One of the most common requests we hear from business users: "I have a data file. I need a report. I have two hours." This is exactly the problem our Data Analysis Agent was built to solve.

In this tutorial, we'll walk through a real example: taking a messy CSV of sales data through all four pipeline stages and producing a polished executive report — complete with insights, anomalies, recommendations, and a downloadable PDF.

The Input: A Real Sales Dataset

For this example, we used a 2,400-row CSV containing: monthly sales by region, product category, sales rep, channel (online/offline), and customer segment. Typical mid-market company data — not clean, not terrible.

Stage 1 — Data Profiling
🔍 What Does This Data Look Like?
The agent profiles the dataset: row count, column types, missing value percentages, data ranges, and unique value distributions. It also identifies data quality issues — in our example, 3.2% of revenue values were null and two region columns had inconsistent naming.
Stage 2 — Statistical Analysis
📊 What Do the Numbers Say?
Descriptive statistics, correlation analysis, distribution analysis, and time-series decomposition (if applicable). For our sales data: average order value by channel, revenue concentration by customer segment, and month-over-month growth rates.
Stage 3 — Pattern Detection
🔬 What's Interesting?
The agent identifies anomalies, outliers, trends, and unexpected patterns. In our example: Q3 showed anomalously high performance in the Western region (later traced to a product launch), and the enterprise segment showed 3x higher average order value but 70% lower purchase frequency.
Stage 4 — Executive Report
📋 What Should Leadership Know?
A structured executive report with: business performance summary, 5 key insights with supporting data, 3 anomalies requiring attention, and strategic recommendations ranked by estimated impact. Downloadable as PDF, Excel, or CSV.

The Output: What We Got

The full pipeline took under 3 minutes. The output included:

  • Executive summary (5 paragraphs, board-ready language)
  • Revenue by segment with growth trajectory analysis
  • Channel performance comparison with ROI estimates
  • 3 flagged anomalies with hypotheses for each
  • 4 strategic recommendations ranked by revenue impact

The PDF download was 12 pages. A human analyst doing the same work would typically produce a similar output in 2-3 days.

Tips for Getting the Best Output

  • Clean column headers first. Replace spaces with underscores, remove special characters. The agent handles messy data, but clean headers produce better insights.
  • Include a date column. Time-series patterns are often the most valuable insights — the agent can only detect them if a date field exists.
  • Add context in your prompt. "This is monthly sales data for a B2B SaaS company" gives the agent domain context that significantly improves recommendation quality.
📊 Try the Data Analysis Agent

Upload any CSV and get a structured executive report in minutes. Available to Pro and Business subscribers. First run is free for all users.