We have secured critical CIO approval to establish and own our dedicated Azure Cloud Infrastructure for Data & AI Platforms. We've already proven we can execute: our Data Intelligence Platform (Azure Databricks, Microsoft Fabric, Unity Catalog, Microsoft Purview) is fully operational—delivered in just 7 weeks through agile collaboration between our Enterprise Data Architect, Cloud Solution Architect, and an Azure Cloud Infrastructure Engineer from EIS.
This momentum positions us perfectly for the next challenge: the Mazda Sales Promise Agent must be operational within 3 months to support enterprise-wide MSP execution and dealer enablement.
- MSP launch timing requires AI-powered support for project execution, dealer enablement, and compliance monitoring
- 500+ users across dealer network and internal teams need intelligent assistance NOW
- Current manual coordination creates bottlenecks and inconsistent dealer experience
- We've proven rapid delivery capability—now we apply it to AI
- Data Intelligence Platform already operational provides the foundation
- Copilot Studio enables rapid deployment using out-of-the-box capabilities
- Atlassian Confluence/Jira and SharePoint integration provides immediate knowledge access
- $8-12M annual value from improved MSP execution and dealer support
We have the approval, the proven delivery model, the operational platform, and the business case. Every week we delay is a week without intelligent dealer support and competitive advantage.
Annual benefits now being captured and tracked, but stalled due to lack of foundational capabilities
Accelerating towards realizing this value each week
AI initiatives faced extensive coordination and delays due to Enterprise Infrastructure Services (EIS) dependencies, creating bottlenecks and slowing innovation.
We've successfully transitioned to infrastructure ownership with a major achievement: Data Intelligence Platform fully operational in 7 weeks.
- Azure Databricks: Data engineering and ML workloads operational
- Microsoft Fabric: Unified analytics platform deployed
- Unity Catalog: Data governance framework active
- Microsoft Purview: Compliance and discovery capabilities live
Cross-functional agile team working in 2-week sprints:
- Enterprise Data Architect (leadership and design)
- 2 Cloud Data Infrastructure Engineers (platform implementation)
- 1 Azure Cloud Infrastructure Engineer from EIS (infrastructure support)
- Delivered ahead of schedule through focused execution and collaboration
- Removed silos between teams for adequate access
- Executive support enabled plan execution without outside intervention
- Omotenashi mindset: built for sustainment from day one
- Sharp focus: team stayed dedicated to the objective
With the data foundation operational, we now build AI capabilities on top of this proven platform. The same agile, focused approach that delivered the Data Intelligence Platform in 7 weeks will deliver the Mazda Sales Promise Agent in 3 months.
This isn't a proposal—it's continuing proven momentum.
We've proven the builder/owner model works. In just 7 weeks, our cross-functional team delivered:
- Azure Databricks, Microsoft Fabric, Unity Catalog, and Microsoft Purview
- Fully operational and supporting data workloads today
- Delivered ahead of schedule through agile 2-week sprints
- Team: Enterprise Data Architect + 2 Cloud Data Infrastructure Engineers + 1 Azure Cloud Infrastructure Engineer (EIS)
Based on our Project ELMO:
- Collective Opportunity: Cross-functional collaboration without silos
- Building Capabilities: Focus on sustainable, long-term solutions
- What's Possible: Prioritize and measure continuously
- Enablement: Never enough time, but always enough focus
- Continuous Learning: Fail fast, learn faster
With CIO approval and proven delivery capability, we now expand the builder/owner model to AI:
- Foundation Ready: Data Intelligence Platform operational provides the base
- AI Infrastructure Next: WAF/CAF-aligned Azure AI Foundry and Azure AI Services
- Power Platform Architecture: Enterprise-wide Copilot Studio enablement
- Same Agile Approach: 2-week sprints, cross-functional teams, sharp focus
We're not starting from zero. We have:
- ✓ Operational data platform
- ✓ Proven delivery team and methodology
- ✓ Executive support and CIO approval
- ✓ Track record of ahead-of-schedule delivery
The AI Dojo builds on this foundation with the same empowerment and enablement approach that made Project ELMO successful.
