AI projects fail more often from poor scoping than poor technology:
The operators who deliver don't just understand AI—they understand how to structure work that succeeds.
Step 1: Decompose the Work
Break every deliverable into components:
DELIVERABLE: AI-Powered Customer Inquiry Triage
├─> Data Pipeline
│ ├─> Connect to ticketing system API
│ ├─> Extract and transform inquiry data
│ ├─> Store in vector database
│ └─> Set up refresh schedule
├─> AI Classification
│ ├─> Design category taxonomy
│ ├─> Develop classification prompts
│ ├─> Build retrieval chain
│ └─> Test and tune accuracy
├─> Integration
│ ├─> API for real-time classification
│ ├─> Dashboard integration
│ └─> Alert/routing logic
├─> Measurement
│ ├─> Accuracy tracking
│ ├─> Performance dashboards
│ └─> Feedback loop
└─> Handoff
├─> Documentation
├─> Training
└─> Runbooks
Step 2: Estimate Each Component
For each component, estimate:
| Component | Optimistic | Most Likely | Pessimistic |
|---|---|---|---|
| Ticketing API integration | 8 hrs | 16 hrs | 32 hrs |
| Data extraction | 4 hrs | 8 hrs | 16 hrs |
| Vector DB setup | 4 hrs | 8 hrs | 12 hrs |
| Category taxonomy | 4 hrs | 8 hrs | 16 hrs |
| Classification prompts | 8 hrs | 16 hrs | 32 hrs |
| ... | ... | ... | ... |
Step 3: Apply the PERT Formula
Estimate = (Optimistic + 4×Most Likely + Pessimistic) / 6
This weights toward "most likely" while accounting for uncertainty.
Step 4: Add Buffers
| Project Type | Buffer |
|---|---|
| Well-understood domain, proven approach | 15-20% |
| Familiar domain, new AI application | 25-30% |
| New domain, experimental approach | 40-50% |
Real project plans require roles with responsibilities:
Example from enterprise AI project:
| Role | Responsibilities | Typical Allocation |
|---|---|---|
| Lead Architect | Solution design, technical decisions, quality oversight | 30-40% through project |
| AI Engineer | Infrastructure, model integration, prompt development | 60-80% during build phases |
| Project Manager | Coordination, status, risk management | 20-30% through project |
| Domain Expert | Requirements, data interpretation, validation | 10-20% peaks during phases 1, 4 |
| Data Engineer | Pipeline development, data quality | 50-70% during data phases |
Critical insight: AI projects need more architecture and less pure development than traditional software. The AI does the heavy lifting; humans design and validate.
Traditional: Pick two of (Fast, Good, Cheap)
For AI: Pick two of (Fast, Accurate, General)
| Trade-off | What It Means |
|---|---|
| Fast + Accurate | Narrow use case, well-defined scope |
| Fast + General | Accept lower accuracy, iterate later |
| Accurate + General | Takes longer, more training data/tuning |
Most AI projects should optimize for Fast + Accurate (narrow scope).
General-purpose AI solutions require exponentially more effort. Start narrow, prove value, expand.
| Risk | Indicator | Mitigation |
|---|---|---|
| Scope creep | "Can we also add...?" | Change control process, scope freeze dates |
| Data quality surprise | Data is messier than expected | Early data audit, buffer for cleanup |
| Integration complexity | APIs don't work as documented | Spike early, integration testing |
| Stakeholder availability | Can't get decisions/reviews | Identify backups, schedule in advance |
| Model performance | AI doesn't perform as hoped | Clear go/no-go criteria, pivot options |
| Security/compliance delay | Review takes longer than expected | Engage early, don't surprise security team |
Every AI project has five phases, regardless of size:
Phase 1: Foundation (10-15% of effort)
Phase 2: Data & Infrastructure (20-30% of effort)
Phase 3: Core Development (25-35% of effort)
Phase 4: Validation & Tuning (15-20% of effort)
Phase 5: Deployment & Handoff (10-15% of effort)
A complete delivery plan includes:
1. Executive Summary
2. Solution Overview
3. Phased Timeline
4. Staffing Plan
5. Deliverables Matrix
6. Success Metrics
7. Risk Management
8. Assumptions & Dependencies
AI PROJECT SCOPE: [Project Name]
SUMMARY
- Duration: [X weeks]
- Investment: [Hours or $]
- Team: [Roles]
- Deliverable: [One sentence]
WHAT WE'RE BUILDING
[2-3 sentence description]
WHAT WE'RE NOT BUILDING
- [Explicit exclusion 1]
- [Explicit exclusion 2]
PHASES
Phase 1: [Name] - Week [X-Y]
└─> Deliverables: [list]
Phase 2: [Name] - Week [X-Y]
└─> Deliverables: [list]
[etc.]
SUCCESS LOOKS LIKE
- [Metric 1]: [Target]
- [Metric 2]: [Target]
KEY RISKS
- [Risk 1]: [Mitigation]
- [Risk 2]: [Mitigation]
ASSUMPTIONS
- [What must be true]
DEPENDENCIES
- [What we need from others]
Over time, track your estimates vs. actuals:
| Project | Estimated | Actual | Variance | Cause |
|---|---|---|---|---|
| Inquiry Triage | 240 hrs | 310 hrs | +29% | Data quality issues |
| Review Analysis | 160 hrs | 145 hrs | -9% | API easier than expected |
| Document Search | 120 hrs | 280 hrs | +133% | Scope creep |
Use this data to:
You're asked to scope an AI project to help your merchandising team analyze customer reviews to identify product improvement opportunities.
Requirements:
Create your project scope:
1. Decompose the Work
2. Estimate Each Component
3. Define Phases and Timeline
4. Identify Staffing Needs
5. List Assumptions and Risks
6. Define Success Metrics
In Module 5.4: Stakeholder Communication, you'll learn how to communicate progress, manage expectations, and handle difficult conversations when AI projects hit inevitable bumps.
You've reached the end of this module. Review the checklist below to ensure you've understood the key concepts.