risk Hierarchy: Instead of flat data tables, we introduced visual risk indicators that dynamically bubble up anomalies (e.g., velocity spikes, linked compromised credit cards) while keeping secondary data accessible via progressive disclosure.
Relational Account Mapping: We transformed dense text logs of related accounts into an interactive visual graph, allowing investigators to instantly trace syndicates and shared entity fingerprints with a single click.
Automated Decision Logging: We automated routine documentation by auto-generating case summaries based on the investigator’s recorded actions, saving minutes per case and improving audit consistency across global sites.
Measurable Business Impact
By rooting the redesign in deep ethnographic research and validating concepts iteratively with the team, Nautilus BRI 2.0 delivered tangible operational ROI:
Substantial Reduction in Time-at-Task: Eliminating tool-toggling and streamlining data visualization allowed investigators to evaluate cases significantly faster.
Higher Decision Accuracy: Elevating high-risk signals reduced false positives, protecting legitimate buyers while intercepting fraudulent actors earlier.
Accelerated Onboarding: New hires reached baseline operational velocity weeks faster because the system guided them through the natural investigation workflow rather than requiring extensive memorization of legacy shortcuts.
Global Operational Coherence: Standardized the investigation process across US and international operational centers, providing leadership with unified telemetry.
Takeaways for Enterprise Design Leaders
Designing for enterprise experts requires humility and rigor. When users work in your software 8 hours a day in high-stress conditions, standard consumer UX heuristics aren't enough.
Observe the workarounds: The post-it notes on a user's monitor and the spreadsheets they keep on their desktop are the clearest roadmaps for where your product is failing.
Design for cognitive stamina: Speed is a byproduct of clarity. When you minimize cognitive friction, efficiency and accuracy naturally follow.
Bring research directly to engineering: Sharing raw user recordings and journey maps with product managers and engineers builds shared empathy, transforming design reviews from subjective debates into user-centered problem solving.
I lead product design and research for complex enterprise platforms, AI-enabled workflows, and large-scale systems. If you're building products that turn complex data into intuitive experiences, let's connect on LinkedIn or reach out via email.