CIO Approval Unlocks New Possibilities:
Our recent approval to own Azure Cloud Infrastructure for Data & AI Platforms fundamentally changes what's possible:
Key Strategic Advantages:
- Speed: Direct infrastructure control eliminates dependency bottlenecks
- Integration: Unified Data Intelligence Platform (Databricks, Fabric, Unity Catalog, Purview) enables seamless data-to-AI workflows
- Governance: We control security, compliance, and operational standards aligned to WAF/CAF
- Innovation: Direct access to Azure AI Foundry and AI Services enables rapid experimentation
- Enterprise Enablement: Power Platform architecture scales Copilot Studio across the organization
- Cost Control: FinOps practices and direct Azure management optimize spending
- Microsoft Partnership: FastTrack and direct support accelerate capability building
Market Timing:
- AI technology maturity makes this the right time to build production-grade infrastructure
- Microsoft's investment in Azure AI creates partnership opportunity
- Competitive pressure requires faster AI deployment than dependency models allow
This isn't just about AI solutions - it's about building the platform capability Mazda needs for the next decade.
We propose a measured, risk-managed approach that progressively builds capability while delivering tangible value at each stage. This phased rollout allows us to prove the model, learn from experience, and scale with confidence.
Establish foundations with quick wins with joint development of base infrastructure
Build platform capabilities and expand delivery with partial autonomy
Achieve full autonomy and deliver highest-impact strategic initiatives
Build AI capabilities on our operational Data Intelligence Platform while delivering immediate business value through the Mazda Sales Promise Agent
Immediate Priority - Mazda Sales Promise Agent (Months 1-3)
- Deploy using out-of-the-box Copilot Studio functionality
- Integrate two primary knowledge sources: Atlassian Confluence/Jira and Microsoft SharePoint
- Support strategic intent, business case, project execution, dealer enablement, risk & compliance monitoring
- Enable agentic mesh architecture with child agents for specialized functions
- Deliver enterprise-wide support for MSP execution
AI Infrastructure Build (Parallel Track)
- Establish Power Platform architecture to support Copilot Studio at scale
- Implement WAF/CAF-aligned infrastructure for Azure AI Foundry and Azure AI Services
- Set up governance framework for agent deployment and monitoring
- Integrate AI capabilities with existing Data Intelligence Platform (already operational)
Our Data Intelligence Platform is already operational (delivered in 8 weeks):
- ✓ Azure Databricks - ready for ML workloads
- ✓ Microsoft Fabric - unified analytics available
- ✓ Unity Catalog - governance framework active
- ✓ Microsoft Purview - compliance monitoring live
This phase builds AI capabilities on top of proven infrastructure using the same agile delivery approach that made Project ELMO successful.
Our Data Intelligence Platform team has already demonstrated rapid delivery capability (8 weeks, ahead of schedule). We expand this proven team for AI enablement:
- Enterprise Data Architect: Platform design, architecture governance, cross-team coordination
- Cloud Data Infrastructure Engineers (2): Data platform operations, integration, performance optimization
- Azure Cloud Infrastructure Engineer (EIS): Infrastructure support, Azure resource management
- Dojo Lead: Overall AI vision, CIO alignment, Microsoft partnership management, business case ownership
- Platform Architect (AI Focus): Designs WAF/CAF-aligned infrastructure for Azure AI Foundry, Azure AI Services, and Power Platform
- AI Engineer: Builds Mazda Sales Promise Agent and agentic mesh architecture in Copilot Studio
- Integration Specialist: Connects Copilot Studio with Atlassian Confluence/Jira and SharePoint knowledge sources
Team Size: ~7-8 members (4 existing + 3-4 new)
- Agile 2-week sprints (same approach that delivered Data Intelligence Platform)
- Cross-functional collaboration without silos
- Sharp focus on objectives with executive support
- Omotenashi mindset: build for sustainment from day one
This isn't a new team—it's expanding a proven winner.
Supporting MSP (Mazda Sales Promise) execution and dealer enablement across the enterprise. Eliminates friction, personalizes experience, and unifies customer engagement.
A multi-agent system leveraging:
- Base Copilot Studio Agent: M365 Generative AI Chat Based orchestrator using M365 Apps
- Child Agents:
- Risk & Compliance Agent
- Dealer Support Agent
- Project Execution Agent
- Data Insights Agent
- Atlassian Confluence/Jira (project execution, tracking, requirements)
- Microsoft SharePoint (documentation, policies, dealer resources)
- Public websites (as needed)
- Dataverse and Files
- Connectors (Atlassian integration)
- Flows (automations)
- Prompts (pre-defined and business-defined)
- MCP (Fabric Data Agent integration)
3 months from kickoff
This is a rapid deployment approach leveraging out-of-box Copilot Studio capabilities to deliver value within 3 months while foundational infrastructure is being established in parallel.
Begin testing the AI Agent as required by the MSP team.
Ensure Environment Makers have permissions for build out and connector configurations.
Confirm requirements and timing for MSP launch.
Ensure the Production environment is ready to support the business timing.
