Section 10.4.4 : Data analysis and reporting with AI
🎯 Learning objective
Using AI to analyze datasets, create automated reports, and make data-driven decisions — without being a data scientist, without writing a single line of code, and reducing analysis time by 80%.
AI democratizes data analysis
Before generative AI, analyzing data required skills in SQL, Python/R, or expensive BI tools like Tableau or Power BI. "Citizen analysts" depended on data teams drowning in requests. The result: 73% of data collected by companies is never analyzed (Forrester, 2024).
Today, AI radically changes this landscape:
- →ChatGPT Advanced Data Analysis: upload a CSV, ask questions in plain language, get charts
- →Claude: analyze 500-page documents in one go
- →Google Gemini: connect directly to Google Sheets and BigQuery
- →Microsoft Copilot: analyze in Excel without complex formulas
The result: any professional can now go from raw data to actionable insights in 15 minutes instead of 3 hours.
Complete data analysis workflow with AI
Step 1 — Upload and initial exploration
The first step is always understanding the dataset. Never dive into analysis without this exploratory phase:
[Upload your CSV/Excel file]
Analyze this data file in depth:
1. STRUCTURE:
- Number of rows and columns
- Name, type and description of each column
- Example of first 5 rows
2. QUALITY:
- Missing values per column (count and percentage)
- Detected outliers
- Potential duplicates
- Inconsistencies (future dates, illogical negative values, etc.)
3. DESCRIPTIVE STATISTICS:
- Mean, median, standard deviation for numeric columns
- Frequency distribution for categorical columns
- Time range if dates present
4. RECOMMENDATIONS:
- 5 most relevant analysis questions to explore
- Recommended data cleaning before analysis
- Most promising cross-analyses
Step 2 — Data cleaning
Raw data is rarely clean. Ask AI to clean before analyzing:
Clean this dataset with the following rules:
1. Remove rows with more than 50% missing values
2. For remaining missing values:
- Numeric columns → replace with group median
- Categorical columns → replace with "Not specified"
3. Standardize formats:
- Dates → YYYY-MM-DD format
- Amounts → numbers without $ symbol, decimal separator as period
- Names → title case, removing extra spaces
4. Flag outliers (values > 3 standard deviations) without deleting them
5. Confirm row count before/after cleaning
Generate the cleaned file and a cleaning report summarizing modifications.
Step 3 — Targeted question-based analysis
Ask precise questions rather than "analyze this data":
From the uploaded data, answer these analysis questions:
1. TREND: What is the main trend over the last 12 months?
→ Create a line chart with regression and confidence interval
2. SEGMENTATION: Which segments perform best / worst?
→ Create a horizontal bar chart sorted by performance
→ For each segment, calculate: revenue, growth, market share
3. CORRELATION: Are there notable correlations between [variable A] and [variable B]?
→ Create a scatter plot with regression line
→ Give the correlation coefficient (Pearson) and its interpretation
4. ANOMALY: Are there any anomalies or inflection points in the data?
→ Identify trend changes with exact dates
→ Propose explanatory hypotheses
5. PREDICTION: Based on the trend, what is the projection for the next 3 months?
→ With confidence interval and explicit assumptions
For each answer, provide the key figure, the chart, and an actionable insight in one sentence.
Step 4 — Automated executive report
The final report must be directly presentable to management:
From all the analysis performed, write an executive report for the leadership team.
Mandatory structure:
1. Executive Summary (3 lines — recommendation first, pyramid principle)
2. Key KPIs (table: indicator | current value | change vs Y-1 | target | status 🟢🟡🔴)
3. Main trends (2-3 charts with one-line commentary each)
4. Deep analysis of critical points (max 2 sections)
5. Identified risks (table: risk | probability | impact | mitigation)
6. Actionable recommendations (max 3, with suggested owner and deadline)
7. Appendix: methodology and analysis limitations
Style: concise, factual, decision-oriented. No statistical jargon.
Key figures in bold. Positive changes in green, negative in red.
Each page should be readable independently (standalone).
