Background

Transforming 70,000+ HR Inquiries Into Actionable Insights With AI

Designed and led a zero-to-one AI-driven HR insights platform, unifying five disconnected channels supporting 70,000+ monthly inquiries.

Role
UX/Product Owner
Impact
5 Second Multichannel Analytics
Timeline
3 Years
Industry
Enterprise HR Services
Product Strategy Taxonomy Design Data Visualization Stakeholder Management AI/ML Collaboration User Research System Architecture Zero to One
Industry Validation

"...force multiplier for staff charged with improving the user experience." — Jakob Nielsen

Featured by Jakob Nielsen on UX Tigers for transforming unstructured enterprise data into actionable decision signals with AI. Published October 2023.

Read the case study →
MY ROLE & CONTRIBUTIONS
Product Strategy & Leadership
  • UX Lead & Product Owner for zero-to-one strategy, research, taxonomy, design
  • Pitched to 3 senior stakeholders → 3-month beta at 30% capacity
  • 5-slide deck + UI examples → approval in 2 weeks
  • 80% accuracy target and delivery approach for value regardless of outcome
Data, Taxonomy & AI Enablement
  • Built complete taxonomy for user-needs and classified 15,000+ excerpts → business process codes across 5 HR channels
  • Designed preprocessing rules to filter noise, improve model accuracy
  • Partnered with Data Science lead; performance feedback and taxonomy, they owned model architecture
Design, Visualization & Insight Delivery
  • All dashboard views and data visualizations for analytics product
  • Cross-channel volume charts → needs across channels at once
  • Trend analysis for pre/post process impact validation
  • Co-occurrence heatmaps + weekly reports (accuracy, training, insights)
  • UX Lead & Product Owner for zero-to-one strategy, research, taxonomy, design
  • Pitched to 3 senior stakeholders → 3-month beta at 30% capacity
  • 5-slide deck + UI examples → approval in 2 weeks
  • 80% accuracy target and delivery approach for value regardless of outcome
Product Strategy & Leadership
  • Led zero-to-one product strategy, research, and taxonomy as UX Lead & Primary Product Owner
  • Pitched concept to 3 senior stakeholders, securing buy-in for 3-month beta at 30% capacity
  • Created 5-slide pitch deck with UI examples and addressed resourcing concerns to secure approval in 2 weeks
  • Defined success criteria (80% accuracy target) and delivery approach to ensure value regardless of beta outcome
Data, Taxonomy & AI Enablement
  • Built complete taxonomy for user-needs classification across 5 disconnected HR channels and classified 15,000+ excerpts to actionable business process codes
  • Enforced source-agnostic normalization and preprocessing rules to filter non-informative text and improve model accuracy
  • Partnered with Data Science lead, providing performance feedback and taxonomy while they owned model architecture
Design, Visualization & Insight Delivery
  • Designed all dashboard views and data visualizations for analytics product
  • Created cross-channel volume charts enabling teams to view needs across channels simultaneously
  • Built trend analysis views to compare pre/post process changes for impact validation
  • Designed co-occurrence heatmaps and delivered weekly progress reports showing accuracy, training progress, and insights
  • Led zero-to-one product strategy, research, and taxonomy as UX Lead & Primary Product Owner
  • Pitched concept to 3 senior stakeholders, securing buy-in for 3-month beta at 30% capacity
  • Created 5-slide pitch deck with UI examples and addressed resourcing concerns to secure approval in 2 weeks
  • Defined success criteria (80% accuracy target) and delivery approach to ensure value regardless of beta outcome
The Challenge

70,000 Monthly Inquiries.
No Unified View.

HR Services supported 172,000+ employees globally, fielding questions across five disconnected channels.

Each channel produced data in a different format. Leadership could not reliably identify what employees needed most or where improvements would have the greatest impact.

Our agents have been mentioning an uptick of manager questions related to employee changes. But we had no way to quantify it across channels.
Senior Process Manager Change Management

Impact of the Problem

5
Siloed Channels No unified view
70K
Monthly Inquiries Manual analysis impossible
0%
Accuracy Baseline Cannot measure improvement
The Approach

Pitch To Approval In Two Weeks

My pitch deck with value proposition and example UI mockups secured a 3-month beta at 30% capacity from myself and a data engineer partner.

The key was integrating research into an upcoming initiative so the beta would produce actionable learning regardless of outcome.

Beta Targets Achieved

5 Channels Unified
Chat (Bot), Chat (Agent), Form, Phone, Survey
80%
Accuracy Achieved
1,500+ Requests Analyzed
Taxonomy added for top 10 "Change Management" needs
80% classification accuracy

Validated Model in 3 Months

I led product strategy, user research, stakeholder workshops, and taxonomy development, partnering with a Data Science Lead who owned model architecture and technical decisions.

01
The Pitch
5-slide deck + mock UI → buy-in for a 3‑month beta at 30% capacity
02
Compliance Review
Data handling, privacy, and security requirements across all five channels
03
Pre‑Processing Strategy
Filtered non-informative text so the model learned needs—not noise
04
Taxonomy Workshops
Weekly sessions with process owners to classify language variations
05
Initial UI Design
First stakeholder-facing UI, designed for trust
06
Continuous Testing
Ongoing validation and iteration to maintain accuracy
07
80% Accuracy Beta
Quantified 10 top process specific needs across chat and surveys at 80% accuracy
08
Reflection Report
Outcomes and insights to influence scaling approach
Critical Foundation

Pre-Processing Strategy

Before training, I identified that a high percentage of text data was noise (greetings, acknowledgments, procedural language), which was increasing our model’s scope of work. I classified 5,000+ excerpts to create a “not applicable” taxonomy that filtered out low value sentences before ingestion.