Infrastructure is contacting TechM to build the environments.
- Week 1-2: POC environment setup, permissions configuration
- Week 3-4: Atlassian and SharePoint connector integration
- Parallel: Begin Azure infrastructure planning and Microsoft gap assessment and validations using support hours
- Week 1-2: Base Copilot Studio agent development with knowledge sources
- Week 3-4: Child agent development (Risk & Compliance, Dealer Support, Project Execution, Data Insights)
- Parallel: Power Platform architecture deployment, initial governance framework
- Week 1-2: Production environment deployment, user acceptance testing
- Week 3-4: Enterprise rollout, user training, adoption support
- Parallel: Data Intelligence Platform initial deployment (Databricks, Fabric foundations)
- Mazda Sales Promise Agent operational with sign off from the business
- Agentic mesh architecture proven with 4 child agents
- Platform foundation established for Phase 2 expansion
- $8-12M annual value delivery path validated
This accelerated timeline proves rapid delivery capability while building infrastructure foundation.
Our Data Intelligence Platform is already fully deployed and operational (delivered in 7 weeks):
- ✓ Azure Databricks: Data engineering and ML workloads running
- ✓ Microsoft Fabric: Unified analytics platform active
- ✓ Unity Catalog: Data governance enforced
- ✓ Microsoft Purview: Compliance and discovery operational
Building AI capabilities on this proven foundation:
- Partner Assessments & Validation: Leverage Microsoft Support Hours for Azure Foundations
- AI-Specific Guidance: Copilot, Azure AI Foundry, and Azure AI Services validation support
- Power Platform Expertise: Copilot Studio enterprise architecture and best practices
- WAF/CAF Implementation for AI: Extend Well-Architected Framework and Cloud Adoption Framework to AI services
- Azure AI Foundry: Model development and deployment infrastructure
- Azure AI Services: Pre-built AI capabilities (document intelligence, speech, vision)
- Power Platform Architecture: Enterprise-grade Copilot Studio deployment supporting agentic mesh
- Connect Copilot Studio agents to Fabric data sources via MCP (Model Context Protocol)
- Leverage Unity Catalog for AI model governance
- Use Purview for AI compliance monitoring and data lineage
- Databricks for custom model training when needed
Extend existing Data Intelligence Platform governance to AI workloads with enterprise-grade controls we own and manage.
We're not starting from scratch—we're building the AI layer on operational infrastructure.
- Mazda Sales Promise Agent Operational: Enterprise-wide deployment supporting MSP execution, dealer enablement, project tracking, risk & compliance
- Agentic Mesh Architecture Proven: Base orchestrator with 4 specialized child agents (Risk & Compliance, Dealer Support, Project Execution, Data Insights)
- Knowledge Integration Complete: Atlassian Confluence/Jira and SharePoint fully integrated as primary knowledge sources
- User Adoption Metrics: 500+ users across dealer network and internal teams within first 3 months
- Power Platform architecture operational and supporting Copilot Studio at scale
- Azure AI Foundry and Azure AI Services infrastructure deployed with WAF/CAF compliance
- Governance framework operational for agent deployment and monitoring
- AI capabilities integrated with existing Data Intelligence Platform
- Data Intelligence Platform (Databricks, Fabric, Unity Catalog, Purview) already supporting AI workloads
- Proven agile delivery model (2-week sprints) applied to AI development
- Cross-functional team collaboration continuing from Project ELMO success
- Integration between AI agents and data platform operational
- Second major delivery in 3 months (following 7-week Data Intelligence Platform success)
- Proven builder/owner model delivering both infrastructure and solutions
- Patterns Documented with Plan - Do - Check - Act (PDCA) leveraging Microsoft & EEARB as validation mechanisms
- Team capability demonstrated across data and AI domains
- Executive confidence in rapid delivery model reinforced
The Mazda Sales Promise Agent demonstrates we can deliver AI solutions as fast as we delivered the data platform.
