Partnership • Product Design • AI • UI/UX

AI-Powered Anomaly Detection for Workiva

Revolutionizing how financial professionals identify data inconsistencies in collaborative spreadsheets through intelligent automation and intuitive design.

Duration
3 months
Team
Nandha, Ranga
Role
Product Designer, UX Researcher
Tools
Figma, Python, Tableau
Focus Area
Anomaly Detection

The Challenge

Workiva users were spending 73% more time than necessary reviewing collaborative reports for data anomalies and formatting inconsistencies. With datasets growing exponentially and compliance deadlines tightening, manual detection methods were failing at scale.

73%
Additional review time
6
User interviews conducted
30%
Increase in revision cycles

Research & Discovery

Understanding the Problem Space

Through extensive user research with Business Analysts and MIS professionals, I uncovered critical pain points in the current anomaly detection workflow. Users were relying heavily on domain knowledge and manual scanning, leading to inconsistent results and missed outliers.

1

User Interviews

6 participants across Business Analysts and MIS students using structured 18-question framework

2

Thematic Analysis

Identified 3 core themes: automatic flagging, learnability, and visualization needs

3

Competitive Analysis

Evaluated existing solutions and identified gaps in enterprise-grade anomaly detection

Key User Insights

"Ideally it would tell me what the problem could be as well. So let's say there's an increase in acquisitions this month. Maybe we can tell me that it's possibly because this variable seems to have too many missing values."

— Business Analyst, Interview Participant

"An alert system basically like the moment it finds out an alert and it should be real time. It should actually tell you where exactly is the error instead of just telling that there is something wrong."

— MIS Graduate Student, Interview Participant

1

Flagging

Users need automated systems that can identify anomalies and provide contextual explanations for why data points are flagged as unusual.

2

Real-time Alerts

Immediate notification systems that pinpoint exact locations of errors rather than general warnings about data quality issues.

3

Visual Hierarchy

Clear color-coding and visual indicators that make anomalies immediately recognizable within large datasets.

User Personas

Novice User Persona

Sarah Mitchell

Age: 27
Role: Junior Analyst
Experience: 1 year
Tech: Moderate
"I need tools that guide me through the process and explain what's happening at each step."

Goals

  • Learn anomaly detection best practices quickly
  • Gain confidence in identifying data issues
  • Understand why certain data points are flagged

Pain Points

  • Overwhelmed by complex datasets
  • Unsure which anomalies are critical
  • Needs contextual help and guidance

Tools Used

Excel Workiva Google Sheets
Experienced User Persona

David Chen

Age: 42
Role: Senior Analyst
Experience: 15 years
Tech: Advanced
"I need speed and accuracy. Show me the anomalies, let me decide what to do with them."

Goals

  • Process large datasets efficiently
  • Customize detection parameters for specific use cases
  • Integrate findings into automated workflows

Pain Points

  • Manual detection is time-consuming
  • Inconsistent results across team members
  • Difficulty prioritizing high-impact anomalies

Tools Used

Workiva Python Tableau SQL

Wireframes

Wireframes

Low-fidelity wireframes exploring different approaches to anomaly visualization

Design Solution

AI-Powered Anomaly Detection Interface

The solution integrates machine learning algorithms with an intuitive interface that provides real-time anomaly detection, contextual explanations, and actionable insights directly within the Workiva spreadsheet environment.

Design Solution

High-fidelity prototype of the integrated anomaly detection system

Impact & Results

The solution addresses critical workflow inefficiencies and positions Workiva as a leader in AI-powered financial data analysis.

45%
Reduction in review time (projected)
89%
Anomaly detection accuracy

Key Learnings