SmartKatch is an AI-powered financial risk intelligence platform that helps enterprises detect financial errors, fraud, anomalies, and compliance risks. It continuously analyzes transactions across invoices, payments, procurement, accounting, and other financial processes. Using AI-driven analytics, pattern detection, and semantic matching, SmartKatch identifies risks that traditional rule-based systems may miss. The platform enables organizations to prevent financial leakage, strengthen controls, and automate continuous auditing.
My Role
As a Design Consultant, I led end-to-end UX design from problem framing and ideation to task analysis, interaction design, visual design, and detailed execution. Worked with CXOs, Engineering and Product team.
Product Features
-
Predictive analytics : analyzes historical and time-series transaction data to identify abnormal patterns and predict potential risks.
-
Semantic matching : goes beyond exact matching to identify near matches and variations in transaction data, useful for invoice and duplicate detection.
-
Pattern & anomaly detection : continuously looks across transactions to identify unusual behavior, suspicious activities and potential fraud.
-
Continuous monitoring : rather than periodic sampling, SmartKatch analyzes financial data continuously/near-real-time and flags exceptions for action.
-
Forensic analysis : investigates transaction patterns and provides insights into the potential root causes of financial exceptions.
-
Actionable intelligence : converts detected risks into alerts, recommendations and corrective/preventive actions rather than simply reporting anomalies.
Existing Design
-
Legacy, tabular interface: The experience relies heavily on dense tables and static data blocks, reflecting an outdated enterprise dashboard pattern rather than a task-oriented workflow.
-
Limited contextual information: Numbers are presented without sufficient context, making it difficult for users to understand what happened, why it matters, and where they should focus.
-
Poor mental-model alignment: The grouping of Scanned, Potential Errors, Errors, Prevented, To Recover, Closed, etc. requires users to interpret the system's terminology and manually build their own understanding of the audit process.
-
No clear next actions: The interface primarily reports status but doesn't guide users toward what needs attention next, which exceptions to investigate, or how to resolve them.
-
Weak visual hierarchy: Important insights and high-risk issues don't stand out clearly. The uniform presentation of information makes prioritization difficult.
-
Outdated visual language: The pale blue/grey color palette, borders, dense grids, and minimal use of visual indicators create a dated enterprise-software experience.
-
Limited interaction and exploration: The dashboard behaves largely as a reporting screen rather than an interactive investigation workspace where users can drill into anomalies and understand their underlying evidence.
-
No AI-assisted intelligence: The existing experience primarily presents aggregated results; it doesn't leverage AI to detect patterns, explain anomalies, prioritize risks, recommend actions, or assist auditors in their investigation.
User Persona
1. Finance Controller / CFO
Goal: Maintain financial accuracy, reduce leakage, and ensure compliance across business transactions.
Needs: Executive-level visibility, risk trends, audit trails, and actionable insights without reviewing thousands of transactions manually.
Pain points: Fragmented financial data, delayed audits, hidden leakage, and difficulty identifying high-risk transactions.
2. Accounts Payable Manager
Goal: Process invoices accurately and efficiently while preventing duplicate payments and overbilling.
Needs: Automated invoice validation, PO matching, exception identification, and clear explanations for discrepancies.
Pain points: Manual reconciliation, high invoice volumes, duplicate invoices, pricing/quantity mismatches, and time-consuming investigations.
3. Internal Auditor / Forensic Auditor
Goal: Identify financial irregularities, fraud patterns, and control weaknesses.
Needs: AI-assisted investigation, transaction relationships, anomaly detection, evidence trails, and explainable findings.
Pain points: Sampling-based audits, large datasets, disconnected evidence, and significant time spent tracing the root cause of exceptions.
4. Finance Operations Analyst
Goal: Resolve exceptions quickly and keep financial processes moving.
Needs: Prioritized alerts, recommended actions, supporting evidence, and collaboration with vendors and internal teams.
Pain points: Too many alerts, unclear priorities, repetitive investigation, and switching between multiple systems.
Design Goals
Design an AI-powered financial risk intelligence platform that transforms complex, manual auditing into a transparent and proactive workflow.
The experience should enable finance teams to connect their data sources, configure AI agents, automatically analyze transactions, identify anomalies and exceptions, explain the underlying evidence, and route high-risk cases for human review—while keeping humans in control of important financial decisions.
Key UX objectives
-
Simplify complexity: Turn complex AI and financial analysis into intuitive workflows.
-
Make AI explainable: Show why an invoice was flagged and provide supporting evidence.
-
Prioritize risk: Help users focus on high-value and high-risk exceptions rather than reviewing everything.
-
Keep humans in control: Allow auditors to validate, override, investigate, and approve AI findings.
-
Create an audit trail: Maintain a clear history of AI analysis, evidence, decisions, and actions.
-
Enable proactive auditing: Move from periodic, sample-based audits to continuous transaction monitoring.
AI Workflow Builder — Design Approach
As a product designer, I designed the AI workflow builder around the mental model of accountants rather than the complexity of the underlying AI system. Instead of exposing technical AI configurations, I structured the experience as a simple, visual sequence of familiar accounting activities —Invoice Intake → PO Matching → Validation → Forensic Analysis → Exception Detection → Human Review → Report.
I used a node-based workflow to make the relationship between AI agents, data sources, and outcomes immediately understandable. Each agent represents a clear business task, while contextual panels provide configuration, data sources, and explanations without disrupting the main workflow.
The interaction flow follows a progressive-disclosure approach: accountants can first understand the overall process, then select an individual agent to configure or investigate it. The workflow clearly communicates what the AI is doing, why it is doing it, and what action the user needs to take next.
The goal was to make a traditionally complex, technical AI workflow feel familiar, transparent, and controllable for finance users—without requiring them to understand how the AI works underneath.
Redesign of Existing Functions
