slice icon Context Slice

Analysis Output Principles

Apply these guidelines to all data analysis outputs regardless of domain.

Data Presentation

  1. Always show the data — Include actual tables with numbers, not just prose summaries
  2. Calculate rates and percentages — Raw counts are less useful than conversion rates, percentages, and ratios
  3. Show distributions — Averages hide important variation; include ranges or breakdowns
  4. Identify outliers — Call out unusual data points that might skew results or warrant investigation

Actionability

  1. Be specific about recommendations — "Reach out to Acme Corp" not "improve retention"
  2. Quantify impact — "5 customers at $50K ARR each" not "some customers at risk"
  3. Prioritize actions — If recommending multiple things, rank by impact or urgency
  4. Connect to decisions — Frame findings in terms of decisions the user needs to make

Data Quality & Limitations

  1. Flag data quality issues — Missing columns, sparse data, suspicious values, inconsistent formats
  2. Acknowledge sample size — If data is sparse (< 20 rows), warn that results may not be statistically significant
  3. Note parsing assumptions — If dates, currencies, or formats required interpretation, say what you assumed
  4. State what's missing — What the data can't tell you; what additional data would improve the analysis

Handling CSV Data

Column Detection

  • If a required column is missing, ask the user which column to use
  • Look for common aliases (e.g., "Status" / "Stage" / "Phase" for pipeline stage)
  • Note which columns were used for each analysis dimension

Data Validation

  • Check for obvious data issues: negative amounts, dates in the future, impossible percentages
  • Count rows with missing values in key columns
  • Identify potential duplicates if there's an ID column

Date Handling

  • Default to ISO format (YYYY-MM-DD) when ambiguous
  • Note timezone assumptions if timestamps are present
  • Handle partial dates gracefully (e.g., "Jan 2024" → first day of month)

Output Structure

Every analysis should include:

  1. Summary — Key metrics at a glance
  2. Data tables — Actual numbers in tabular form
  3. Insights — What the data reveals (patterns, outliers, trends)
  4. Recommendations — 2-3 specific actions based on findings
  5. Limitations — Data quality notes and what wasn't captured