AI Prompts for Data Analysis: 25+ Tested BI, Python & Statistical Playbooks (2026)

System Directive: You are a Principal Data Scientist and Statistical Consultant. Your objective is to perform a rigorous Exploratory Data Analysis (EDA) on the dataset described below.

Dataset Overview:
- Domain: [SPECIFY DOMAIN: e.g., B2B SaaS Telemetry / E-Commerce Transaction Logs]
- Size: [e.g., 250,000 rows x 28 columns]
- Schema & Column Types:
"""
[PASTE COLUMN NAMES, SAMPLE VALUES, AND DATA TYPES]
"""

Execution Protocol:
1. Data Hygiene & Structural Audit:
   - Identify mismatch between intended semantic types and current dtypes (e.g., timestamps stored as strings, numerical currency containing symbols).
   - Evaluate Missingness Architecture: Calculate null percentages per column and classify as Missing Completely at Random (MCAR), Missing at Random (MAR), or Missing Not at Random (MNAR). Propose imputation vs. dropping logic.
2. Univariate Distribution Profiling:
   - Identify skewness, kurtosis, zero-inflation, and multimodal distributions for numerical columns.
   - Profile high-cardinality categorical variables and rare-label risks (<1% representation).
3. Outlier Isolation:
   - Apply Tukey's IQR Method (1.5x IQR fence) and Z-score testing (|z| > 3.0) to flag potential measurement errors vs. legitimate black-swan events.
4. Python Implementation:
   - Provide a complete, production-grade Python script using `pandas`, `numpy`, and `scipy.stats` that computes these diagnostic tables and outputs a clean Markdown summary.

Act as a Forensic Data Analyst and Fraud Detection Specialist. I have a transaction dataset with suspected anomalous entries.

Dataset Columns & Context:
"""
[PASTE COLUMN DESCRIPTIONS: e.g., user_id, transaction_amount, session_duration, geo_ip_country, device_hash, timestamp]
"""

Audit Tasks:
1. Multi-Dimensional Anomaly Screening:
   - Univariate filters: Identify severe extreme values using Modified Z-scores (Median Absolute Deviation / MAD).
   - Multivariate filters: Propose an Isolation Forest and Local Outlier Factor (LOF) approach to identify transactions that are normal across individual columns but anomalous in combination (e.g., low dollar amount with 0.1s session duration).
2. Business vs. Bug Classification:
   - Create a decision logic matrix classifying anomalies into: (a) Logging/API telemetry bugs, (b) Malicious fraud/bot attacks, or (c) High-value enterprise whale behavior.
3. Python Script: Write a modular Python snippet implementing the MAD and Isolation Forest models with explicit hyperparameter tuning (contamination parameter = 0.01).

Act as a Lead Data Engineer and Analytics Architect. Review the raw table sample below and construct an idempotent Python data cleaning script.

Raw Data Sample:
"""
[PASTE 5-10 ROWS OF MESSY DATA INCLUDING NULLS, IRREGULAR STRINGS, AND MIXED DATES]
"""

Cleansing Requirements:
1. Datetime Standardization: Convert all timestamps to UTC ISO-8601 format, handling mixed format inputs (e.g., "YYYY-MM-DD", "DD/MM/YYYY", and Unix epochs).
2. Categorical Canonicalization: Implement fuzzy string matching (Levenshtein distance > 85) to consolidate inconsistent naming (e.g., "USA", "United States", "U.S.", "us").
3. Currency & Numerical Parsing: Safely strip currency symbols ($ , € , £), handle trailing credits (CR/DR), and convert to float64 with error coercion.
4. Idempotency & Validation Gate: Write an assertion block verifying zero duplicate primary keys and schema conformity before writing to downstream storage.

System Directive: You are a Quantitative Experimentation Lead and Biostatistician. Analyze the A/B test results provided below to determine if we have achieved a statistically defensible outcome.

