Analysis & Reasoning with Claude
Claude is a capable reasoning partner for structured analysis, logic problems, decision frameworks, and scenario evaluation. Getting reliable output from these tasks requires understanding both where Claude is strong and where you need to apply additional scrutiny.
Quantitative Analysis
Claude can reason about numbers and perform moderate arithmetic, but it is not a calculator or data analysis tool in the strict sense:
Where Claude is reliable
- Interpreting quantitative results (what does this number mean in context?)
- Setting up the right analytical framework for a problem
- Writing code (Python, R, SQL) that performs calculations on your data
- Identifying the right statistical approach for a given question
- Explaining statistical concepts and results clearly
Where Claude makes errors
- Multi-step arithmetic with large numbers (can make calculation errors)
- Complex probability calculations done in prose (not code)
- Financial modelling without executing actual code
- Anything requiring precise numerical precision at scale
Best practice: For quantitative analysis, ask Claude to write code that does the calculation, then run the code. Don't ask Claude to compute large arithmetic in prose — the error rate is too high for anything beyond simple arithmetic.
Structured Reasoning Prompts
For complex reasoning tasks, structuring the prompt explicitly improves output quality significantly:
- Step-by-step breakdown: "Think through this step by step before giving me your answer." This forces Claude to reason through the problem before committing to a conclusion.
- Enumerate assumptions: "Before answering, state the key assumptions you're making." This surfaces hidden assumptions that may be wrong.
- Separate reasoning from conclusion: "First explain your reasoning in full, then give me your conclusion in a single sentence."
- Consider counterarguments: "What's the strongest argument against your conclusion?" Asking Claude to steelman the opposite position improves balanced analysis.
Logic Problems and Scenario Analysis
Claude handles formal logic problems, deductive reasoning puzzles, and scenario analysis reliably when:
- The problem is unambiguously stated — ambiguous premises lead to multiple valid interpretations and inconsistent answers
- You ask Claude to show its work — multi-step logic is more reliable when each step is explicit
- The problem scope is defined — "assume X, Y, Z are the only relevant factors" prevents Claude from importing outside considerations
For scenario analysis, give Claude the specific scenarios to evaluate: "Evaluate these 3 scenarios against criteria A, B, and C: [scenarios]. Be explicit about which criteria each scenario satisfies."
Decision Analysis
Claude is a useful thinking partner for structured decision-making. Frameworks that work well:
Pros/cons analysis
"Give me a structured pros/cons for each option. Then weight the pros and cons by: [your criteria]. What does the analysis suggest?"
Weighted decision matrix
"I'm evaluating [options] against criteria: [list with weights]. Score each option 1–5 on each criterion and compute weighted totals. Show your reasoning for each score."
Pre-mortem analysis
"Assume it's 12 months from now and this decision turned out badly. What went wrong? Now assume it went well — what made it succeed?" This surfaces risks and success factors before committing.
When to Use Chain-of-Thought Prompting
Chain-of-thought (CoT) prompting — asking Claude to reason through a problem before answering — is most valuable when:
- The task involves multiple reasoning steps that build on each other
- Errors in early steps would compound into a wrong final answer
- You need to verify Claude's reasoning, not just trust the conclusion
- The problem has multiple valid interpretations that need to be resolved before answering
Use the phrase "Think step by step" or "Let's work through this together — first [step 1], then [step 2]..." to trigger more deliberate reasoning. For highly complex problems, Claude's extended thinking mode (available on Sonnet/Opus via API) performs even deeper deliberation before responding.
Checklist: Do You Understand This?
- For quantitative analysis, ask Claude to write code to do the calculation — don't ask it to compute large arithmetic in prose
- Structured reasoning improves with: step-by-step instructions, explicit assumption listing, and asking for counterarguments
- State logic problems unambiguously and ask Claude to show its work at each step
- Use pros/cons, weighted matrices, and pre-mortem patterns for decision analysis
- Chain-of-thought prompting ("think step by step") is most useful for multi-step reasoning tasks where early errors compound