Workiva

Workiva is a SaaS company whose platform connects data, documents, and teams so organizations can manage financial reporting, sustainability/ESG reporting, audit, risk, and broader GRC work in one place. They describe this as “assured integrated reporting.”

Team Members

Team of 2

Technologies

Figma

Python

Tableau

Power BI

Figma

Python

Tableau

Power BI

Figma

Python

Tableau

Power BI

Duration

4 Weeks

What are we going to do?

Spreadsheets are used everywhere for tracking sales, managing budgets, organizing data, and more. But when spreadsheets grow large, it becomes hard to spot mistakes or unusual patterns. For example, if a product suddenly sells 5 times more than usual or a monthly expense doubles without reason, how can users quickly catch these issues before they become costly problems?

The Challenge

At Workiva, we aim to simplify how people work with data. Your challenge is to design an AI-powered anomaly detection solution that helps users quickly spot unusual patterns (anomalies) in spreadsheet data, without needing to be a data expert.

What are we going to do?

Design Process

Empathize: Conducted user interviews to identify real challenges in detecting anomalies in spreadsheets.

Define: Synthesized findings into key pain points and user needs using affinity mapping.

Ideate: Generated feature ideas and prompt types based on user needs and trust concerns.

Prototype: Created high-fidelity screens with guided flows, anomaly summaries, and contextual actions.

Test: Due to time constraints, we couldn’t conduct formal usability testing. Instead, we relied on internal critiques and quick feedback loops during prototyping to keep the design aligned with user needs.

Research Phase

User Interviews

Affinity mapping

User personas

User journey map

User Interview

Key Insights Across Participants

Desired AI Capabilities:

Visual, intuitive anomaly notifications integrated into spreadsheet views.
Clear, understandable explanations for each anomaly flagged.
Flexibility and user control over anomaly detection settings and actions.

Visual, intuitive anomaly notifications integrated into spreadsheet views.
Clear, understandable explanations for each anomaly flagged.
Flexibility and user control over anomaly detection settings and actions.

Visual, intuitive anomaly notifications integrated into spreadsheet views.
Clear, understandable explanations for each anomaly flagged.
Flexibility and user control over anomaly detection settings and actions.

AI Trust & Transparency:

Users trust AI more if provided transparent, detailed reasoning behind flagged issues.
Preference for hybrid approach: AI suggests anomalies and solutions, but users retain decision-making control.

Users trust AI more if provided transparent, detailed reasoning behind flagged issues.
Preference for hybrid approach: AI suggests anomalies and solutions, but users retain decision-making control.

Users trust AI more if provided transparent, detailed reasoning behind flagged issues.
Preference for hybrid approach: AI suggests anomalies and solutions, but users retain decision-making control.

Common Challenges:

Difficulty in visually interpreting large datasets quickly.

Time-consuming and error-prone manual anomaly checks.

Difficulty in visually interpreting large datasets quickly.

Time-consuming and error-prone manual anomaly checks.

Difficulty in visually interpreting large datasets quickly.

Time-consuming and error-prone manual anomaly checks.

Affinity Mapping

Affinity Mapping Summary

Theme 1: Detection Challenges

Users frequently face frustration when working with large datasets in spreadsheets.
Manual scanning is tedious and prone to error, especially when the anomalies are subtle or context-based (e.g., a number that looks normal alone but is off-trend).
Scripts help but require setup and aren’t user-friendly for non-technical users.
Many struggle to get a quick, clear overview of what’s going on in the data.

Users frequently face frustration when working with large datasets in spreadsheets.
Manual scanning is tedious and prone to error, especially when the anomalies are subtle or context-based (e.g., a number that looks normal alone but is off-trend).
Scripts help but require setup and aren’t user-friendly for non-technical users.
Many struggle to get a quick, clear overview of what’s going on in the data.

