Navi: Designing BCG's AI Knowledge Assistant
How strategic UX leadership turned a fragmented search-and-chat experience into a clearer, more trusted AI knowledge workflow.
Headline Results
Executive Summary
Product transformation at a glance.
Problem
Inherited a powerful but underperforming knowledge assistant with significant adoption and satisfaction challenges.
- Navi usage was low relative to KN's user base; early features felt limited or unintuitive for everyday work
- Auto-routing users to Navi from KN search disrupted workflows and caused a 9% ESAT decline in Q1 2025
- Fragmented UX with inconsistent patterns across chat, search, and agent selection
- No shared UX strategy, research cadence, or measurement framework for the Knowledge portfolio
- Content gaps, relevance issues, and performance problems eroded user trust
What I Did
Built a UX operating model to diagnose root causes, validate solutions, and drive measurable improvements.
- Established continuous UX research cadence with dedicated UX researchers (10+ studies, 50+ participants for Navi-specific work)
- Designed and advocated for key UX interventions: Chat default removal, homepage redesign, agent discoverability, Deep Research workflows
- Drove alignment to GenAI design system standards, successfully persuading the PO to replatform Navi
- Managed product tradeoffs using data (e.g., quantifying the adoption vs. satisfaction tension from auto-routing)
What We Shipped
Shipped targeted UX improvements, new capabilities, and a strategic product direction.
- Removed the Navi default toggle from KN search to reduce workflow disruption (addressing 8-11% of dissatisfaction)
- Designed and shipped usability quick wins: improved agent discoverability, simplified agent selection, clearer search-to-chat navigation
- Designed the Internal Deep Research capability enabling deeper multi-source synthesis
- Designed the Navi homepage/landing page redesign with intent-driven agent selection
- Integrated BCG Internal content as a new data source (addressing 21% of negative feedback)
Impact (as of Nov '25)
Delivered measurable product, user, and business outcomes.
- Navi ESAT recovered from 39% low (Dec '24) to 59% (Jun '25), a +20pp swing
- Adoption grew 1.8× to ~18k chats/week with 9k monthly active users
- Time savings: ~0.5 hrs/wk (CT) — up from 20 min baseline
- Retired legacy search infrastructure saving $400K annually
- 100% of Deep Research beta users reported it adds value
- Active user engagement: A/C 70%, PPPL 70%, MDP 55%
Context & Problem
BCG's knowledge is one of the firm's core assets, and how effectively employees can access it is vital to BCG's competitive advantage. Navi had meaningful capabilities, but the experience was fragmented. Satisfaction was declining, and there was no shared UX strategy to better understand its users or improve the experience.
What looked like an adoption problem was actually an experience strategy problem.
The team had not yet clearly defined which user behaviors and use cases were best served by search, chat, or hybrid workflows, how agents should be used, or how to build trust with its outputs.
When I joined as Product Design Lead, my role was to bring structure to that complexity, establish an evidence-based UX practice, and create a design strategy that could improve satisfaction, grow adoption, and earn trust without waiting for every content gap to be fixed.
My Goals, Role & Responsibilities
My mandate as Product Design Lead was to bring structure to a fragmented experience and establish a repeatable UX operating model for the Knowledge portfolio.
- 1Define the platform UX strategy
Clarify how KN search, Navi chat, and agent-based experiences should work together, and prioritize the UX interventions that would improve adoption, satisfaction, trust, and navigability.
- 2Map user behavior by cohort and task
Understand how CT, BST, and MDP users search, synthesize, validate, and apply knowledge so the right mode and workflow could support the right use case.
- 3Build a durable UX operating model
Establish KPI alignment, research cadence, and cross-functional decision-making so UX activities were integrated into roadmap planning, product tradeoffs, and outcome measurement.
My Role & Responsibilities
Decision Rights & Operating Model
I owned the UX strategy and design direction for Navi and its key KN touchpoints within the Knowledge portfolio. Product owned business prioritization. Engineering owned technical implementation. Decisions were made jointly, with UX providing evidence-based recommendations grounded in research and analytics.
Responsibility Pillars
UX Strategy & Vision
Owned the cross-product UX strategy for Navi and KN touchpoints, defining how search, chat, synthesis, and agent experiences should work together and setting the experience direction for the platform.
