Case study / Enterprise AI platform & design system

Bay6 AI — From Generic Vendor to Enterprise Intelligence Platform

Led the end-to-end UX transformation, product naming architecture, and enterprise design system for Bay6 AI (formerly Xemplar AI)—turning fragmented bespoke algorithms into a cohesive product suite that drives 40–60% support deflection and 90%+ predictive accuracy.

40–60%

Support query volume deflected through automated self-service workflows

90%+

Production model accuracy across insurance risk and retention scoring

< 90 Days

Deployment time from initial discovery to live enterprise integration

+300%

Increase in qualified enterprise demo requests from C-suite buyers

Platform brand and interface motion overview Full HD overview

Role

Principal Product & Visual Systems Designer

Industry verticals

Insurance, Financial Services, Higher Education

Compliance standards

SOC 2 Type II Certified, FERPA Compliant

Core products

Agent6, Model6, Connect6, Forge6, PolicyBuddy™

01 / The strategic pivot

The "Generic AI Vendor" Trap

As Xemplar AI evolved into a structured enterprise platform, its market perception lagged behind. Enterprise buyers (CIOs, VP Claims, Chief Risk Officers) were fatigued by vague algorithmic claims and prolonged experimental pilots that failed to reach operational production.

Legacy identity

Xemplar AI

Positioned as a custom project-based AI services shop with scattered tools

Rebranded identity

Bay6 AI ("Built to Think")

Unified enterprise intelligence platform grounded in operational fundamentals

The Name Origin: "Basics" Core philosophy

The name Bay6 originates from the word basics, articulating a foundational design principle: enterprise AI cannot solve high-stakes business problems without mastering core operational requirements—contextual relevance, data privacy, existing system integration, and verifiable ROI.

The rebrand addressed structural architecture rather than superficial aesthetics, transitioning clients from lengthy experimentation to reliable production within 90 days.

Legacy: Xemplar AI

Perceived as custom dev agency
Isolated scripts and disconnected tools
Opaque predictive models with drift uncertainty
Disconnected chatbot handoffs dropping context
12–18 month prolonged POC cycles

Transformed: Bay6 AI

Integrated enterprise intelligence platform
Unified "6-System" architecture (Model6, Agent6)
Transparent model explainability and drift monitoring
Context-preserving human-in-the-loop handoff
Predictable sub-90-day production roadmap

Figure 1.1 · Strategic positioning framework: Think, Integrate, and Scale Responsibly Architecture blueprint

02 / Product architecture

The Think6 Cognitive Framework & Product Naming

To earn trust in highly regulated industries like insurance underwriting and student admissions, we introduced the Think6 Framework. Every platform capability is designed through six cognitive dimensions.

Figure 2.1 · The "6-System" product hierarchy: Unifying standalone tools into a cohesive platform Platform ecosystem

Lens 01

Think Knowledge

Ingests dark enterprise data—unstructured PDFs, policy manuals, and claims logs—making institutional knowledge instantly searchable.

Lens 02

Think Signals

Monitors behavioral and operational patterns to detect customer intent, fraud indicators, and churn risks early.

Lens 03

Think Intelligence

Deploys reasoning agents that adapt dynamically to user nuance rather than failing on rigid scripted decision trees.

Lens 04

Think Systems

Integrates cleanly into existing core platforms (Guidewire, Salesforce, Ellucian) via Model Context Protocol (MCP) and REST APIs.

Lens 05

Think Outcomes

Every interface is engineered toward measurable business goals: inquiry deflection, loss ratio reduction, and improved retention.

Lens 06

Think Forward

Continuous feedback loops where models learn from human operator corrections, steadily compounding in accuracy over time.

03 / Visual language & design system

The Enterprise Design System

Balancing technological capability with corporate restraint. We formulated a structured design system with clear typography hierarchy and accessible semantic color tokens.

Figure 3.1 · Bay6 visual identity specification: Wordmark geometry, color palette, and typography system Design system sheet

Emerald

#10B981

Primary action, active resolution, system status

Slate Blue

#3B82F6

Telemetry streams, API integrations, data badges

Amber

#F59E0B

Model drift alerts, human handoffs, warnings

Violet

#8B5CF6

Reasoning gates, multi-stage agent pipelines

Deep Canvas

#0B0F17

Base surface, card layering, high contrast

Figure 3.2 · Material and layout moodboard: Surface hierarchy and typographic clarity Design exploration

Motion study: Modular glyph transitions and surface depth Interaction reel

04 / Interface walkthroughs

Core Product Interfaces & Schematics

Full-resolution screen captures and architectural schematics designed to simplify high-frequency workflows for enterprise operators and technical managers.

Figure 4.1 · Live enterprise platform homepage: Value proposition and workflow positioning Live production capture

Figure 4.2 · Self-service claims and policy intake schematic Automated resolution workflow

Bring Us the Workflow That Slows You Down

Self-service claims intake with real-time verification and zero-drop CSR escalation.

