project-cost

Project Cost Estimation

Estimate total API costs for multi-step projects and recurring workflows.

Gathering requirements

Ask the user about:

1. Task type: What will Claude do? (writing, analysis, coding, classification, etc.)
2. Volume: How many items/documents/requests?
3. Input size: Average size of each input (words, pages, or tokens)
4. Output size: Expected output length per request
5. Frequency: One-time or recurring? If recurring, how often?
6. Model preference: Which model tier? (default to Sonnet if unsure)

Cost calculation method

For each step in the workflow:

1. Estimate input tokens (content + system prompt + conversation history)
2. Estimate output tokens (expected response length)
3. Apply model pricing
4. Multiply by volume
5. Add buffer of 20% for retries, errors, and prompt iteration

Common workflow templates

Document processing pipeline

• Input: document tokens + system prompt (~500 tokens)
• Output: typically 20-40% of input length for summaries, 5-10% for extraction
• Factor: number of documents

Content generation

• Input: brief/instructions (~1000 tokens) + reference material
• Output: 500-3000 tokens per piece typically
• Factor: number of pieces

Classification/triage

• Input: item text + classification criteria (~800 tokens)
• Output: ~200 tokens per classification
• Factor: number of items (often high volume, use Haiku)

Code generation

• Input: spec + context (~2000 tokens)
• Output: 500-5000 tokens per component
• Factor: number of components

Output format

Present as a cost breakdown table:

Step	Model	Requests	Input tok/req	Output tok/req	Cost
...	...	...	...	...	...
Total					$X.XX

Then add:

• Monthly projection: if recurring
• Annual projection: if recurring
• Cost optimization tips: specific to their workflow
• Break-even analysis: vs manual labor cost if relevant