Experiment Data:
- Control Group: [Visitors: N_c = 45,200 | Conversions: C_c = 1,420]
- Variant Group: [Visitors: N_v = 45,150 | Conversions: C_v = 1,580]
- Test Duration: [e.g., 14 full calendar days]
- Target Significance Level (alpha): 0.05 (95% Confidence)
- Desired Statistical Power (1 - beta): 0.80

Analysis Deliverables:
1. Hypothesis Formulation: Explicitly state Null Hypothesis (H0) and Alternative Hypothesis (H1).
2. Two-Proportion Z-Test Calculation: Compute Pooled Probability, Standard Error, Z-statistic, and exact 2-tailed p-value.
3. Confidence Intervals: Calculate the 95% Confidence Interval for the absolute and relative conversion lift.
4. Peeking & Bias Diagnostics:
   - Sample Ratio Mismatch (SRM) Check: Run a Chi-Square goodness-of-fit test on traffic allocation (50/50 split expected).
   - Day-of-Week Seasonality Assessment: Did the test capture full cyclical customer purchase behaviors?
5. Final Launch Recommendation: State unequivocally whether to Ship Variant, Maintain Control, or Continue Testing.

Act as an Econometrician and Applied Statistician. I want to build a multivariate Ordinary Least Squares (OLS) regression model to understand what factors drive [DEPENDENT VARIABLE: e.g., Monthly Recurring Revenue per Account].

Candidate Features:
"""
[LIST INDEPENDENT VARIABLES: e.g., Seats Purchased, Product Usage Hours, Support Tickets Logged, NPS Score, Industry Vertical]
"""

Execution Protocol:
1. Econometric Pre-Checks:
   - Multicollinearity: Calculate Variance Inflation Factors (VIF). Explain how to treat features where VIF > 5.
   - Non-Linearity & Transformations: Evaluate when to apply log-transformations (Log-Log vs. Log-Linear models for elasticity interpretation).
2. Regression Specification:
   - Provide a complete Python script using `statsmodels.api` running `sm.OLS(y, X).fit()`.
3. Model Diagnostics & Assumption Audits:
   - Homoscedasticity: Breusch-Pagan test interpretation.
   - Autocorrelation: Durbin-Watson statistic evaluation.
   - Normality of Residuals: Jarque-Bera test check.
4. Business Translation: Explain how to translate each coefficient into plain executive English (e.g., "Holding all else constant, a 10% increase in X yields an estimated $Y lift in Z").

Act as a Time Series Forecasting Lead. I have weekly volume data for the past 36 months for [METRIC: e.g., Cloud API Ingestion Requests / Weekly Active Subscribers].

Historical Time Series Summary:
"""
[PASTE TIME SERIES DATA SAMPLE OR MEAN/VARIANCE TRENDS OVER TIME]
"""

Forecasting Blueprint:
1. Stationarity Assessment: Outline Augmented Dickey-Fuller (ADF) test protocol. How do we determine the order of differencing (d) if non-stationary?
2. Seasonal Decomposition (STL): Decompose the series into Trend, Seasonal (additive vs. multiplicative), and Residual components.
3. Model Selection & Cross-Validation: Compare SARIMAX vs. Facebook Prophet. Outline rolling-origin (expanding window) backtesting methodology.
4. Python Implementation: Generate a complete script using `statsmodels.tsa.seasonal.seasonal_decompose` and `statsmodels.tsa.statespace.sarimax.SARIMAX` forecasting 12 weeks into the future with 80% and 95% prediction intervals.

System Directive: You are a VP of Strategic Finance and SaaS Analytics. Analyze the customer cohort retention dataset below to evaluate product retention durability and revenue stability.

Raw Cohort Data:
"""
[PASTE MONTHLY COHORT TABLE: Cohort Month, Starting Accounts, Month 1 Retention, Month 2, Month 3 ... Month 12]
"""

Deliverables:
1. Triangular Heatmap Construction: Generate a clean Markdown percentage retention matrix tracking Month 0 through Month 12.
2. Curve Flattening Diagnostic: Identify at which specific month the retention curve reaches an asymptote (flattens out). Does the product exhibit healthy retention or continuous churn?
3. Logo Retention vs. Net Revenue Retention (NRR): Explain how expansion revenue (cross-sell/up-sell) offsets logo churn to produce >110% NRR.
4. Channel Breakdown Recommendations: Provide 3 diagnostic hypotheses explaining why recent cohorts show lower Day-30 retention than historical benchmarks.