Product Design Leadership
Led end-to-end product design across the highest-impact problem areas: homepage redesign, agent discoverability, Deep Research workflows, search-to-chat handoffs, and new data-source experiences.
Research Leadership
Established an ongoing research partnership with dedicated UX researchers across 10+ Navi-specific studies. Provided design assets, prototypes, and concept stimuli to diagnose root causes and validate direction.
Experience Architecture
Designed the experience architecture for agent routing, hidden agent selection, multi-source responses, and hybrid search + chat workflows. Ensured coherence across KN, Navi, and Deckster touchpoints.
Design System & Ops
Drove Navi's adoption of the unified GenAI design system to improve consistency and reduce maintenance overhead. Aligned interaction patterns with other BCG GenAI tools to reduce cognitive switching costs for users moving across products.
Cross-Functional Leadership
Served as strategic design partner to the PO, helping align product, engineering, and UX around portfolio priorities, sequencing, and tradeoffs. Managed design decisions with data, including the adoption vs. satisfaction tension from auto-routing.
Research: Evidence → Decisions
We used a sequence of research methods across analytics, in-product feedback, interviews, surveys, and concept testing to clarify which problems were UX issues, which were content or platform issues, and where design could create the most leverage.
Key Takeaway
Targeted research made a fragmented UX legible enough to untangle. Insights showed that Navi's challenges were not just about adoption. The experience was fragmented across three dimensions: mode clarity, interaction architecture, and trust.
Research Scale
10+
Navi-specific research studies (within broader K&D portfolio of 20+ studies)
50+
Research participants across Navi studies spanning CT, BST, and MDP cohorts
80–100
In-app feedback submissions per cycle via Usabilla + quarterly Digital Pulse survey data
Three Research Threads That Shaped the Strategy
Research did not reveal one root cause. It surfaced three distinct but related sources of fragmentation that each required a different design response.
1Mode clarity: search and chat served different use cases
Design implication: Preserved both interaction paths; defined clearer search-to-chat handoffs and clarified which mode should support which use cases.
2Interaction architecture: hidden agents made capabilities hard to navigate
Design implication: Redesigned agent discovery, simplified selection, and reduced exposed complexity through clearer pathways and orchestration.
3Trust: UX issues were compounded by content, relevance, and performance gaps
Design implication: Separated UX problems from content/platform issues and prioritized trust-building interventions across content access, relevance, and performance.
User Personas
Synthesized across seven UXR studies, these personas capture the recurring jobs-to-be-done and behavioral patterns that shaped how search, chat, and hybrid workflows needed to work together.
The Ramping Generalist
Get smart fast
Associate / Consultant users who rely on chat-first exploration to ramp up quickly on a new client situation, topic, or practice area.
They expect a usable first answer with context, relevant materials, and directionally-right synthesis. When Navi falls short, they switch to ChatGPT for brainstorming and return to KN for artifacts.
The Delivery Lead
Create client work
Project Leaders and Principals who move fluidly between search, chat, and browse while building proposals, slides, and client-ready outputs.
They want Navi to behave like a top associate: explain why a recommendation is relevant, connect to next steps, and reduce the 30-40% of time they currently spend validating outputs.
The Targeted Retriever
Find a specific artifact, fact, or person
Senior Vantage specialists, experienced CT users, and cross-role users who arrive with a specific deck, slide, benchmark, policy, expert, or internal fact in mind.
Their behavior is transactional and search-first, even when prompts sound conversational. They have near-zero tolerance for friction, forced mode decisions, weak title matching, or missing BCG Internal answers.
The Research Synthesizer
Deep synthesis and benchmarking
Research Analysts, senior KNA users, Vantage experts, and GenAI power users who run structured, multi-source research workflows.
They treat prompts like code, skim outputs for tables and citations, and rely on Deep Research for high-value synthesis. Their biggest bottleneck is manual verification and weak follow-up depth.
Quantifying the Auto-Routing Tradeoff
Auto-routing users to Navi from KN search increased adoption but tanked satisfaction, making the tradeoff visible in both analytics and user feedback.
ESAT measurement for Navi began in October 2024, so the first point in the chart and table marks the start of available Navi satisfaction tracking. Navi itself launched earlier, in January 2024.