Mitigation: Microsoft Support partnership, phased rollout, experienced platform architects
Mitigation: Strategic hiring, Microsoft training programs, external consulting for knowledge transfer
Mitigation: WAF/CAF frameworks from day one, security reviews at each milestone, automated compliance monitoring
Mitigation: Parallel workstreams, choose pilot use cases that can run on partially-complete platform, incremental platform maturity
Mitigation: Unity Catalog as central governance layer, clear data architecture, integration testing at each phase
- Data Intelligence Platform delivered in 7 weeks (Azure Databricks, Fabric, Unity Catalog, Purview)
- Proven agile delivery model with 2-week sprints
- Cross-functional team collaboration working effectively
- Executive support and empowerment validated
- Mazda Sales Promise Agent operational in 3 months
- Azure AI infrastructure deployed with WAF/CAF compliance
- Power Platform architecture supporting Copilot Studio at scale
- Agentic mesh architecture proven with 4 child agents
- $8-12M annual value from Mazda Sales Promise Agent
- Two consecutive rapid deliveries (8 weeks for data, 3 months for AI)
- Platform reliability and performance demonstrated
- Team capability proven across data and AI domains
- Deploy multiple concurrent AI projects on mature platform foundation
- Build advanced MLOps pipeline leveraging Databricks and AI Foundry
- Scale team with confidence in proven delivery model
- Expand agentic mesh to additional use cases
Momentum Multiplier: Each success builds confidence and capability. Data Intelligence Platform → Mazda Sales Promise Agent → Enterprise AI at scale.
We're not building from zero—we're accelerating from proven success.
Phase 2 represents the critical transition from dependency to capability. We establish the technical infrastructure and team capacity needed to operate at scale, while simultaneously expanding our delivery of high-value AI solutions.
Build dedicated Azure Landing Zone with dev/test environments, CI/CD pipelines, and governance guardrails
Grow to ~12 core members with critical platform engineering and additional delivery capabilities
Tackle 3-5 projects simultaneously, demonstrating increased throughput and value
Operationalize Definition of Ready/Done, RAI checklists, and automated compliance checks
Building on Operational Foundation:
With Data Intelligence Platform operational since Phase 0 (delivered in 7 weeks) and AI infrastructure established in Phase 1, Phase 2 focuses on maturity and advanced capabilities:
Our operational Data Intelligence Platform (Databricks, Fabric, Unity Catalog, Purview) gets advanced capabilities:
- Advanced ML workloads and model training in Databricks
- Real-time data streaming and processing
- Enhanced Unity Catalog governance for AI models and sensitive data
- Expanded Purview integration for AI compliance monitoring
- Advanced Azure AI Foundry capabilities for complex model deployment
- Expanded Azure AI Services integration (document intelligence, speech, vision)
- Enterprise-scale Copilot Studio deployment across multiple use cases
- MLOps pipeline connecting Databricks, AI Foundry, and production deployment
- Multi-environment strategy (dev, test, prod) fully operational across data and AI platforms
- Disaster recovery and business continuity capabilities
- Advanced monitoring and observability across all platforms
- Cost optimization and FinOps practices
- Continued Microsoft Support for advanced scenarios
- Early access to preview features and roadmap alignment
- Joint innovation initiatives
The foundation is operational—Phase 2 adds sophistication and scale.
A cornerstone of Phase 2 is establishing the machine learning operations capability that enables rapid, reliable model deployment:
Data scientists experiment and train models in Azure ML workspace with full tracking and versioning
Automated pipeline runs model validation, bias checks, security scans, and performance benchmarks
Successful builds deploy to test environment for user acceptance and integration testing
Approved models deploy to production with monitoring, logging, and automated alerting enabled
Continuous tracking of model performance, data drift, and business metrics with feedback loop
Growing from 5 to ~12 members requires strategic hiring across multiple disciplines:
Sets up Landing Zone, implements Infrastructure-as-Code, manages cloud resources. Initially contractor, convert to FTE.
Builds CI/CD pipelines, implements model registry and versioning, ensures DevSecOps practices. FTE by end of phase.
Specialized in NLP, ML modeling, and data analytics. Mix of FTE hires and contractors for surge capacity.
Add 1-2 more FVLs to manage parallel workstreams. Domain expertise in sales, aftersales, and supply chain management
With expanded capability, we target medium-complexity, high-value opportunities requiring data integration and ML expertise:
Annual Value: $1.86M
Automates cross-checking of import duty documents, flagging discrepancies and identifying refund opportunities. Requires integration with customs databases and SAP.
Annual Value: $750K
AI assistant for warranty team to instantly answer dealer inquiries by referencing warranty policies and technical service bulletins. Scales Q&A capability to broader audience.
Annual Value: $972K
Bilingual (Spanish/English) AI support for Mexico operations. Tests localization capability and cross-regional collaboration.
Mazda imports thousands of vehicles and parts annually, each subject to customs duties. Manual reconciliation of declarations against invoices and payments is time-consuming and error-prone, leading to:
- Unclaimed duty refunds
- Compliance risks from documentation gaps
- Significant auditor time on repetitive checks
Machine learning model trained on historical audit data identifies patterns and anomalies:
- Ingests data from customs systems and SAP
- Flags discrepancies for auditor review
- Learns from auditor decisions to improve
- Prioritizes highest-value opportunities
This project validates our ability to handle complex data integration and deploy ML models requiring ongoing learning—essential capabilities for future initiatives.
