oc-p12-fashion-insta/README.md

AI Project Scoping - Fashion-Insta

📘 Ce projet est également disponible en français 🇫🇷

Source of truth : https://git.gregoiremureau.com/openclassroom-ai/oc-p12-fashion-insta

git clone https://git.gregoiremureau.com/openclassroom-ai/oc-p12-fashion-insta.git

Dataset: see fetch-data.sh (outside git if large).

📋 Project Context

This project was carried out as part of an AI Product Manager assignment at Fashion-Insta, a fashion company that sells clothing items through a network of physical stores and an e-commerce website.

The objective was to carry out the complete scoping of a strategic AI project: the development of a mobile application for recommending clothing items based on photos, in order to present the project to the executive committee (COMEX) for approval.

🎯 Objectives

The project aims to develop a mobile recommendation application allowing users to:

1. Take a photo of themselves wearing their favorite clothes
2. Get recommendations for items in the same clothing style
3. Boost online and in-store sales

Mission assigned: Carry out the entire scoping of the project following an agile SCRUM methodology and convince the COMEX to launch the project.

📊 Scoping Approach

The project scoping took place over several phases across 3 weeks:

Phase 1: Requirements gathering and backlog

• Formalization of user stories based on business needs
• Prioritization using the MoSCoW method (Must have, Should have, Could have, Won't have)
• Identification of the MVP (Minimum Viable Product)
• Review with the Product Owner

Phase 2: Estimation and profitability

• Estimation of effort per user story (in person/days)
• Breakdown by profile (Data Scientist, Developer, DevOps, etc.)
• Calculation of development costs
• Estimation of Azure costs (initial infrastructure and production)
• Profitability analysis (ROI, break-even point)
• Sprint planning

Phase 3: GDPR compliance and risks

• Analysis of personal data (CNIL data processing register)
• Identification of legal and ethical issues
• Analysis of potential model biases
• Complete project risk analysis
• Mitigation action plan

Phase 4: COMEX presentation

• Summary of objectives and expected gains
• Presentation of the agile SCRUM methodology
• Sprint planning and organization
• GDPR and ethical issues
• Risk mitigation

📁 Project Structure

├── P12_01_backlog.xlsx
├── P12_02_tableur.xlsx
├── P12_03_présentation.pdf (ou .pptx)
├── P12_04_ExportedEstimate_initial.xlsx
└── P12_05_ExportedEstimate_production.xlsx

1. P12_01_backlog.xlsx

Project backlog with prioritized user stories

This file contains:

• User stories: Formalization of functional and technical needs from the user's point of view
• MoSCoW prioritization:
  • Must have: Features essential to the MVP
  • Should have: Important but non-blocking features
  • Could have: Desirable features
  • Won't have: Features excluded from the current scope
• Effort estimation: Development time per user story
• Acceptance criteria: Validation conditions for each story
• Dependencies: Relationships between user stories

Focus: AI user stories (recommendation, image processing, ML models)

2. P12_02_tableur.xlsx

Spreadsheet for effort and cost estimation

This file includes several tabs:

"Effort by profile" tab:

• Breakdown of effort by user story
• Profiles: Data Scientist, ML Engineer, Full-Stack Developer, DevOps, Product Owner, Scrum Master
• Calculation of person/days per profile
• Application of daily rates (TJM)

"Development costs" tab:

• Total initial development cost
• Breakdown by profile and by sprint
• Fixed vs variable costs

"Recurring costs" tab:

• Annual maintenance: 15% of development cost
• Production Azure infrastructure costs (annual)
• Cost trends over 3-5 years

"Profitability analysis" tab:

• Estimated annual gains (increase in sales)
• Cumulative costs year after year
• ROI and break-even point calculation
• Profitability chart

"Sprint planning" tab:

• Breakdown into sprints (duration, content)
• Assignment of user stories per sprint
• Milestones and deliverables

3. P12_03_présentation.pdf (or .pptx)

Presentation for the COMEX

This presentation includes:

Slide 1: Executive summary

• Summary of key points (not presented orally)
• Project vision in one slide

Slides 2-3: Objectives and expected gains

• Increase in sales (quantified)
• Improved customer experience
• Success KPIs (adoption rate, conversion rate, satisfaction)

Slides 4-5: Required resources

• Human: Team and profiles needed
• Technical: Azure infrastructure
• Financial: Initial investment and recurring costs

Slides 6-8: Agile SCRUM methodology

• Principles of agility
• Roles (Product Owner, Scrum Master, Development Team)
• Benefits for the project

Slides 9-11: Organization and tracking

• Daily Scrum (daily standup)
• Sprint Review (end-of-sprint demo)
• Retrospective (continuous improvement)
• Burndown chart (progress tracking)

Slides 12-13: Sprint planning

• Time breakdown
• Content of each sprint
• Key milestones and deliverables

Slides 14-16: Legal and ethical issues

• GDPR principles
• Personal data management (photos, preferences)
• Potential model biases (diversity, representativeness)
• Transparency and consent

Slides 17-19: Risks and mitigation

• Technical risks (model performance, scalability)
• Organizational risks (team availability)
• Financial risks (budget overrun)
• Legal risks (GDPR non-compliance)
• Action plan for each critical risk

Appendix: Full backlog

• List of user stories with prioritization
• MVP user stories clearly identified

4. P12_04_ExportedEstimate_initial.xlsx

Azure cost estimation for the initial phase

This file contains the estimation of Azure cloud infrastructure costs for:

