Analysts are finding that generative AI can speed up certain parts of their work significantly — particularly writing up findings and generating initial code for data manipulation. The harder question is knowing when to trust the output and when to verify it independently.
The seminar format
This is a half-day event, not a course. The format is demonstration-heavy: a facilitator works through three realistic analysis scenarios live, using ChatGPT with Code Interpreter, GitHub Copilot, and a locally-run model for comparison. Participants observe, ask questions, and then attempt a short version of each scenario themselves.
Scenarios covered
The three scenarios are: cleaning and summarizing a messy CSV dataset, generating a narrative report from structured financial data, and writing Python functions for exploratory visualization. Each scenario includes at least one deliberate error introduced by the AI, which participants are asked to identify.
Limitations worth knowing
The seminar does not promise that AI will replace analytical judgment. Several examples demonstrate cases where the model produced statistically plausible but incorrect interpretations of the data. These cases are examined carefully rather than dismissed.
Facilitator Miroslava Hendrych holds a background in applied statistics and has spent two years testing AI tools against traditional analytical methods in consulting contexts. She presents her findings without advocacy for any particular tool.