While full deployment is reserved for Phase 3, we begin groundwork on what will become our highest-impact initiative:
- Analyze historical lead data to understand patterns and outcomes
- Build proof-of-concept AI agent on test data
- Develop response generation using GPT models fine-tuned on vehicle information
- Validate technical feasibility with 1-2 pilot dealers
- Identify integration points with dealer CRM systems
This de-risks the Phase 3 full rollout, ensuring we understand the technical and operational requirements before scaling to all 540 dealers.
Volume requiring follow-up
Current 90-day conversion
At 1.5% improvement
Platform build begins. Team onboarding. Project discovery for Audit AI and Warranty Agent.
Platform MVP ready. Development sprints in full swing. First pipeline deployment to non-prod.
End-to-end testing complete. Warranty Agent ready for deployment.
Warranty Agent goes live. Audit AI in final testing. Platform refinements based on learnings.
Audit AI deployed. Mexico Agent in development. Dealer Lead pilot begins.
Phase 2 review. Updated EIS OLA. Phase 3 planning and stakeholder alignment.
With multiple concurrent projects, we institute formal coordination mechanisms and quality gates:
Before entering development: data sources identified and access approved, success metrics defined, RAI considerations documented, stakeholders aligned on scope
Before production: all tests pass, CI/CD pipeline executes successfully, security scans clean, documentation complete, training materials ready, business owner sign-off obtained
Bi-weekly synchronization across workstreams to manage dependencies and shared resources. Led by Delivery RTE.
Monthly executive review of progress, interim results, and escalations. Ensures leadership alignment and rapid decision-making.
Live in production
Vs. Phase 1 delivery
Tasks handled internally
Phase 2 Incremental Benefit: $5-7M annually
Cumulative Portfolio Value: $8-10M annually
Phase 2 Investment: ~$900K
The platform investment yields immediate dividends through faster delivery cycles, with projects moving from 4-5 months to 2-3 months from kickoff to production.
Impact: Could slow project development
Mitigation: Expert contractor for rapid build. EA/CISO/EEARB involvement in design for buy-in. Phased approach starting with basic dev environment before full capabilities.
Impact: Potential knowledge gaps or delayed productivity
Mitigation: Begin recruiting in Phase 1. Use contractors as interim. Pair new hires with Phase 1 team members. Allocate time for Mazda context training.
Impact: Multiple projects competing for shared resources
Mitigation: Stagger project start dates. Clear prioritization framework. RTEs and FVLs coordinate frequently. Maintain backlog flexibility to adjust if needed.
Impact: Solutions built but not properly utilized
Mitigation: Half-time Adoption Lead ensures each solution has business owner. Training sessions and quick reference guides. Monitor usage metrics and address barriers.
Phase 2's success creates the conditions for Phase 3's transformative impact. By demonstrating responsible operation of our own platform while accelerating delivery, we earn the trust needed for full Builder/Owner autonomy.
- Zero security incidents despite increased velocity
- All projects passed privacy and compliance reviews
- EIS change requests reduced from 10 to 3 per phase
- Average turnaround improved from 4 weeks to 1 week
- Documented adherence to governance frameworks
Based on Phase 2 performance, we negotiate an updated OLA with EIS that further reduces touchpoints:
- Pre-approved operations whitelist for Dojo
- Production deployment authority within guardrails
- EIS monitoring role vs. gatekeeper role
- Exception-based engagement model
Phase 3 represents the full realization of the Builder/Owner model. The AI Dojo now operates as an AI factory—a high-velocity, highly-governed capability that continuously delivers solutions driving revenue, cost savings, and competitive advantage.
Complete authority to design, build, test, deploy, and maintain AI solutions with only high-level oversight
Tackle the highest-impact, most complex initiatives including dealer-facing and revenue-generating solutions
Fast flow mode with frequent releases, quick iterations, and ability to respond to new demands in weeks not months
Robust governance scales with velocity—policy-as-code, automated checks, and mature risk management
Building on infrastructure ownership established in Phase 1, Phase 3 represents operational excellence:
- Self-service capabilities for AI teams to provision resources within governance guardrails
- Automated compliance monitoring and remediation
- Advanced FinOps with cost allocation and optimization
- Platform-as-a-service model for internal customers
- 99.9% platform uptime with automated failover
- Continuous platform updates with zero-downtime deployments
- Advanced security posture with threat detection and response
- Comprehensive observability and performance optimization
- Supporting 5-8 concurrent AI solution deliveries per quarter
- Power Platform/Copilot Studio deployed across multiple business units
- Data Intelligence Platform serving analytics and AI workloads enterprise-wide
- Center of Excellence model with reusable patterns and accelerators
The crown jewel of the AI Dojo's portfolio: an intelligent system that transforms how Mazda dealers engage with potential customers.