• Design phase: Development environment
• Model training phase:
  • Compute instances (GPU for deep learning)
  • Training data storage
  • ML services (Azure Machine Learning)
  • Computer vision services (Computer Vision API)

Export: Azure Pricing Calculator

5. P12_05_ExportedEstimate_production.xlsx

Azure cost estimation for production

This file contains the estimation of Azure cloud infrastructure costs for:

• Application production:
  • Compute instances for the recommendation API
  • Database (product catalog, user history)
  • Blob storage (user images)
  • CDN (Content Delivery Network)
  • ML inference services
  • Monitoring and logging services
  • Bandwidth

Calculation: Monthly and annual costs based on a traffic estimate

🛠️ Methodology

Agile SCRUM Method

Roles:

• Product Owner: Alicia (VP Product)
• Scrum Master: To be defined
• Development Team: Data Scientists, ML Engineers, Developers, DevOps

Ceremonies:

• Sprint Planning: Sprint planning (1-2 weeks)
• Daily Scrum: 15-minute daily standup
• Sprint Review: End-of-sprint demonstration
• Sprint Retrospective: Continuous improvement

Artifacts:

• Product Backlog: Prioritized list of user stories
• Sprint Backlog: User stories selected for the sprint
• Increment: Potentially shippable product at the end of the sprint

MoSCoW Prioritization

• Must have (M): Essential to the MVP, non-negotiable
• Should have (S): Important but can be postponed
• Could have (C): Desirable if budget/time available
• Won't have (W): Excluded from the current scope

Effort Estimation

Techniques used:

• Planning Poker for collaborative estimation
• Story Points or Person/Days depending on the user stories
• Team velocity to adjust forecasts

📈 Profitability Analysis

Costs

Initial investment:

• Development: Total cost of sprints
• Azure infrastructure (initial phase)
• Team training

Recurring costs (annual):

• Maintenance: 15% of development cost
• Azure infrastructure (production)
• Support and enhancements

Gains

Estimated gains:

• Increase in online conversion rate
• Increase in average basket size
• Customer retention (churn reduction)
• Cross-selling and up-selling

ROI calculation:

ROI = (Gains cumulés - Coûts cumulés) / Coûts cumulés × 100

Break-even point: Date at which cumulative gains equal cumulative costs

Visualization

Chart showing:

• Cumulative costs curve
• Cumulative gains curve
• Break-even point (intersection of the curves)

🔒 GDPR Compliance

Personal data processed

• User photos: Sensitive data (images of people)
• Clothing preferences: Behavioral profiling
• Browsing history: Tracking of interactions
• Account data: Email, name, contact details

Data processing register (CNIL)

For each AI processing activity:

• Purpose: Personalized item recommendation
• Data categories: Images, preferences, behavior
• Retention period: Defined and justified
• Security measures: Encryption, pseudonymization
• Recipients: Data team, partners (if applicable)
• Transfers outside the EU: Azure Europe region (compliance)

GDPR Principles Applied

• Explicit consent: Opt-in for the use of photos
• Right of access: Users can view their data
• Right to erasure: Deletion of data upon request
• Portability: Export of personal data
• Minimization: Collection of only necessary data
• Transparency: Clear information about data use

⚖️ Ethical Issues

Model bias

Identified risks:

• Representation bias: Under-representation of certain body shapes, skin colors, clothing styles
• Cultural bias: Recommendations skewed toward Western standards
• Gender bias: Gendered clothing stereotypes

Action plan:

• Diversification of the training dataset
• Regular audit of recommendations
• User testing on varied populations
• User feedback capability to improve the model

Transparency and explainability

• Inform the user about how the recommendation system works
• Ability to understand why an item is recommended
• Option to disable personalized recommendations

⚠️ Risk Analysis

Methodology

Assessment grid:

• Probability: Low / Medium / High
• Impact: Low / Medium / Critical
• Criticality: Probability × Impact

Spectre 7D Checklist:

1. Data: Quality, availability, bias
2. Development: Skills, technologies
3. Deployment: Infrastructure, scalability
4. Dependability: Reliability, performance
5. Detectability: Monitoring, alerts
6. Diversity: Inclusion, fairness
7. Documentation: Traceability, compliance

Main Identified Risks

Technical risks:

• Insufficient performance of the recommendation model
• Latency too high in production
• Scalability difficulties under increased load

Organizational risks:

• Unavailability of resources (Data Scientists)
• Turnover in the project team
• Delays in deliverables

Financial risks:

• Azure budget overrun
• Underestimated development costs
• ROI taking longer than expected

Legal risks:

• GDPR non-compliance (financial penalties)
• User complaints about data management
• Disputes related to model bias

Mitigation Plan

For each critical risk:

• Preventive action: Measure to reduce probability
• Corrective action: Measure if the risk materializes
• Owner: Person in charge of follow-up
• Deadline: Implementation date

Examples:

• GDPR risk → Compliance audit with the DPO before each sprint
• Bias risk → Dataset diversification and regular testing
• Scalability risk → Load testing and elastic cloud architecture

🎓 Skills Developed

• End-to-end AI project scoping
• Agile SCRUM methodology
• Estimation and costing of technical projects
• Profitability analysis (ROI, break-even)
• GDPR compliance and personal data management
• Risk identification and mitigation
• Ethical issues in AI (bias, transparency)
• Stakeholder management (COMEX)
• Project communication and presentation

👤 Author

Grégoire Mureau
Completion date: 2025

---

This project was carried out as part of the AI Engineer program