AIAP Topic Courses·Data Analysis·17:51

Claude (Opus) and ChatGPT for Data Analysis: a Role-Defined Prompt, a Targeted Question, and a PDF Report

Data Analysis trainer (unnamed) Marketing-analytics practitioner; runs four AI data tools on one e-commerce orders dataset

▶LMS lesson

The short version

Upload the CSV to Claude and write a role-defined prompt: 'you are an expert in data analysis... do the entire univariate, bivariate and multivariate analysis, find relationships, give recommendations and insights to the team'. Context matters - 'I want my model to understand why'. Opus chosen for power; it took 4-5 minutes, not the promised 2-3 (0:00-0:06).
Same prompt in ChatGPT in parallel: it starts with a data dictionary and plain-text steps; Claude mostly ran JavaScript, ChatGPT Python (one session's observation). Targeted follow-up: kurta orders in cancelled state by platform - a three-variable question (0:06-0:11).
Ask for a PDF of the analysis (images need a separate ask). Claude's report: delivery and cancellation KPIs, a 67%-expedited / 0%-success finding, Meesho highest AOV, Amazon volume leader, Myntra most cancellations, recommendations (0:12-0:20).

At a glance, three clicks deep

Skim here first: the closed row is the glance, open is the study card with the key points and timestamps, and the ↓ link drops to that concept's full write-up below.

01The role-defined, open-ended analysis promptRole + goal + named analyses + audience.›

Role + goal + named analyses + audience.

Prompt (0:01-0:05)

ChatGPT comparison (0:06-0:09)

Report (0:13-0:20)

↓ Full write-up of this concept

02Targeted questions and PDF exportNarrow question after broad report;›

Narrow question after broad report; export.

Q&A (0:09-0:11)

PDF (0:12)

↓ Full write-up of this concept

The concepts in full

The role-defined, open-ended analysis prompt

Tell it who it is, what the data is, what the three analyses are called, and who the recommendations are for. Then wait five minutes.

Prompt pattern: role ('you are an expert in data analysis'), goal ('uncover insights from this dataset'), the vocabulary ('univariate, bivariate and multivariate analysis, relationships between variables'), the deliverable ('recommendations and insights to the team'), optional framing ('solve this as an assignment'). Context tells the model why it is doing this. Opus was chosen for capability; runtime ~4-5 minutes. Result: executive summary with KPIs, an 'expedited delivery crisis' (67% of customers chose expedited, 0% delivered), platform insights (Meesho AOV ~708, Amazon volume, Myntra cancellations), category and geography, strategic recommendations. ChatGPT with the same prompt: data dictionary first, plain-text steps, Python.

Why it matters

The vocabulary from lessons 2-7 is what makes this prompt work.

This idea elsewherebuilds onThe four types

Targeted questions and PDF export

After the big report, ask the small question: how many kurta orders were cancelled, per platform. Then ask for the PDF.

Follow-up prompt: 'how many orders for the item kurta are in cancelled state across different e-commerce platforms' - category x status x platform, a multivariate question answered in seconds. Then 'give me a PDF of this analysis and recommendations'; images were not included by default - ask for them.

Why it matters

The chat becomes an analyst you can interrogate.

Tools referenced

Tool	Coverage	Moment	Context
Claude	demonstrated		Opus model; CSV upload, report, PDF
ChatGPT	demonstrated		Same prompt in parallel

Action items

Resources mentioned

Resources

docLesson page on the platform

Extraction notes

This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

Transcript corrections applied

The transcript says	The trainer actually means
rule defining	role-defining
Charge GPT / Chart GPT / ChargeGPD	ChatGPT
cloud / Cloud AI	Claude
Mishou / Misho / MISO	Meesho
main dot com / mail dot com	Make.com

True on recording day — verify before relying

Model names, runtimes, file limits and which language each tool runs are a single-session snapshot; the dataset figures (67%, AOV 708) are specific to the trainer's file