- 70,000 leads per month across 540 dealers
- Current 23% baseline close rate within 90 days
- Speed of response strongly correlates with conversion
- BDC staff struggle to respond promptly and consistently
- Each 1% improvement = 8,400 additional sales annually
- AI agent ingests leads via email or CRM API integration
- Generates personalized responses using GPT fine-tuned on vehicle data
- Incorporates dealer-specific inventory and pricing
- Provides draft to salesperson or auto-responds based on dealer preference
- Continuous learning from successful conversions
Mid-case scenario detail: 12,600 additional sales × $2,408 profit per vehicle = $30.3M incremental profit, plus $3.2M in labor savings from automated follow-up, totaling $33.5M net annual benefit.
Build production-grade agent infrastructure. Integrate with dealer CRM systems. Establish vehicle data feeds. Create response templates and quality guardrails.
Expand beyond Phase 2 test to 10-15 dealers. Measure conversion impact. Refine based on dealer feedback. Train model on successful interactions.
Present results to dealer advisory councils. Develop training materials and support resources. Coordinate with Sales Operations on rollout plan.
Phased deployment to all 540 dealers in waves. Provide white-glove support for early adopters. Monitor performance and address issues rapidly.
Continuous improvement based on conversion data. Expand features based on dealer requests. Measure ROI and document success stories.
Building on the Mazda Sales Promise Agent success from Phase 1, Phase 3 expands the agentic mesh architecture enterprise-wide:
- Manufacturing Operations Agent (production planning, quality monitoring)
- Supply Chain Intelligence Agent (logistics optimization, supplier coordination)
- Customer Experience Agent (post-sale support, warranty processing)
- Financial Planning Agent (budget analysis, forecasting support)
- Cross-agent orchestration for complex multi-domain workflows
- Proactive insights and recommendations based on pattern detection
- Integration with Azure AI Foundry for custom model deployment
- Real-time data integration via Fabric Data Agents
- 2,000+ active users across all business functions
- 10+ specialized child agents supporting various domains
- Automated agent deployment and lifecycle management
- Comprehensive analytics on agent performance and business impact
Expected Value: $15-20M annually from operational efficiency and decision support across the enterprise
Building on earlier successes, we scale and integrate our AI assistants into a comprehensive knowledge platform:
Phase 1 success expanded with enhanced capabilities and integration
Phase 2 deployment supporting dealer technical inquiries
Level-1 helpdesk automation for password resets and software requests
Spanish for Mexico, French for Canada expanding accessibility
Cross-agent insights on common questions and knowledge gaps
Phase 3 brings existing analytics work into reliable production deployment:
- Owner Retention Models: Predict which customers are at risk of switching brands and trigger proactive engagement
- Marketing Propensity: Target campaigns to highest-probability prospects
- Inventory Optimization: ML-driven predictions for optimal dealer stock levels
- Service Demand Forecasting: Anticipate parts needs and service center capacity
The MLOps infrastructure built in Phase 2 enables these models to run reliably with automated retraining, monitoring, and integration into operational systems—transforming proof-of-concepts into business value.
Phase 3 team reaches ~18-20 members organized into specialized squads with platform support:
5-6 members dedicated to dealer-facing AI including Lead Agent and Sales Promise. Includes FVL, developers, dealer relationship manager.
5-6 members focusing on internal process AI for HR, Finance, Audit, and IT. Includes FVL, data scientists, integration specialists.
5-6 members providing shared services: platform engineering, MLOps, architecture, RAI oversight, change management.
15-20% of capacity allocated to exploring emerging AI capabilities, technical debt reduction, and continuous learning.
- Quarterly PI Planning: All teams align on objectives for next 10-12 weeks
- Two-Week Sprints: Regular delivery cadence with demos and retrospectives
- Daily Standups: Within squads for coordination and blocker removal
- Weekly Sync: Cross-squad dependencies and resource allocation
- Monthly Steering: Executive review of progress and strategic decisions
Squads own their solutions in production:
- On-call rotations for critical systems
- Monitoring dashboards for each application
- Incident response procedures
- Support ticket triage and resolution
- Continuous optimization based on usage patterns
This DevOps approach ensures solutions remain healthy and high-performing long after initial deployment.