Stage 1: Ingestion

Inbound Customer

Mobile web chat, WhatsApp, or voice IVR interaction.

Stage 2: ProcessingIntent Extraction
Connect6 parses policy number, location, and severity level.

Stage 3: Integration

Policy Verification

Real-time Guidewire API check for coverage details.

Stage 4: Outcome

Resolution or Handoff

64% direct automated resolution; 36% context-aware CSR handoff.

Figure 4.3 · Model6 predictive analytics portal Model-as-a-Service

Figure 4.4 · Connect6 conversational self-service Conversational AI

Figure 4.5 · Agent6 autonomous enterprise agents Agentic workflows

Figure 4.6 · PolicyBuddy insurance AI module Domain solution

Figure 4.7 · Forge6 custom solution architecture: Model Context Protocol (MCP) and enterprise connectors Integration map

Figure 4.8 · Model6 predictive telemetry and drift monitoring console Model version 2.4

Commercial Auto Claim Risk Engine

Prediction accuracy window · 90-day telemetry

94.2% Current Accuracy

May 01 (Baseline: 89.1%) Jun 15 (Retrained: 92.4%) Jul 30 (Current: 94.2%)

Feature Importance Ranking

Telematics speed variance 38%
Prior loss ratio history 26%
Fleet maintenance age 18%
Operator shift duration 12%

Data drift status: Stable (0.014 PSI)

Figure 4.9 · Connect6 dual-view customer chat and CSR context card Human-in-the-loop escalation

Customer Chat Stream Channel: Mobile Web

I had a collision on highway I-80 near Exit 42. My car won't start and I have my kids with me. Need a tow right now.

I have noted your location on I-80 Exit 42. Confirming policy #CA-90412 includes full roadside towing and vehicle rental benefits.

Escalated to human representative: High urgency emergency assistance

Sarah (Live Agent): "Hello Michael, I see your location at Exit 42. Tow truck is dispatched and estimated in 14 minutes. Should I arrange a rental car for pickup?"

Representative Context Card Full context preserved

Policyholder Michael Chang

Policy status #CA-90412 · Active

Priority level High Urgency

Generated claim ID #CLM-2026-9914

Figure 4.10 · Agent6 node-based workflow builder Higher education admissions pipeline

1. Ingestion

Document Upload

Applicant uploads academic transcript via portal.

→

2. Extraction

Entity Parsing

Agent6 OCR extracts GPA, credit hours, and school code.

→

3. Reasoning

Merit Evaluation

Automated rules check for scholarship eligibility.

→

4. Sync

Banner SIS Sync

Verified data writes securely to core database.

05 / Outcomes & business impact

Measurable Business & UX Outcomes

Restructuring the platform architecture, establishing the Think6 Framework, and improving operational visibility delivered verified gains across enterprise benchmarks.

Performance indicator	Legacy baseline (Xemplar AI)	Bay6 AI platform	Operational impact
Support query deflection	12% – 18% (scripted FAQ bot)	40% – 60%	Substantial drop in routine inquiries; eliminates representative burnout during peak seasons.
Predictive scoring accuracy	72% (experimental prototypes)	90%+ in production	High-confidence loss ratio forecasting and early intervention for policy and student retention.
POC-to-production cycle	12–18 months prolonged pilots	< 90 Days	Standardized Model6 Lite onboarding and pre-trained domain templates reduced deployment friction by 75%.
Website bounce rate	74% (generic vendor positioning)	26% (-65% drop)	Interactive workflow diagrams immediately clarified operational value to enterprise buyers.
Enterprise demo inquiries	~8–10 per quarter	+300% Inbound surge	Clear product naming architecture enabled technology leaders to present adoption proposals internally.
Support representative satisfaction	2.9 / 5.0 (fragmented context)	4.8 / 5.0 Rating	Single-pane AI Context Card eliminated repetitive user questions during live escalations.

06 / Testing & design iterations

Design Decisions & Usability Iterations

Key adjustments derived from 24 usability review sessions with enterprise system architects, claims adjusters, and customer service teams.

Iteration 01

Replacing decorative visuals with workflows

Initial explorations included abstract 3D elements. Enterprise evaluators found them detached from daily operations.

Design adjustment: Transitioned to interactive workflow diagrams with clear input, processing, and output stages.

Iteration 02

Preventing notification fatigue

Frequent pop-up banners during shift peaks caused representatives to dismiss critical customer alerts.

Design adjustment: Implemented a persistent, organized context panel displaying severity indicators and pre-filled identifiers.

Iteration 03

Clarifying model performance telemetry

Complex statistical error charts made it difficult for operational managers to know when models required review.

Design adjustment: Introduced clean feature-weight bar charts and straightforward stability status indicators.