ESAT trend over time
Oct 2024 to Nov 2025
| Period | ESAT | Context |
|---|---|---|
| Oct '24 | 61% | First ESAT captured; pre-toggle baseline in-app satisfaction before Navi default was enabled |
| Nov '24 | 44% | Auto-route to chat enabled (toggle ON); sharp decline after routing KN searches into Navi chat |
| Dec '24 | 39% | Lowest point; users frustrated by forced chat-only paths |
| Jan '25 | 43% | Slight recovery as team begins addressing feedback |
| Feb '25 | 47% | Continued recovery |
| Mar '25 | 48% | Q1 2025 Digital Pulse confirms 9% decline vs. prior quarter |
| Jun '25 | 59% | Q2 readout; +7pp vs. prior quarter after UX interventions and quick wins |
| Nov '25 | 55% | Q4 readout; slight pullback. CT 51%, BST 59%. Content gaps remain the key barrier. |
What We Heard
“Nobody puts anything useful on KN — all the good stuff lives in people's personal SharePoints. This is really a cultural issue vs. an IT issue.”
“I usually search for internal policies but it seems to have a bias for client material.”
Cohort Use Case Model
Research surfaced distinct use case patterns by cohort that informed both design and roadmap decisions.
Senior CT users with deep expertise favor targeted material retrieval
Associates / Consultants lean toward chat for broader topic exploration
Expert lookup, policy queries, quick BCG-internal answers
Complex research starting with search, then deepening via Navi synthesis
Cohort Engagement by Role (Oct '25)
Research Content To Review
I kept the earlier insight framing here so we can compare it against the newer three-thread narrative before deciding what to keep or cut.
Auto-routing to chat disrupted workflows and lowered satisfaction by 9%
Design implication: Removed Navi as default toggle; designed clearer paths between search and chat modes
Users use Search and Chat differently depending on task and cohort
Design implication: Designed a hybrid search+chat model rather than forcing a single modality; preserved both interaction paths
Agent selection caused cognitive load; users didn't know which agent to choose
Design implication: Redesigned homepage for intent-driven agent discovery; proposed auto-routing via orchestration layer
Content gaps (not UX) drive 55% of negative feedback
Design implication: Prioritized BCG Internal integration as data source; shifted product narrative toward content strategy alongside UX
Solution
Navi provides a GenAI-powered knowledge assistant embedded within Knowledge Navigator — supporting search, chat-based synthesis, expert discovery, and deep research across BCG's proprietary and external data sources.
KN Search + Navi Chat Integration
Knowledge Navigator interface showing the search bar with Navi integration — search results with an embedded Navi AI Summary panel providing a chat-synthesized overview alongside traditional document results
Let users search and chat seamlessly within the same experience, preserving both interaction models.
Reduces friction between search and chat modes; organically introduces Navi's synthesis capabilities downstream in the user's journey.
How I Led the Transformation
I used a systematic approach to diagnose, design, validate, and measure — building an evidence-based UX practice that shifted Navi from reactive feature development to proactive, research-led product decisions.
Diagnose & Frame
Understand the root causes of low adoption and satisfaction, map the knowledge discovery journey, and establish a problem frame that guided roadmap priorities.
Activities
- Joined the product team and onboarded to KN/Navi's ecosystem, data sources, and user landscape
- Conducted discovery research including user interviews and survey analysis across CT, BST, and MDP cohorts
- Mapped use case differences between search and chat modalities across cohorts
- Analyzed in-app feedback (Usabilla, 80-100 submissions/cycle) and Digital Pulse survey data
Key Outcomes
- Confirmed that search and chat serve different mental models and use cases — forcing users into one mode lowers satisfaction
- Identified core pain points: content relevance (55% of negative feedback), vague prompting (29%), performance (14-20%), search-vs-chat confusion (8-11%)
- Established a feedback categorization framework that linked dissatisfaction drivers to actionable product priorities
- Surfaced cohort-specific patterns: Senior CT users favor search for targeted retrieval; A/Cs lean toward chat for exploration
Dissatisfaction driver analysis
Categorized dissatisfaction drivers that directly informed roadmap prioritization — content gaps, not UX, were the primary barrier, but UX interventions could address ~35% of negative feedback.
