"Will ChatGPT really make a proper WBS if I ask it to?"
Surprisingly, it makes it much better than expected.
And in just 10 minutes, it creates a draft that systematically breaks down hundreds of tasks.
Of course, AI-generated WBS is a "draft."
It only becomes a perfect WBS when human professional verification and improvement are added.
But AI's role as a starting point is beyond imagination.
Today, I'll share technical implementation methods and core prompt engineering strategies for efficiently generating WBS using LLM (Large Language Model).
Why is LLM Good at Making WBS?
There are reasons LLM is particularly strong at WBS generation.
First, it grasps the entire context at once.
It can see the big picture of the project while thinking about detailed tasks.
Second, hierarchical thinking is natural.
Creating tree structures is one of language models' basic abilities.
Third, it automatically recognizes dependencies.
It knows "API development is only possible after database design is done."
Fourth, it maintains consistency.
It writes hundreds of tasks in the same format and detail level.
Above all, it knows patterns because it learned from numerous project documents.
Practice: Generating E-commerce WBS
Step 1: Basic Prompt
Please create a WBS for the following project:
Project: E-commerce Platform Development
Duration: 3 months
Team Size: 5 people
Key Features: Product Management, Shopping Cart, Payment, Delivery Tracking
Even this much produces a basic WBS.
But depth may be insufficient and may not fit our situation.
Step 2: Advanced Prompt (Key!)
Generate a WBS for the e-commerce platform development project.
Constraints:
- 3-level depth (Level 1 > Level 2 > Level 3)
- Each task broken down to 8 hours or less
- Include development/test/deployment stages
- Include time estimates (optimistic/realistic/pessimistic)
Output Format:
1. [Category] (Total Time)
1.1 [Subcategory] (Time)
1.1.1 [Task] (O:Optimistic/R:Realistic/P:Pessimistic)
Tech Stack:
- Frontend: React, TypeScript
- Backend: Node.js, PostgreSQL
- Infrastructure: AWS, Docker
Requesting this specifically completely changes the quality.
Step 3: ChatGPT Response Example
AI responds like this:
1. Project Planning and Design (120h)
1.1 Requirements Analysis (32h)
1.1.1 Stakeholder Interviews (O:4h/R:6h/P:8h)
1.1.2 Functional Specification Writing (O:6h/R:8h/P:12h)
1.1.3 Non-functional Requirements Definition (O:3h/R:4h/P:6h)
1.2 System Design (40h)
1.2.1 Architecture Design (O:6h/R:8h/P:12h)
1.2.2 DB Schema Design (O:4h/R:6h/P:8h)
1.2.3 API Design (O:6h/R:8h/P:10h)
2. Development (480h)
2.1 Backend Development (200h)
2.1.1 Auth/Authorization System (O:16h/R:24h/P:32h)
2.1.2 Product Management API (O:20h/R:28h/P:40h)
...
Such a structured WBS appears in just 30 seconds!
Prompt Engineering Tips
1. Make Context Clear
prompt = f"""
Project Type: {project_type}
Tech Stack: {tech_stack}
Team Size: {team_size}
Constraints: {constraints}
Output Format: {output_format}
"""
More specific = better results.
2. Provide Few-shot Examples
Example:
1. Frontend Development (100h)
1.1 Component Development (40h)
1.1.1 Header Component (O:2h/R:3h/P:4h)
Please write the entire WBS in this format.
Showing examples makes AI follow accurately.
3. Iterative Improvement
Don't expect perfect WBS in one go.
First: "Create WBS" → Basic structure
Second: "Add test tasks" → Add detailed tasks
Third: "Apply PERT estimation" → Time estimation
Improve conversationally like this.
Automating with Python
import openai
class WBSGenerator:
def __init__(self, api_key):
openai.api_key = api_key
def generate_wbs(self, project_desc, constraints=None):
prompt = self._build_prompt(project_desc, constraints)
response = openai.ChatCompletion.create(
model="gpt-4",
messages=[
{"role": "system", "content": "You are a project management expert."},
{"role": "user", "content": prompt}
],
temperature=0.3, # Set low (consistency)
max_tokens=3000
)
return self._parse_response(response)
Setting temperature low produces more consistent results.
Actual Performance Comparison
Writing WBS manually takes 4-6 hours.
Tasks are limited to about 50-80, and consistency is poor.
Most importantly, high chance of missing important tasks.
Using LLM produces 100-200 tasks in 10-15 minutes.
Consistency is high, and chance of omission is low.
Of course, domain specificity or team characteristics are hard to reflect.
But humans can review and adjust this.
Limitations and Compensation of LLM WBS
Limitations
Domain Specificity: May not know unique processes of specific industries.
Latest Technology: May not know technology after 2024.
Team Characteristics: Cannot reflect our team's unique work style.
Compensation Methods
Domain experts review,
Reflect past project data,
Apply team velocity.
AI makes 80%, humans complement 20%.
This is the most efficient method.
Practical Checklist
When generating WBS with LLM:
- [ ] Provide project context clearly
- [ ] Specify tech stack specifically
- [ ] Include output format examples
- [ ] Request 3+ levels depth
- [ ] Specify time estimation method
- [ ] Expert review of generated WBS
- [ ] Adjust to reflect team characteristics
Conclusion: AI is Tool, Judgment is Human
WBS made by ChatGPT is a starting point.
Create a draft quickly, review and adjust with the team.
Make 80% in 10 minutes, team completes remaining 20%.
This is project management in the AI era.
If PMs used to think alone for half a day,
Now you can create a draft in 10 minutes and improve with the team.
Faster, more accurate, more collaborative.
Need systematic WBS management with AI? Check out Plexo.
Top comments (0)