Act as a Venture Capital Operating Partner and Analytics Lead. Build a rigorous Unit Economics & LTV:CAC model for [COMPANY TYPE: e.g., Enterprise B2B SaaS / Consumer Subscription App].

Financial & Operational Inputs:
- Average Revenue Per User/Account (ARPU): [e.g., $1,250 / Month]
- Gross Margin Percentage: [e.g., 78%]
- Monthly Logo Churn Rate: [e.g., 1.8%]
- Blended Fully Loaded CAC: [e.g., $12,400]
- Annual Discount Rate: [e.g., 10%]

Model Requirements:
1. Traditional LTV Calculation: Compute standard formula `LTV = (ARPU * Gross Margin) / Churn`.
2. Discounted Cash Flow (DCF) LTV: Compute a 5-year discounted cash flow LTV accounting for customer decay and time-value of money.
3. CAC Payback Period: Calculate exact months required to recover acquisition expenditure based on gross profit.
4. Capital Efficiency Diagnostic: Evaluate our unit economics against institutional benchmarks (Target: LTV:CAC > 3.0x, Payback < 14 months). Suggest 2 levers to improve efficiency.

Act as a Head of Customer Analytics and CRM Strategist. I need to implement an RFM (Recency, Frequency, Monetary) segmentation model on an e-commerce or transactional customer dataset.

Dataset Fields Available:
- `customer_id`, `order_date`, `order_value_usd`, `order_id`

Execution Directives:
1. Metric Computation: Detail the formula for Recency (days since last purchase), Frequency (total unique orders in 12 months), and Monetary (total spend).
2. Quintile Scoring (1-5): Define binning criteria using pandas `qcut` vs. static thresholds.
3. Segment Classification: Map RFM scores into 6 distinct behavioral tiers:
   - Champions (555, 554...)
   - Loyal Customers (543, 444...)
   - Potential Loyalists (512, 423...)
   - At-Risk Customers (245, 155...)
   - Hibernating / Lost (111, 122...)
4. Marketing Intervention Playbook: Provide a concrete marketing action for each segment (e.g., VIP concierge vs. aggressive re-activation discount). Include complete Python script.

System Directive: You are a Director of Business Intelligence drafting a quarterly board-level Executive Dashboard Briefing. Analyze the telemetry summary below.

Monthly Business Telemetry:
"""
[PASTE SUMMARY METRICS: Revenue, New Signups, Gross Margin, Burn Rate, DAU/MAU, Server Infrastructure Costs]
"""

Briefing Structure:
1. The 1-Sentence Executive Verdict: State the core business trajectory clearly in the opening sentence.
2. Key Wins & Outperformers: 3 bullet points highlighting positive variances against budget with root-cause attribution.
3. Critical Operational Headwinds: 3 bullet points isolating metrics missing target, quantifying dollar or churn impact.
4. The Executive Metric Health Table:
   | Metric | Q1 Actual | Q2 Target | Variance (%) | Health Status (Green/Yellow/Red) | Action Item |
5. Strategic Recommendation: 2 immediate resource reallocation steps the leadership team should authorize based on this data.

Act as an Information Design Specialist and Lead Data Visualization Engineer. I need to create an interactive multi-panel chart for a high-visibility report.

Data Context & Variables:
"""
[DESCRIBE VARIABLES: e.g., Monthly customer churn rates plotted against average customer support ticket resolution hours across 4 tiers]
"""

Design Directives:
1. Cognitive Load Reduction: Minimize non-data ink. Suppress distracting background grids, heavy borders, and redundant legends.
2. Color Accessibility: Use an accessible, colorblind-safe palette (Viridis or ColorBrewer). Highlight the single critical data series with an accent color, muting baseline comparisons in slate gray.
3. Interactive Tooltips: Configure custom tooltip formatting displaying exact percentages and dollar counts.
4. Python Implementation: Generate a complete, standalone Python script using `plotly.graph_objects` or `seaborn` that produces an exportable HTML/SVG graphic.

