Financial Proposal Correlation Analysis
Enhancing proposal tools through data-driven insights
Project Duration
Q2 2024 - Q3 2024
My Role
As Lead UX Researcher, I designed and executed the research plan to analyze financial proposals. I applied machine learning (hierarchical clustering, correlation analysis) and quantitative methods to uncover usage patterns, template co-occurrence, and categorization issues. I facilitated stakeholder workshops and translated findings into actionable recommendations that improved categorization accuracy and template reuse.
Team
Project Overview
This quantitative UX research analyzed 6,793 financial proposals (created July 2022 - July 2024) to optimize template effectiveness and streamline creation. Leveraging ML techniques like hierarchical clustering and correlation analysis, we identified that 30.7% of proposals were miscategorized ('Other') and 437 were named 'test'. Key findings included Q1 peak usage (1,193 proposals in Q1 2023) and high update rates (90% of proposals). The insights led to a 43% improvement in categorization accuracy, a 51% increase in template reuse, a 26% reduction in search time, and a 17% faster average proposal creation time.
Methods
Tools
The Challenge: From Proposal Volume to Proposal Value
The existing financial proposal system, despite high usage (6,793 proposals created), suffered from significant inefficiencies. Advisors lacked clear guidance on template selection, leading to inconsistent proposal quality and suboptimal use of the platform's capabilities.
Template Underutilization & Misuse
Were advisors using the *right* templates effectively?
Template Underutilization & Misuse
Advisors often struggled to find or select the most appropriate templates, leading to many proposals being poorly categorized (30.7% as 'Other') or inconsistently named (437 as 'test').
Lack of Data-Driven Insights
How could data improve proposal strategy?
Lack of Data-Driven Insights
No systematic understanding of proposal effectiveness, template co-usage, or usage pattern variations hindered data-driven optimization and strategic enhancements.
Inefficient Creation & Discovery
Why was proposal creation and retrieval cumbersome?
Inefficient Creation & Discovery
Poor categorization and naming made finding existing proposals or efficiently creating new ones difficult, causing wasted time and duplicated advisor effort.
Impact of Inefficiencies
- •Significant advisor time wasted in proposal creation and searching.
- •Inconsistent quality and branding in critical client-facing documents.
- •Missed opportunities to leverage successful proposal strategies firm-wide.
- •Difficulty in planning resources for peak demand periods (e.g., Q1 usage spikes).
This Research Aimed To:
- •Quantify template usage patterns and identify co-occurrence with ML.
- •Apply clustering to discover natural groupings for better categorization.
- •Provide actionable insights for enhancing template design and discovery.
- •Reduce inefficiencies in the overall proposal creation workflow.
The Path Forward: Unlocking Proposal Intelligence
These challenges highlighted a clear need to move beyond anecdotal feedback. A deep, data-driven analysis using quantitative methods and machine learning was essential to understand actual usage, identify inefficiencies, and provide a foundation for a more intelligent and effective proposal system.
Key Research Questions
To transform the proposal system from a mere document generator into a strategic tool, we centered our quantitative UX research around understanding actual usage, identifying inefficiencies, and pinpointing opportunities for ML-driven enhancements. Our inquiry focused on how advisors create, manage, and leverage proposals in their daily workflows.
Understanding Usage & Temporal Dynamics
To map out when, how often, and by whom proposals are created and modified, identifying peak periods and update behaviors.
- Q1:
What are the primary usage patterns of proposals across quarters and months, and are there discernible seasonal trends?
- Q2:
How frequently do advisors update existing proposals, and what is the typical lifecycle or modification frequency over time?
- Q3:
What are the dominant proposal categories by volume, and how does usage vary across different company segments or advisor types?
Optimizing Template Effectiveness & Interrelations
To identify which templates are most utilized, how they relate to each other, and how to improve their discovery and application.
- Q1:
Which specific proposal templates are most frequently used, and are there underutilized templates with high potential value?
- Q2:
What correlations exist between different proposal template types, suggesting common co-usage or bundling opportunities?
- Q3:
How can template categorization and naming conventions be improved to enhance discoverability and reduce misclassification (e.g., 'Other', 'test')?
From Inquiry to Insight: Paving the Way for ML-Driven Solutions
These research questions were pivotal in dissecting the complex proposal ecosystem. The answers formed the bedrock for applying machine learning techniques—like hierarchical clustering and correlation analysis—to not only understand current behaviors but also to architect a more intelligent, efficient, and data-driven proposal generation and management system. The goal was to transform raw usage data into actionable strategies for platform enhancement.
Research Methods
To unlock insights from 6,793 financial proposals and optimize template effectiveness, we employed these advanced analytical techniques:
Hierarchical Clustering & Dendrograms
Revealing Template Structures
Applied agglomerative clustering to visualize proposal template relationships and identify natural groupings for consolidation using dendrograms.