50% → 80%+ Accuracy Gains
Testing results reached 80%+ accuracy more quickly and more predictably
Hello, I need assistance with
transferring my worker to
another manager. Can you
please guide me through the
process? It's urgent and I want to
ensure everything is done
correctly. Hello, I need assistance with transferring my worker to another manager. Can you please guide me through the process? It's urgent and I want to ensure everything is done correctly.
Ongoing Collaboration

Taxonomy Workshops

I designed weekly workshops with process stakeholders to build and refine taxonomy using domain expertise, aligning on language variations and how employees describe their needs. These sessions doubled as discovery, strengthening shared understanding and trust.

Trust + Credibility
High-participation sessions drove trust, discovery, and alignment.
Solutions

From Data Chaos to Actionable Clarity

Unified data across all five channels gave leadership a single view of what employees were actually struggling with by consolidating 70,000+ monthly inquiries into one product.

Exposed patterns previously hidden in siloed systems.
Enabled easy measurement of process improvement impact.
Distinguished preventable vs. expected inquiries to target cost reduction.
Success Metric
5 Seconds
Time to generate insights that previously took weeks of manual analysis
Cross-Channel Inquiry Volume
Expected
Preventable
Need Tier Survey Chat (Bot) Chat (Agent) Phone TOTAL
A

Measuring Impact

Second only to "show me the highest-volume problem areas" was: "Show me if our changes actually worked."

By focusing on pain-point volume, the tool enabled teams to benchmark, detect volume shifts, and quantify impact across channels.

Trend Analysis Before/After Measurement Impact Validation
Recruiting and Onboarding
4 Needs selected
Measured Impact
60% Reduction
In high-volume Recruiting & Onboarding pain points over one year through targeted identification and measurement.
B

Co-Occurring Needs Discovery

Users often ask multiple things in a single inquiry. The heatmap surfaced relationships that process teams weren't aware of, like how "background check" questions almost always came with "start date" questions.

Agent Training Gaps: Identified areas where agent answers didn't align with what users were actually asking.

Faster QA: Cross comparison of clusters made incorrect matches obvious.

User Journey Insights: Discovered associated needs that appeared together, revealing how employees navigate HR processes.

Heatmaps Dependency Discovery Quality Assurance
Recruiting and Onboarding
Chatbot Accuracy Improvement
+80% Accuracy
Identified 1,500+ incorrect chatbot responses, enabling targeted taxonomy improvements.
C

Actionable Intelligence for a Historic Company Separation

I transformed classified inquiry data into executive-ready reports that guided decisions throughout a historic three-way separation. Leaders could see problem areas by company, region, and team. This enabled targeted support and process changes management before and after the split.

Separation Intelligence Leadership Reports Company / Region / Team
Report views: by company, region, and team; separation planning detail
OUTCOMES: SUCCESS METRICS
5 Second

Multichannel Analytics

Weeks of manual analysis reduced to seconds. One product unifying 70,000+ monthly inquiries across five channels.

80%
Chatbot Accuracy Improvement

Identified 1,500+ incorrect responses via taxonomy alignment. Retraining improved chatbot accuracy 80%, reducing agent escalations.

60%
R&O Pain Point Reduction

Led a workshop creating a taxonomy measuring top 15 Recruiting and Onboarding requests at 80% accuracy. Targeted inquiry volume dropped 60% within one year.

70K+
Monthly Inquiries Unified

Converted classified inquiry data into executive insight reports by company, region, and team. Enabled proactive support before and after enterprise split.

2022 → 2024
Beta to Enterprise Scale

Feb–Apr 2022 Beta (10 needs). Sept 2022 production, ongoing enhancements through Dec 2024, 50+ needs with 80%+ accuracy.

Conclusion

Reflecting on the Journey

Below I've captured what worked well, what I'd do differently, and what I'm taking forward: a retrospective on building a zero-to-one product from pitch to production.

What Worked Well

  • Pitch-first approach: mockups secured buy-in from senior stakeholders before building
  • 80% accuracy target gave leadership a clear, trustworthy benchmark
  • Weekly taxonomy workshops built domain expertise and trust while classifying 15,000+ excerpts
  • Source-agnostic architecture made adding new channels seamless
  • Before/after measurement let teams benchmark, detect volume shifts, and quantify impact

What I'd Do Differently

Invest in preprocessing discovery earlier

We could have added roughly 50% more needs during beta if we'd surfaced preprocessing rules and noise patterns sooner. That would have unblocked taxonomy and classification earlier.

Build automated conflict detection earlier

Manual QA cycles slowed iteration. Catching conflicts programmatically would have accelerated feedback loops.

Create a self-service contribution path

Process owners had valuable edge cases but no easy way to submit them. A contribution portal would have improved coverage faster.

Core Learning

The future of enterprise analytics is hybrid. Custom taxonomy helped us prove value with precision and trust; exploring LLMs earlier would have accelerated scaling. The optimal path: leverage LLMs for faster insight discovery, then apply human-in-the-loop taxonomy where specificity and accountability matter most.