If every AI-assisted analysis starts with a blank chat window, you do not have a workflow yet. You have a habit.

A repeatable workflow is more useful because the same structure can work across SQL, Python, data cleaning, documentation and exploratory analysis.

Step 1: Define the task

Write the question, expected output and important business rules before asking AI for code.

Step 2: Give the smallest useful context

Include schema, sample rows, constraints and relevant definitions. More context is not automatically better context.

Step 3: Ask for diagnosis before a rewrite

When debugging, ask AI to identify hypotheses and propose small tests first. This keeps the reasoning visible.

Step 4: Generate the draft

Now use AI for SQL, Python, documentation, test cases or alternative approaches.

Step 5: Validate independently

Run row-count checks, distinct-key checks, reconciliations and edge-case tests. Compare the output with known examples.

Step 6: Keep the useful pattern

Save the workflow, not just the prompt. Next time, you should be improving a process rather than rebuilding one.

The result

The analyst moves from “What prompt should I use?” to “Which stage of my workflow should AI help with?” That is a much more useful question.

I turned this way of working into my first ebook for analysts who want practical AI-assisted workflows rather than a collection of prompt tricks.

Also read: The AI Analyst Workflow: From Question to Shippable Analysis.