What this means: By mapping relationships, we found natural clusters of templates (e.g., 'IRA' and 'IRA, Proposal'), which guided consolidation for a 43% improvement in categorization accuracy.
Hierarchical Template Relationships
Correlation Matrix Analysis
Uncovering Template Co-Usage
Calculated and visualized correlation coefficients between 19 proposal template types to uncover co-usage patterns and inform template design.
What this means: This identified templates frequently used together (e.g., 'Investment Strategy' often paired with 'Performance Review'), leading to a 51% increase in effective template reuse through better suggestions.
Template Co-usage Matrix
Predictive Modeling
Forecasting Proposal Demand
Developed predictive models based on historical data (6,793 proposals) to forecast future proposal usage trends by category and company.
What this means: Models forecasted peak proposal demand in Q1 (used for resource planning) and identified that 90% of proposals are updated, highlighting the need for efficient editing tools.
Usage Forecast by Quarter
Quantitative Dashboard Analysis
Tracking KPIs & User Segments
Leveraged dashboard analytics on 6,793 proposals and 6,102 updates to track KPIs, user segments, and identify key trends (e.g., Q1 peaks).
What this means: Dashboards showed 30.7% of proposals were miscategorized as 'Other' and 437 named 'test', directly leading to UI changes that reduced search time by 26%.
Key Dashboard Metrics
Foundation for Process: These analytical methods (Clustering, Correlation Analysis, Predictive Modeling, Dashboard Analysis) were crucial for our research. They enabled us to identify template relationships, understand co-usage patterns, forecast trends, and derive actionable insights, directly informing the optimization of the proposal generation system.
Research Process
Our systematic approach to analyzing proposal system usage and informing engineering decisions
Usage Analysis
Data Extraction
Power BI Analysis
Advanced Analytics
Executive Delivery
Usage Analysis
Collected quantitative and qualitative data on how financial advisors were using the proposals system and identified key usage patterns.
Data Extraction
Navigated complex database structures to locate and extract the necessary proposal usage data for comprehensive analysis.
Power BI Analysis
Imported data into Power BI for cleaning, transformation, and creation of foundational statistical analyses and visualizations.
Advanced Analytics
Applied regression analysis, machine learning, and LLMs to uncover deeper patterns in how advisors used proposal features together.
Executive Delivery
Presented findings to VP of Product with clear recommendations about engineering resource allocation for the proposals platform.
Key Findings
Our machine learning analysis of 6,793 financial proposals revealed these critical insights about advisor proposal creation and usage patterns.
Quarterly Usage Patterns
Analysis showed consistent Q1 peaks in proposal creation (e.g., 1,193 in Q1 2023), indicating predictable seasonal demand for proposal tools.
Category Fragmentation
Hierarchical clustering revealed that 30.7% of proposals were in 'Other'. Dendrograms showed 4-5 natural template families suitable for consolidation.
Naming Inconsistencies
Text analysis found 437 proposals named 'test', highlighting poor naming practices. Consistent naming correlated with template co-usage in the correlation matrix.
Template Co-Usage Patterns
Correlation matrix analysis revealed strong positive correlations (dark purple areas) between specific template types, indicating common advisor workflows.
Key Opportunity
By addressing category fragmentation (consolidating based on ML) and implementing AI-assisted naming, we can improve proposal searchability by over 26% and increase template reuse by 51%, significantly boosting advisor efficiency.
Impact & Outcomes
How our financial proposal analysis transformed document creation workflows and delivered measurable efficiency improvements.
Research Insights
Identified 30.7% proposals in 'Other' category
Found 437 proposals named 'test', hindering search
Revealed template co-usage patterns via correlations
Implementation
Consolidated categories using ML cluster analysis
Implemented AI-assisted naming suggestions
Developed guided template selection system
User Impact
43% improvement in categorization accuracy
26% reduction in proposal search time
51% increase in template reuse
Efficiency Gains
Reduction in average proposal creation time due to better template discovery and reuse (51% increase).
Improved Analytics
Improvement in categorization accuracy provided cleaner data for more reliable business reporting and trend analysis.
Standardization
Reduction in time spent searching due to standardized naming conventions, facilitating better knowledge sharing.
Proposal Workflow Transformation
Before Implementation
Fragmented categories, large 'Other' group
Inconsistent naming ('test', duplicates)
Limited template reuse, high search times
After Implementation
Consolidated, ML-derived categories (43% better)
AI-assisted naming & standardization
51% increased template reuse, 26% faster search
Visuals
Proposal Data Overview Dashboard
Shows proposal counts by quarter (6,793 total) and updates (6,102) with company distribution.
Reveals Q1 trends with highest proposal creation (1,193 in Q1 2023).
Displays top company concentration with a single company accounting for 659 proposals.
