Overview of the demo
In a clear walkthrough, Teacher's Tech demonstrates how an AI agent built with Claude Code turns raw spreadsheets into a business analysis in about ten minutes. The video uses a fictional company, Basecamp Brew Co., and provides the exact sales and customer files so viewers can follow along step by step. As a result, viewers can see how to set up a project, run the agent, and iterate on reports without writing formulas or building pivot tables themselves.
Moreover, the creator timestamps each stage of the process, which helps users jump to sections like setting up VS Code, creating a CLAUDE.md file, and reviewing the sales and customer reports. The sequence emphasizes building a reusable workflow, then applying it to the 12 months of sales data and 200 customer records. Consequently, the demo highlights both practical setup details and the agent’s actual outputs so viewers can reproduce the steps.
Setting up the agentic workflow
First, the video shows how to initialize the project in VS Code and add the three core files: sales data, customer data, and the prompt file named CLAUDE.md. The host explains the structure of CLAUDE.md, which contains the agent’s goals and the reusable instructions that guide the workflow each time it encounters a new data file. This design lets the agent ask clarifying questions, check data quality, and then perform analysis consistently across datasets.
Next, Teacher's Tech demonstrates using a “plan mode” and writing a workflow prompt so the agent knows how to break tasks into subtasks and when to seek more input. The video emphasizes clear prompts that tell the agent what to look for, such as trends, anomalies, and specific recommendations. In doing so, the tutorial helps viewers build a pattern they can reuse for other business problems without rebuilding the logic each time.
How the agent analyzes data
During execution, the agent inspects the spreadsheets for issues, summarizes the data, and then performs deeper analysis to uncover trends and flags. It runs the sales analysis first, focusing on profit as requested, and then moves to the customer dataset to examine acquisition channels and repeat purchase behavior. Importantly, the agent carries out these steps without manual formulas or pivot tables, relying on its workflow to generate formatted tables and narrative summaries.
In addition, the agent responds to the user’s follow-up answers. For example, when prompted to prioritize profit and to watch a suspected underperforming kiosk, the agent narrows the analysis and highlights channel-level performance. When asked to lead with spend and frequency for customer analysis, it digs into repeat versus one-time buyers and attempts to cross-reference joinable fields with the sales data, while flagging any limitations in the joins.
Key findings and recommendations
The video’s guided prompts shape the report: it leads with profit metrics, provides channel comparisons, and ends with clear recommendations tailored to the business. The host asks the agent to include actual numbers and formatted tables, and the output includes both high-level trends and specific next steps. As a result, viewers can see how the agent transforms plain data into actionable advice that a small business might use to plan the next quarter.
Moreover, the agent follows a final section that lists the top three actions the business should take next, which aligns with the initial prompt requirements. The recommendations reflect tradeoffs in focus: for instance, whether to reallocate marketing spend based on short-term revenue or long-term customer value, and whether to investigate underperforming channels before cutting them. This framing helps decision-makers weigh immediate gains against strategic investments in customer retention.
Tradeoffs and practical challenges
However, the video also surfaces important tradeoffs. Automation speeds analysis and lowers the barrier to entry, but it depends on clean, joinable data and well-crafted prompts; otherwise, the agent may miss context or flag misleading patterns. Consequently, businesses must pair agent outputs with human review to validate assumptions and to investigate anomalies that require domain knowledge.
Another challenge is iteration: while the agent produces a useful first pass, refining the prompts and rerunning analyses improves accuracy and relevance, which takes time and testing. In addition, cross-referencing datasets can expose limitations when fields do not join cleanly, prompting careful data preparation or conservative interpretation of any linked insights. Thus, teams should plan for a short cycle of validation and adjustment after the agent’s initial report.
Why reusable workflows matter
Finally, the video underlines the value of building reusable agentic workflows: once you create a robust prompt and file structure, you can apply it to multiple datasets and save time on future analyses. That reusability supports consistency across reports and helps teams scale their analytic capacity without hiring extra specialists. Therefore, the approach suits small businesses and teams that need repeatable, fast insights from routine data.
In conclusion, Teacher's Tech presents a practical, repeatable path to use Claude Code as a business analyst, and the hands-on demo makes clear both the potential and the limits of agentic analysis. Viewers who follow the steps can quickly test the technique on their own files, while keeping in mind the need for data checks and human oversight to turn automated outputs into reliable business decisions.
