Day 12 — Data privacy, ethics and 12 practice prompts
What must never leave your building, and twelve prompts to practise everything from this course.
Assume anything you paste may be kept
Consumer AI tools may retain what you send, use it for training, or expose it in a support process. Enterprise agreements often say otherwise - but that is a contract your organisation has to have signed, not a default.
- Customer names, phone numbers, email addresses, account numbers
- Employee salaries, appraisals, ID numbers
- Card details, bank details, anything financial and identifying
- Anything covered by an NDA or a client contract
- Unreleased financial results
- Passwords, connection strings, API keys - even in a code snippet
You almost never need the real values
Every prompt in this course works on the shape of the data. Replace values with fake ones of the same shape and the answer is identical.
| Instead of | Paste |
|---|---|
| Rajesh Kumar, 9876543210 | Customer A, 9999999999 |
| Actual salary figures | Numbers of the same magnitude |
| Real client names in a query | Client_1, Client_2 |
| Your production connection string | Nothing at all - it is never needed |
Being honest about AI use
Norms are still forming, but a few things are already clear in an analytics job:
- Check before you send. Owning the output means verifying it - see Day 11.
- Say so if asked. Nobody minds AI-assisted SQL. People do mind being misled about it.
- Do not present generated commentary as analysis you did. If the model wrote the reasoning, you have not analysed anything yet.
- Follow your organisation’s policy. If there is not one, ask - do not assume permission.
Twelve practice prompts
Each one has a starting prompt in the box. Improve it until the workbench scores above 85, then compare with the suggestion underneath.
1. A formula from a description
2. Explain an error
3. A SQL query from a schema
4. Clean a messy column
5. Classify with examples
6. Summarise for an executive
7. Convert between tools
8. Document a measure
9. Debug a join
10. Report commentary with no invented causes
11. Ask for assumptions
12. Critique an analysis
Where to go next
AI makes a competent analyst faster. It does not make anyone an analyst, because every prompt in this course assumed you could tell whether the answer was right.
That judgement comes from the tools themselves. The Excel, SQL, Power BI and Tableau tutorials on this site are all free, all runnable, and all build exactly the judgement this course depends on.
Try these yourself
- List six things you would never paste into a public AI tool.
- Rewrite a real prompt of yours with all identifying values replaced.
- Improve practice prompt 3 until it scores above 85.
- Improve practice prompt 6 until it scores above 85.
- Write your own rule for when you would tell a colleague AI helped.