Portfolio

From Fragmented Knowledge to Orchestrated Intelligence:

A Multi-Agent Transformation

CLIENT

Eli Lilly

Pharmaceutical

Strategy

RESPONSIBILITIES

Strategy

Conversation Design

Research

Orchestration

ROLE

AI Integration & Strategy Lead

SOFTWARE

TEAM

Conversation Designer

Product Owner

Data Scientist

Security Stakeholder

The System Problem

After multiple HR platform rollouts and IT system migrations, Lilly’s knowledge environment had fractured

12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information

12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information

12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information

12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information
12+ repositorie
Knowledge scattered across 12+ repositories (SharePoint, intranet, PDFs, KBsr
500+ duplicate tickets
500+ duplicate tickets per month, often triggered by conflicting answer
weeks to propagate updates
Policy updates taking weeks to propagate across channel
no single source of truth
No single source of truth, leading to employee distrusu
$2–3M lost productivity
Estimated productivity loss: ~$2–3M annually, based on 15,000 employees losing 30 minutes weekly searching for policy information

Trigger: During benefits enrollment season, confusion peaked as employees received conflicting policy guidance. This created compliance exposure and employee frustration—escalating the issue to the Csuite

Orchestrated Intelligence

Why Now?

The urgency was undeniable:

Hybrid work expansion made self-service mission-critical

Compliance audits revealed risks from inconsistent policy disseminatio

Retention pressures raised the cost of bad employee experience

Executives recognized fragmented knowledge as both a strategic liability and a compliance threat

The mandate: solve knowledge fragmentation or risk penalties, attrition, and loss of trust.

The Orchestration Strategy

Previous fixes had failed: unified search engines, centralizing SharePoint content, or deploying a single chatbot. They couldn’t resolve conflicts or enforce compliance in real time.

Our orchestration approach:

Agent Registry: Core quartet Listener → Retriever → Policy Sentinel → Content Coach — handled 80% of queries; specialized agents (Evaluator, Audit Logger, Escalation) covered edge case

Governance Layer: conflict resolution hierarchy → Compliance > Brand > Brevit

Observability: audit-ready lineage of every answer (input → prompt → source → approvalr

Feedback Loops: conversation analytics feeding prompt and policy update

Rollout Plan: shadow mode → assist mode → controlled automation

Vendor Role: Partnered with OneReach (orchestration platform vendor) while I owned the system strategy, governance design, and enterprise adoption roadmap.

The Human Impact

This wasn’t just technical; it reshaped experience

Employee search time: PTO lookup dropped from 15 minutes → 2 minute

Agent confidence

For the first time, I feel confident I'm saying the right thing about FMLA without checking three different docs. Service Desk Agenu

Leadership trust: Audit-ready lineage shortened approval cycles and reduced compliance anxiety

Vendor Role: Partnered with OneReach (orchestration platform vendor) while I owned the system strategy, governance design, and enterprise adoption roadmap.

The Transformation Results

Scale: Served 15,000+ employees, handling ~2,000 queries daily across 300+ policy documents

System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks

System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks

System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks
Process-level
Review cycles cut from 5–7 days → 3–4 days, saving ~200 hours/month
Duplicate tickets reduced, freeing service desk for higher-value issues
Human-level
Containment up, AHT down in pilot categories
20% increase in employee satisfaction with HR service
25% fewer compliance-related content errors post-implementation
System-level
Achieved one source of truth across HR and IT policy
Policy updates propagated in days, not weeks

Business outcome: ~15% reduction in HR service costs

Lessons from Friction

Transformation is never clean:

Resistance from content owners:

Fear of losing control. Solved by reframing them as orchestration designers instead of reviewers

Rollout misstep:

Rushed from shadow to assist mode too quickly, eroding trust in one pilot. Slowing the transition restored adoption·

Technical constraint:

Legacy HRIS lacked APIs. Built intelligent extraction layers to monitor changes and auto-update the knowledge graph.

Takeaway: Governance isn t a gate—it s a compass. Success depended as much on cultural alignment as technical architecture.

Strategic Choices

This was AI strategy, not just information architecture:

Modeling: Selected smaller, specialized models over giant LLMs to minimize hallucinations, improve latency, and win compliance approval

Competitive evaluation: Assessed ServiceNow and others, but none could handle real-time policy conflict resolution at scalò

Team leadership: Directed a cross-functional team of 8, aligned C-suite stakeholders, and drove adoption with OneReach as platform partner

Traditional IA organizes static content. Our AI strategy orchestrated dynamic, conflicting knowledge sources with real-time governance and self-improving feedback loops.