Speed without control is recklessness. Our governance model scales with our velocity through automation and clear accountability:
Azure Policy and automated scans enforce security and compliance standards. No sensitive data in external AI prompts detected automatically.
Pipeline includes bias testing, security scanning, performance validation. Issues block deployment until resolved.
All solutions instrumented with telemetry. Dashboards track usage, performance, errors, and business outcomes.
RASCI matrices define who's Responsible, Accountable, Supportive, Consulted, Informed for every process and decision.
Quarterly governance reviews identify improvement opportunities. Retrospectives after incidents lead to process updates.
- Govern: AI Steering Committee, policies, and risk register operational
- Map: Risk assessment standard for every use case
- Measure: Defined metrics for accuracy, fairness, performance tracked continuously
- Manage: Incident response procedures and model retraining triggers established
- Leadership commitment via Steering Committee (EEARB)
- Comprehensive planning and risk management
- Training and competency programs
- Operational controls from intake to deployment
- KPI tracking and quarterly management reviews
- Feedback loops for continuous improvement
By Phase 3 end, MNAO has the option to pursue external certification of our AI management system—demonstrating leadership in responsible AI governance to stakeholders, regulators, and customers.
New AI capabilities delivered continuously
Idea to production
for moderate projects
Cumulative benefit by Year 2 end
Benefit vs. two-year total investment
- Dealer Lead AI: $21.9M (mid-case)
- Sales Promise AI: $50M program optimization
- Phase 1-2 Solutions: $8-10M maintained and enhanced
- Additional Projects: $10-15M
- Permanent AI development function
- Platform for continuous innovation
- Proven governance model
- Cross-trained team with deep Mazda knowledge
Risk: Incorrect AI output causes customer or business harm
Mitigation: Human-in-loop for critical communications initially. AI constrained to known topics. Regular quality reviews. Phased rollout with opt-in approach. Comprehensive testing and validation.
Risk: Users resist AI adoption fearing job replacement or impersonal service
Mitigation: Position AI as assistant not replacement. Show data on success improvements. Celebrate early adopters. Emphasize how AI makes jobs more interesting by handling repetitive work.
Risk: Skilled team members recruited away by external offers
Mitigation: Competitive compensation. Positive team culture. Career growth opportunities. Highlight meaningful impact of work. Exposure to cutting-edge technology and problems.
Risk: Autonomy leads to unchecked problems
Mitigation: Monthly EEARB reviews. Open invite for Internal Audit. Transparent reporting. Maintain high visibility even as we move fast.
At each phase, leadership can adjust investment and velocity based on appetite for speed versus risk. We've designed three option tracks:
Philosophy: Minimize cost and risk with lean teams and sequential delivery
Best For: Risk-averse culture or budget constraints. Proves concept before scaling.
Trade-off: Slower value capture, may lose competitive ground
Philosophy: Optimize for sustainable growth with proven ROI at each stage
Best For: Most organizations balancing speed and prudence
Trade-off: None—best risk-adjusted return
Philosophy: Maximize speed with heavy upfront investment
Best For: Urgent competitive pressure or executive mandate for rapid transformation
Trade-off: Higher risk of inefficiency if processes aren't mature
$350K - Mazda Sales Promise Agent & infrastructure foundation (3 months)
$900K - Platform maturity & expansion
$3.5M - Full operation (annual)
Cumulative Two-Year Investment: $4.75M
- Phase 1 Benefit: $10-15M/year from Mazda Sales Promise Agent
- MSP execution efficiency and dealer enablement
- Reduced project coordination overhead
- Faster dealer onboarding and support
- Improved compliance monitoring
- Phase 2 Incremental: +$15-20M/year (Customs Audit AI, expanded agents)
- Phase 3 Incremental: +$35-50M/year (Dealer Lead AI, enterprise agent mesh)
Year 2 Run-Rate: $60-85M annual benefit
Benefit-to-cost ratio over two-year period
The Mazda Sales Promise Agent delivers immediate, substantial ROI that funds subsequent phases, with total annual benefits reaching $60-85M.
Every week we delay this initiative costs Mazda $1.49 million in unrealized value from our identified use case backlog.
Consider just the Dealer Lead AI opportunity: at current volumes and conversion rates, we're missing ~700 additional sales per month that this solution could enable. That's $1.8 million in lost profit monthly, or $21.6 million annually.
Meanwhile, competitors are rapidly deploying AI across their operations. The gap between early movers and followers grows wider each quarter. This isn't about keeping pace—it's about competitive survival.
