Documents, Data, and Artifacts

Everything before this module was text-in, text-out. This module introduces the work that occupies most business hours: files. CSVs, contracts, handbooks, financial data. Claude reads uploaded files, analyzes their contents, compares documents, finds patterns in data, and produces formatted outputs. For participants who think of Claude as a writing assistant, this is the module that recategorizes it as a business analyst.

The productivity claim becomes concrete here. Analyzing a messy 50-row CSV by hand takes an hour. Claude does it in under a minute. Reading a 20-page contract for dollar amounts takes 45 minutes. Claude does it in 30 seconds. The savings are not marginal. They are order-of-magnitude.

The CSV contains deliberate problems that Claude should catch. Blank units_sold fields in several rows. A deal marked as “Pipeline” status mixed in with “Closed Won” records, which would inflate revenue if included. A customer listed as “Unknown Customer” with no sales rep assigned. A row with a discounted price well below the standard price, suggesting a negotiated deal that may not be representative. Missing sales rep names on two records.

The exercise file lists all expected findings. Key metrics: approximately $850K total revenue (exact figure depends on whether the Pipeline deal and rows with missing data are included or excluded). The interesting trends: Enterprise segment outperforming SMB, one rep (Johnson) carrying a disproportionate share, Q4 seasonal pattern visible in deal timing.

Know these numbers before the demo. When Claude surfaces them, confirm aloud: “That is correct. The data has five quality issues, and Claude found all of them because we asked it to look.”

The employee handbook excerpt contains several policies that a careful HR attorney would flag. The attendance policy states that three consecutive days of no-call, no-show constitute “voluntary resignation.” This is legally risky in many jurisdictions because it can be interpreted as constructive termination without proper notice or process. The jury duty policy requires employees to “remit” (return) any compensation received from the court, which some states prohibit. The electronic communications policy grants broad monitoring authority without specifying limits, which may conflict with state privacy laws. The drug testing policy references “reasonable suspicion” testing without defining the criteria for suspicion, opening the door to inconsistent application.

You do not need to be an employment lawyer to teach this demo. You need to know that Claude will flag these issues and that the flags are legitimate. If a participant who works in HR challenges a specific finding, acknowledge it: “The specifics depend on your state’s laws. The point is that Claude identified policy language worth reviewing with counsel. That is the analysis skill.”

The financial CSV shows six rows: Q4 2024 through Q3 2025 (actuals), Q4 2025 (actual), and Q4 2025 (budget). The story: Q4 2025 missed the revenue budget by approximately $600K. Gross margin compressed from a budgeted 42% to actual 37.6%. Headcount came in at 192 versus a budgeted 202, suggesting either a hiring freeze or attrition. Operating expenses ran close to budget despite the revenue miss, meaning the company did not cut costs to match the shortfall.

Claude should connect these data points into a narrative: “Revenue missed plan by X. Margin compressed because Y. Headcount ran below plan, which may have contributed to the revenue miss through reduced capacity.” This is the kind of analysis a FP&A analyst produces for CFO reviews. The prompt should specify the audience (CFO), the format (variance narrative), and the scope (explain the Q4 miss vs. budget).