Horizontal stacked bar showing the breakdown of negative feedback: Relevancy 18%, BCG Internal 21%, Performance 14%, Expert Search 11%, Search vs Chat 8%, Usability 8%, Content 13%, Deep Research/Capabilities 4%, Unknown 3% (N=72 Usabilla; supplemented by Digital Pulse)
Design & Validate UX Interventions
Rapidly design and validate UX solutions to address the ESAT decline, starting with high-impact quick wins and evolving toward a strategic homepage redesign.
Activities
- Conducted rapid 3-week discovery and concept testing with 10 participants and 50 survey respondents
- Developed UI/UX concepts for: KN homepage experience, PA-specific agent marketplace, search+chat hybrid, Navi toggle treatment
- Tested two agent layout concepts (card marketplace vs. flat list) and validated preferences by cohort
- Designed quick win UX improvements: chat default removal, agent discoverability, search-to-chat navigation clarity, visual language alignment with other BCG GenAI tools
Key Outcomes
- Validated that Navi should not be the default homepage experience without further refinement
- Confirmed user enthusiasm for Deep Research but confusion about PA-specific agents
- Produced 6 actionable UX recommendations adopted into the product roadmap
- Shipped toggle-off (addressing 8-11% of dissatisfaction) and usability quick wins that contributed to the +7pp ESAT recovery
Six UX Recommendations (adopted into roadmap)
- Remove the Navi toggle from the KN search bar to reduce workflow disruption
- Integrate a Navi AI 'Search Summary' on KN search results page (similar to Google Gemini's approach)
- Provide a clear, intuitive way to switch between KN search and Navi
- Improve agent discoverability by moving away from dropdown selection, aligning with other GenAI product patterns
- Use straightforward visual language (e.g., broom icon for new chat) aligned with other BCG GenAI tools
- Continue exploring alternative UI treatments for Navi on the homepage
Concept testing results
Concept testing revealed that layout preferences correlate with use case: card format supports exploration, flat list suits focused users.
Side-by-side screenshots of Concept 1 (Card Marketplace, preferred by 6/10 participants) and Concept 2 (Flat List, preferred by 4/10 participants) with annotated pros/cons
Design Advanced Capabilities
Design the interaction model and workflows for Navi's advanced capabilities — Deep Research, multi-source synthesis, and PA-specific agents — to enable deeper, more valuable knowledge work.
Activities
- Benchmarked industry patterns for deep research interaction (ChatGPT, Gemini) to align internal differentiators with user expectations from 3rd party tools
- Designed the Deep Research workflow: from prompt input → iterative reasoning loops → structured output (executive summary, BCG perspective, external perspective, comparison) → citation preview and validation
- Designed the multi-source response model enabling Navi to combine answers from KN Materials, Experts, Web, BCG Internal, LSEG, and Transcript Library
- Designed the PA agent marketplace and selection UX, testing both card and flat-list layouts
Key Outcomes
- Deep Research design validated with 7 user interviews: 100% reported DR adds value
- Proposed a hybrid 'search+chat' model for the platform
- Agent orchestration design enables >95% accuracy in auto-routing queries to the right data source
- Deep Research beta launched to 1,000+ power users with positive reception
Deep Research workflow
Designed a tiered capability model from quick answers to deep research, informed by industry benchmarking and validated through user interviews.
Annotated flow diagram showing the three tiers of Navi capability: Question & Answer (quick, <30s) → Reasoning (multi-source synthesis, 3-5 min) → Deep Research (complex, iterative, ~20-25 min). Each tier maps user inputs → system states → outputs → validation steps.
Navigate the Auto-Routing Tradeoff
Quantify and communicate the tradeoff between adoption and satisfaction caused by auto-routing users to Navi, and demonstrate that durable behavior change persisted after the setting was reversed.
Activities
- Combined analytics with qualitative research to surface the adoption vs. satisfaction tradeoff when Navi was auto-enabled
- Tracked ESAT decline from 61% → 39% during the auto-routing period (Oct–Dec '24)
- Analyzed post-reversal data to determine whether the adoption spike was durable
- Recommended and implemented the toggle-off, monitoring recovery
Key Outcomes
- Demonstrated that post-reversal adoption remained above baseline as satisfaction recovered — indicating durable positive behavior change, not just forced usage
- ESAT recovered from 39% (Dec '24) to 59% (Jun '25), a +20pp swing
- Chat volume stabilized at ~25k chats/week post-reversal (vs. ~13k pre-toggle), then naturally leveled to ~17-18k
- This analysis became a reference example of data-driven product decision-making
Adoption vs. satisfaction tradeoff visualization
Data-driven analysis of the auto-routing tradeoff: adoption spiked but satisfaction cratered. Post-reversal, adoption remained above baseline as satisfaction recovered — demonstrating durable behavior change.