Users frequently face frustration when working with large datasets in spreadsheets.
Manual scanning is tedious and prone to error, especially when the anomalies are subtle or context-based (e.g., a number that looks normal alone but is off-trend).
Scripts help but require setup and aren’t user-friendly for non-technical users.
Many struggle to get a quick, clear overview of what’s going on in the data.

Theme 2: Expectations from AI Tools

There’s strong interest in an AI solution that goes beyond just flagging cells.
Users want clear, understandable explanations of why something is flagged, ideally with a suggested next step.
Visual indicators like highlights, icons, and summary dashboards are preferred.
Features that save time, reduce cognitive load, and surface trends clearly are highly desired.

There’s strong interest in an AI solution that goes beyond just flagging cells.
Users want clear, understandable explanations of why something is flagged, ideally with a suggested next step.
Visual indicators like highlights, icons, and summary dashboards are preferred.
Features that save time, reduce cognitive load, and surface trends clearly are highly desired.

There’s strong interest in an AI solution that goes beyond just flagging cells.
Users want clear, understandable explanations of why something is flagged, ideally with a suggested next step.
Visual indicators like highlights, icons, and summary dashboards are preferred.
Features that save time, reduce cognitive load, and surface trends clearly are highly desired.

Theme 3: Trust & Control

Trust in AI depends heavily on transparency and user agency.
Participants prefer the AI to make suggestions, not automatic decisions, especially when working with sensitive or high-stakes data.
The ability to review, approve, or undo AI recommendations is key.
Users also want visibility into what the AI is doing and why.

Trust in AI depends heavily on transparency and user agency.
Participants prefer the AI to make suggestions, not automatic decisions, especially when working with sensitive or high-stakes data.
The ability to review, approve, or undo AI recommendations is key.
Users also want visibility into what the AI is doing and why.

Trust in AI depends heavily on transparency and user agency.
Participants prefer the AI to make suggestions, not automatic decisions, especially when working with sensitive or high-stakes data.
The ability to review, approve, or undo AI recommendations is key.
Users also want visibility into what the AI is doing and why.

Theme 4: Customization & Integration

Participants want the tool to work within familiar environments like Excel, Google Sheets, or Power BI.
Many expect to customize anomaly thresholds, define rules, or toggle between beginner and advanced modes.
Some want integration with Python or formula-based logic, while others want pre-built templates to reduce setup time.

Participants want the tool to work within familiar environments like Excel, Google Sheets, or Power BI.
Many expect to customize anomaly thresholds, define rules, or toggle between beginner and advanced modes.
Some want integration with Python or formula-based logic, while others want pre-built templates to reduce setup time.

Participants want the tool to work within familiar environments like Excel, Google Sheets, or Power BI.
Many expect to customize anomaly thresholds, define rules, or toggle between beginner and advanced modes.
Some want integration with Python or formula-based logic, while others want pre-built templates to reduce setup time.

User Personas

User personas

User Journey Map

User journey map

Ideation - Solutions

User Flows

Ideation Sketches

User Flows

Ideation Sketches

User Flows

Ideation Sketches

High Fidelity Prototype

What We Tried To Solve With Our Design

Detection Challenges:

Users rely on manual checks, which are slow and often miss subtle errors like duplicates or unusual drops in values.

Users rely on manual checks, which are slow and often miss subtle errors like duplicates or unusual drops in values.

Users rely on manual checks, which are slow and often miss subtle errors like duplicates or unusual drops in values.

Expectations from AI Tools:

Users want helpful suggestions, not full automation. They expect clear explanations and visual cues without losing control.

Users want helpful suggestions, not full automation. They expect clear explanations and visual cues without losing control.

Users want helpful suggestions, not full automation. They expect clear explanations and visual cues without losing control.

Desired AI Capabilities:

AI should flag issues, explain them simply, and suggest actions—ideally through in-sheet highlights and plain-language insights.

AI should flag issues, explain them simply, and suggest actions—ideally through in-sheet highlights and plain-language insights.

AI should flag issues, explain them simply, and suggest actions—ideally through in-sheet highlights and plain-language insights.