Putting it together · for data analysts
Building AI workflows in data analysis
One prompt can write one query. Recurring analysis needs more. A weekly metrics review involves pulling data, validating it, computing changes, investigating outliers, and writing the summary. Asking one chat to do all of it gives a different answer each week and no record of how it got there.
Putting it together means breaking recurring work into steps, keeping the code that computes numbers separate from the text that explains them, and putting a human check where errors cost the most. The goal is analysis the team can rerun and audit.
Where it comes up
A weekly metrics notebook
The notebook pulls data with saved queries, runs validation checks, and computes the changes. AI drafts the narrative only after those steps pass, using the computed figures as its input.
Documenting a data pipeline
AI drafts a description of each transformation from the code. You add the business rules and known issues that the code does not show, and a teammate reviews the result.
Investigating an anomaly
Use AI to list candidate explanations, then write a query to test each one. You record which explanations the data ruled out and which remain open.
What it looks like at each band
E1 · Foundational
Uses AI for one step at a time, such as a single query, and moves results between steps by hand. Each new analysis starts from a blank prompt. Results vary between runs.
E2 · Proficient
Saves queries and prompts that work and reuses them. Keeps calculations in code rather than asking a chat to compute. Has a fixed point where the numbers are checked before anything is written up.
E3 · Distinguished
Designs workflows with clear handoffs between code, AI, and human review. Uses AI for drafting and explanation and code for every number. Documents the workflow so a teammate can run it and get the same result.
E4 · Exceptional
Builds the team’s analysis system and tracks where errors enter it. Adds automated validation where repeated mistakes occur and removes steps that no longer add value. Other analysts adopt these workflows because they produce consistent, auditable results.
Common mistakes
Asking the chat to be the calculator
A number produced in conversation cannot be rerun or audited. Have AI write the code, then compute with the code.
Narrating before validating
If the narrative step runs on unvalidated data, it will explain an error with confidence. Put the validation checks before the write-up.
No record of the steps
An analysis that lives in a chat history cannot be repeated next month. Keep the queries, code, and prompts in the notebook or repository.
A worked example
Producing the monthly product metrics report for leadership.
Each month, export five tables to spreadsheets, paste them into a chat, and ask for a summary of what changed.
The model computes changes in conversation, so figures cannot be checked or rerun. The format shifts each month. There is no point where someone confirms the numbers before leadership reads them.
Step 1: saved queries pull the month’s data into a notebook. Step 2: validation cells check row counts and totals against the warehouse. Step 3: code computes each metric and its change. Step 4: AI drafts the narrative from the computed table, using a fixed template. Step 5: you check each sentence against the table and add context the data cannot show. Step 6: send.
Every number comes from code that can be rerun. The model only writes prose from figures you have validated. The human checks sit at fixed points, and a teammate could run the report from the notebook.
The full study guide for this skill, for every role, is Putting it together. The other five skills for data analysts are on the AI skills for data analysts page. A one-page summary is in the AI workflow checklist.
See where you stand
The free check grades putting it together and the other five skills in about fifteen minutes. Nobody sees the grade but you.
Then build the skills and prove it. The three steps for data analysts list the courses and templates that fit the work.