Dual-axis chart showing weekly chat volume (bar chart) and ESAT % (line chart) from Oct '24 through Mar '25. Clear visual showing the spike in chats when toggle was ON, the corresponding ESAT drop, and the post-reversal recovery where chats remained elevated above baseline while ESAT climbed back.
4.2 Engagement Trajectory
| Event | Weekly Chats | Notes |
|---|---|---|
| Pre-Sept 2024 release | ~8K/week | Baseline |
| Post Sept 2024 release + campaign | ~30K/week | +292% increase |
| Feb 2025 | 56% monthly active | Peak engagement period |
| June 2025 | 29–54% monthly active by cohort | Declined after default toggle turned off |
| 2026 target | 16K active users/week, 40K chats/week | Set in Dec 2024 ABR |
Continuous Improvement & Design System Alignment
Drive platform alignment to GenAI design standards, establish a measurement loop, and continuously improve based on user signals.
Activities
- Advocated and persuaded the PO to replatform Navi using the GenAI design system, improving cross-product consistency and reducing long-term maintenance burden
- Designed usability improvements: flat navigation, simplified taxonomy, large interaction windows, prompt guidance, 'waiting' indicators, layout optimizations
- Contributed to GenAI design system governance, driving adoption across 15 product teams (peak 139k component insertions in June)
- Ran ongoing ESAT drop investigation research (Q3 2025) with dedicated test plans targeting PPPL cohort satisfaction decline
Key Outcomes
- Navi adopted the GenAI design system (~2,000 component insertions), ensuring pattern consistency across Q, Deckster, Navi, and Proposal Toolkit
- Codified common interactions: agent selection, result display, chat conventions, error-prevention pathways
- Performance improved: page load from 4.2s → 2s; application performance +70%; cost efficiency maintained at $0.04/chat
- Design system adoption reduced cognitive switching costs for users moving across GenAI products
Design system adoption metrics
GenAI design system adoption across the portfolio, with Navi's migration enabling cross-product consistency and accelerating development velocity.
Bar chart showing component insertions by product team over time. Navi at ~2,000 insertions; total portfolio peak at 139k in June.
Outcomes & Impact
Adoption & Engagement
UX Impact
Business Value
What Users Say
“Navi lets me see the slide in the citation — that's great.”
— Knowledge Analyst, Industrial Goods & OEM sectors, Heavy Navi user
“[Deep Research] accelerates research and analysis, especially when combined with KN data, driving faster, higher-confidence insights.”
— Aggregated user feedback, Deep Research beta (7 interviews)
Leadership & Organizational Impact
From Reactive to Evidence-Led
Helped shift the team from reactive feature fixes to proactive, research-backed product decisions. The dissatisfaction driver analysis, ESAT drop investigation, and Deep Research validation all demonstrate this shift.
UX as Strategic Partner
Elevated UX from downstream implementation to upstream strategy. A product leadership partner credited the collaboration with driving "improvements reflected in the next quad survey, as Navi e-SAT improved." Navi's homepage redesign and Deep Research capability were UX-led initiatives.
Cross-Product Design Coherence
Drove Navi's alignment to the GenAI design system, ensuring consistency across Q, Deckster, Navi, and Proposal Toolkit.
Data-Driven Product Tradeoffs
Quantified the adoption vs. satisfaction tension from auto-routing, demonstrating that durable behavior change persisted post-reversal. This analysis established a reference model for evidence-based decision-making across the portfolio.
Key Takeaway
Navi demonstrates how UX leadership can navigate complex product tradeoffs in a GenAI knowledge environment. The transformation required:
- •Diagnosing root causes with research, not assumptions (content gaps ≠ UX gaps)
- •Managing the tension between adoption and satisfaction with data
- •Designing for multiple user mental models (search vs. chat, exploration vs. precision)
- •Aligning platform standards across a multi-product GenAI ecosystem
- •Earning trust as a strategic design partner in a product-engineering-led organization