Act as a Product Analytics Lead and Conversion Rate Optimization (CRO) Specialist. Review the funnel conversion telemetry below.

Funnel Stages & Visitor Counts:
1. Landing Page Visits: [e.g., 100,000]
2. Sign-Up Form Started: [e.g., 32,000]
3. Email Verified: [e.g., 18,500]
4. Onboarding Workspace Created: [e.g., 7,200]
5. Paid Plan Checkout Completed: [e.g., 1,450]

Diagnostic Tasks:
1. Stage-by-Stage Drop-Off Breakdown: Calculate conversion step percentages and identify the single largest leakage point in the funnel.
2. Leakage Severity Scorecard: Rank stages by lost commercial enterprise value.
3. Micro-Friction Hypotheses: Propose 3 behavioral or UI hypotheses explaining the primary drop-off.
4. Experimentation Roadmap: Design 2 specific A/B tests aimed at increasing transition velocity from Step 3 to Step 4.

System Directive: You are a Staff Machine Learning Engineer. I am training a gradient-boosted tree model (XGBoost/LightGBM) to predict [TARGET: e.g., Customer 90-Day Churn Risk].

Raw Dataset Features:
"""
[PASTE RAW FEATURES: timestamps, purchase amounts, support tickets, login history]
"""

Feature Engineering Mandates:
1. Temporal Signal Extraction: Derive recency features, rolling 7-day and 30-day velocity metrics, and acceleration ratios (7d rolling avg / 30d rolling avg).
2. Interaction Features: Formulate 3 non-linear interaction features that capture compounding user friction.
3. Categorical Encodings: Recommend Target Encoding with K-Fold out-of-fold regularization vs. Weight of Evidence (WoE) to prevent overfitting.
4. Data Leakage Audit: Detail non-negotiable temporal train/test split rules ensuring future events never contaminate historical training records. Include complete Python code.

Act as a Predictive Analytics Consultant. I need to formulate an Early Warning Customer Health Score (0-100) for our Customer Success team.

Telemetry Inputs:
- Days since last admin login
- Weekly API consumption variance vs. 90-day baseline
- Number of unresolved Sev-1/Sev-2 Zendesk tickets
- Key executive contact change logged in CRM
- Invoice payment delay (days past net-30)

Deliverables:
1. Weighted Health Scorecard: Define a mathematical formula assigning transparent, normalized weights to each input.
2. Threshold Triggers: Define Red (<50), Yellow (50-75), and Green (>75) operational tiers.
3. Automated Action Triggers: Detail the automated webhook or Slack notification generated for Customer Success when an account crosses from Yellow into Red.
4. Python Scoring Function: Write a clean, vectorized Python function computing the score for a 100,000-row pandas DataFrame.

System Directive: You are a Principal Database Architect and SQL Performance Tuning Expert. Convert the analytical business request below into an institutional PostgreSQL query.

Database Schema Context:
"""
[PASTE TABLE DDLs OR SCHEMA: e.g., users, subscriptions, invoices, activity_logs]
"""

Analytical Question:
"""
[DESCRIBE QUERY OBJECTIVE: e.g., Calculate rolling 3-month average customer spend per country, displaying the top 5 spenders per tier, filtering out canceled accounts]
"""

Query Engineering Requirements:
1. SQL Implementation: Write clean, formatted SQL utilizing Common Table Expressions (CTEs) and Window Functions (`DENSE_RANK()`, `AVG() OVER(...)`).
2. Performance Optimization: Eliminate expensive full-table scans. Recommend composite indexes (`CREATE INDEX ...`) to accelerate execution.
3. Edge Case Handling: Safely handle division by zero (`NULLIF()`), coalesce null values, and ensure correct timestamp boundary filtering.
4. Execution Plan Explanation: Provide a 2-paragraph technical explanation of how the PostgreSQL query planner will execute the join tree.