If we don't act
From lead AI alone
We will track progress through clear, measurable indicators at each phase:
- Throughput: Number of solutions delivered per quarter
- Lead Time: Average days from kickoff to production
- Cycle Time: Average days in active development
- Deployment Frequency: Releases per month
- Benefits Realized: Actual savings/revenue vs. projected
- Adoption Rate: % of target users actively using each solution
- User Satisfaction: NPS or CSAT scores for AI tools
- ROI: Cumulative benefit-to-cost ratio
- Incidents: Production issues per solution per month
- Model Performance: Accuracy, precision, recall tracked continuously
- Compliance: % of projects passing all governance gates
- Technical Debt: Outstanding issues and velocity impact
- Team Satisfaction: Engagement and retention rates
- Knowledge Growth: Skills acquired and certifications achieved
- Platform Uptime: Availability of dev/prod environments
- Autonomy Index: % of work completed without external dependencies
Quarterly business reviews will track these metrics against targets, with executive dashboards providing real-time visibility into Dojo performance.
The AI Dojo would operate within a robust governance framework ensuring alignment with enterprise strategy and risk management:
Quarterly strategy reviews, major investment decisions, cross-functional alignment
Monthly oversight of portfolio, risk reviews, policy updates, phase gate approvals
Reviews use cases for ethical implications, bias testing, fairness assessments
Validates solution designs, ensures standards compliance, approves exceptions
Day-to-day execution, sprint planning, delivery, continuous improvement
Technology alone doesn't deliver value—adoption does. Our change management approach ensures each AI solution achieves its potential:
- Stakeholder Engagement: Early involvement of business owners and end users in design
- Communication Plans: Clear, consistent messaging about what's changing and why
- Training Programs: Role-specific education on using new AI tools
- Champion Networks: Identify and empower enthusiasts to drive peer adoption
By Phase 3, we have a full-time Change & Adoption Lead ensuring every solution has a path to realizing its projected benefits through effective organizational change.
Beyond the two-year roadmap, the AI Dojo evolves into Mazda's permanent AI Center of Excellence:
Continuously exploring emerging AI capabilities and pilot testing new applications
Educating employees across Mazda on AI literacy and hands-on skills development
Documenting lessons learned, reusable components, and proven patterns for AI development
Partnering with Mazda Japan and other regions to share capabilities and insights
Maintaining relationships with AI vendors, academic institutions, and industry consortia
We recommend Option B (Balanced) with immediate Phase 1 execution as the optimal path forward:
- Mazda Sales Promise Agent delivers $8-12M annual value within 3 months
- Proven ROI (23-34x first year) funds subsequent phases without additional justification
- Balances rapid delivery with sustainable infrastructure foundation
- De-risks future phases through early capability demonstration
- Positions us for Phase 2 expansion with operational platform
- Secure Phase 1 Funding: $350K approval for 3-month delivery
- Establish POC Environment: Copilot Studio environment with proper permissions
- Initiate Microsoft Partnership: FastTrack engagement for accelerated deployment
- Confirm Integration Access: Atlassian Confluence/Jira and SharePoint connector permissions
- Assemble Core Team: Dojo Lead, Platform Architect, AI Engineer, Cloud Infrastructure Engineer
- Align with MSP Timeline: Confirm business requirements and launch coordination
Approve Phase 1 funding and team allocation to begin delivery within 2 weeks
The Mazda Sales Promise Agent is not just a pilot—it's immediate business value that proves our builder/owner model.
Annual value within reach in 24 months—starting with $8-12M in the first 3 months
The AI Dojo represents more than a technology initiative—it's a strategic transformation of how Mazda operates in an AI-driven world.
We have everything we need to start:
- CIO approval to own our Azure infrastructure ✓
- Proven technology in Copilot Studio ✓
- Clear business need with MSP launch timing ✓
- Knowledge sources ready (Atlassian, SharePoint) ✓
- $8-12M annual value opportunity in first 3 months ✓
The Mazda Sales Promise Agent is our proof point. In 90 days, we can demonstrate:
- Rapid delivery capability (3 months from approval to production)
- Substantial ROI (23-34x in first year)
- Infrastructure ownership model working
- Foundation for enterprise-wide AI transformation
Every week we delay is a week without intelligent dealer support, efficient MSP execution, and competitive advantage. We have the approval. We have the plan. We have the technology.
The only question is: when do we start?
Our recommendation: Approve Phase 1 funding this week and begin delivery immediately.
Building the AI Dojo to Drive $60M+ in Annual Value