How We Plan and Estimate

Estimation is hard. AI makes us faster, but not infinitely fast. This guide covers how we plan realistic timelines.

Core principle: Plan in outcomes, not hours. "Login works" is better than "8 hours on auth."

Why Estimation Matters

• Sets expectations — stakeholders know when to expect results
• Reveals scope — breaking down work exposes hidden complexity
• Enables prioritization — can't fit everything? Cut the least important
• Tracks progress — milestones show if we're on track

The Planning Process

PRD → Break into milestones → Define tasks → Estimate → Review → Commit

Step 1: Break Into Milestones

Milestones are meaningful checkpoints. Each should deliver something usable.

Good Milestones

Bad Milestones

Example: Keyword Clustering Tool

Week 1 Milestones:
├── M1: User can upload CSV and see keywords parsed
├── M2: Clustering runs (even if slow)
└── M3: Results display with cluster labels

Week 2 Milestones:
├── M4: Progress indicator during clustering
├── M5: Export results to CSV
└── M6: Deployed to Railway

Step 2: Define Tasks

Each milestone breaks into tasks. Tasks should be completable in a few hours.

Task Format

- [ ] [Component/Area] Task description
      Estimate: S/M/L
      Dependencies: None | Task X

Size Guide

Example: Milestone Breakdown

M1: User can upload CSV and see keywords parsed

- [ ] [UI] Create upload component with drag-drop
      Size: M
      Dependencies: None

- [ ] [API] Create /api/upload endpoint
      Size: S
      Dependencies: None

- [ ] [Logic] Parse CSV, extract keywords
      Size: S
      Dependencies: API endpoint

- [ ] [DB] Save keywords to Supabase
      Size: S
      Dependencies: Parse logic

- [ ] [UI] Display parsed keywords in table
      Size: M
      Dependencies: DB save works

Step 3: Estimate Realistically

The Multiplier Rule

What Eats Time

AI Doesn't Eliminate All Work

Claude Code accelerates coding, but you still need time for:

• Thinking about the right approach
• Reviewing and testing AI output
• Fixing edge cases AI missed
• Integration and deployment
• Communication and coordination

Rule of thumb: AI makes you 2-3x faster, not 10x.

Step 4: Create the Timeline

Timeline Template

## [Project Name] Timeline

### Week 1: [Theme]
**Goal**: [What's working by end of week]

Monday-Tuesday:
- [ ] Task 1
- [ ] Task 2

Wednesday-Thursday:
- [ ] Task 3
- [ ] Task 4

Friday:
- [ ] Task 5
- [ ] Buffer for overflow

**Milestone Check**: [What should work]

### Week 2: [Theme]
...

Example: AI Overviews Checker Timeline

## AI Overviews Checker Timeline

### Week 1: Core Functionality
**Goal**: Can check keywords and see results

Monday-Tuesday:
- [ ] Project setup, Supabase schema
- [ ] DataForSEO API client
- [ ] Keyword input form

Wednesday-Thursday:
- [ ] Batch keyword checking logic
- [ ] Results storage and display
- [ ] Basic error handling

Friday:
- [ ] Progress indicator
- [ ] Buffer

**Milestone Check**: Enter keywords, see AI Overview results

### Week 2: Polish & Deploy
**Goal**: Production-ready

Monday-Tuesday:
- [ ] Export functionality
- [ ] Rate limit handling
- [ ] Better error messages

Wednesday-Thursday:
- [ ] Loading states, empty states
- [ ] Mobile responsiveness
- [ ] Railway deployment

Friday:
- [ ] Documentation
- [ ] Team demo
- [ ] Handoff

**Milestone Check**: Live on production, team can use it

Managing Scope

When You're Behind

Priority 1: Cut scope      → Remove "nice to have" features
Priority 2: Reduce quality → Ship rougher edges, polish later
Priority 3: Extend time    → Only if stakeholder agrees

The MoSCoW Method

Scope Negotiation

When someone asks for more:

"We can add [feature], but that means either:
A) Pushing the deadline by X days, or
B) Removing [other feature] from this release

Which would you prefer?"

Never just say yes and absorb the extra work silently.

Common Mistakes

Estimating in "perfect conditions"

Signs: "If I just focus, I can do this in 2 days" (never happens).

Fix: Add buffer for meetings, context switching, bugs, and learning curves.

Not tracking against estimates

Signs: You estimate at the start, never look at it again.

Fix: Check progress against milestones weekly. Adjust early, not late.

Treating estimates as commitments

Signs: Stress when estimates prove wrong, hiding delays.

Fix: Estimates are guesses. Communicate when reality differs. Adjust plans.

Padding everything instead of being honest

Signs: Every task is "Large" because you're scared to commit.

Fix: Estimate honestly, then apply project-level buffer. Transparency builds trust.

Evaluation Checklist

Your planning is working if:

• Milestones are clear and deliverable
• Tasks are small enough to complete in a day
• You know what "done" looks like
• Stakeholders understand the timeline
• You're checking progress weekly
• Scope changes are negotiated, not absorbed

Your planning needs work if:

• You're always "almost done"
• Milestones keep moving
• Tasks turn out to be 3x larger than expected
• Nobody knows when things will be ready
• You're surprised by how much is left

Quick Reference

Estimation Checklist

Before committing to a timeline:

• Broken into milestones with deliverables?
• Tasks sized (S/M/L) with dependencies?
• Buffer added for unknowns?
• Stakeholder aligned on scope and timeline?
• Plan documented and shared?

Time Sinks to Remember

When Scope Creeps

1. Document the new request
2. Estimate the impact
3. Present trade-offs to stakeholder
4. Get explicit decision
5. Adjust plan accordingly