Detailed use cases by function
| Role | Typical data | AI question | Business result | Time saved |
|---|---|---|---|---|
| Marketing | GA4 export, campaign data | "Which channels have the best ROI by segment?" | Ad budget reallocation | 3h → 30min |
| Sales | CRM export (Salesforce, HubSpot) | "Which lead profile converts best? What's the average sales cycle by segment?" | Optimized lead scoring, pipeline prioritization | 4h → 45min |
| HR | Satisfaction survey, turnover data | "Which factors correlate with attrition? Which profiles are at risk of leaving?" | Targeted retention plan | 2 days → 2h |
| Finance | Quarterly P&L, balance sheet | "Which line items are drifting vs. budget? What's the year-end projection?" | Preventive alerts, revised forecasts | 1 day → 1h |
| Operations | Production data, quality | "Where are the bottlenecks? What's the defect rate by line?" | Capacity optimization | 3h → 30min |
| Supply Chain | Inventory, orders, lead times | "Which products risk stockout? What's the optimal inventory level?" | 40% reduction in stockouts | 4h → 1h |
Claude for long document analysis
For 50+ page reports (annual report, market study, audit, contracts, RFP responses), Claude with its 200K token window (~500 pages) is the reference tool:
Annual report analysis:
[Upload the PDF]
1. Extract the 10 most important figures from this annual report
2. Compare 2024 vs 2025 results: identify the 5 most significant gaps
3. Identify mentioned risks: rank them by severity
4. What are the 3 promises made to shareholders? Are they credible given the figures?
5. One-page summary: what an investor needs to know
Market study analysis:
[Upload the PDF]
1. What is the identified TAM/SAM/SOM? How is it calculated?
2. Who are the 5 key players and their positioning?
3. What emerging trends are identified?
4. Which market segments are underexploited?
5. What data is missing or questionable in this study?
Contract analysis:
[Upload the PDF]
1. Identify the 5 most unfavorable clauses for us
2. What are the exit / termination conditions?
3. Are there any automatic renewal clauses?
4. What quantified commitments are we making?
5. Which points should we prioritize for renegotiation?
AI analysis pitfalls to avoid
| Pitfall | Risk | How to avoid it |
|---|---|---|
| Correlation ≠ Causation | AI finds a correlation and you conclude causation | Always ask: "Could this correlation be due to a third factor?" |
| Survivorship bias | Analyzing only current customers (not churned ones) | Include churned customer data in the analysis |
| Pattern hallucination | AI "sees" trends in statistical noise | Ask for confidence interval and statistical significance |
| Non-representative data | Drawing general conclusions from a small sample | Ask: "Is this sample sufficient to conclude? What's the margin of error?" |
| Confirmation bias | Asking leading questions that confirm your hypothesis | Ask the AI: "What evidence contradicts this hypothesis?" |
| False precision | Numbers with 5 decimal places aren't more accurate | Round, question sources, cross-reference with other sources |
Sensitive data and confidentiality — Strict rules
Never share personal data (GDPR), confidential, or proprietary data with public LLMs. Rules to follow:
- →Personal data (names, emails, addresses): anonymize before upload (replace with anonymous IDs)
- →Financial data: use enterprise versions (ChatGPT Enterprise, Claude for Business) that guarantee your data isn't used for training
- →Strategic data: verify the vendor's retention policy — some delete your data after 30 days, others don't
- →When in doubt: create a synthetic dataset with the same structure but fictitious values to test your analyses before using real data
Practical exercise: your first complete AI analysis
Duration: 45 minutes
- →Find a dataset (data.gov, Kaggle, or an export of your own non-sensitive data)
- →Upload into ChatGPT and follow steps 1 through 4 above
- →Ask 3 specific analysis questions (not "analyze everything")
- →Request 2 charts (one trend chart, one comparison chart)
- →Generate a 1-page executive report
- →Verify: are the figures consistent? Are the conclusions logical? Are the recommendations actionable?
Quick Check
What is the best approach for analyzing a 100-page PDF report with AI?
Quick Check
What is the most dangerous pitfall of data analysis with AI?