A step-by-step way to work
Map the question to the layer
Decide whether you need product analytics, BI/warehouse, qualitative, or experimentation — AI can help you place the question.
Query in natural language where available
Many analytics and BI tools let you ask in plain English; use it, then verify the definition it used.
Bridge from 'what' to 'why'
AI can turn a 'what happened' finding into hypotheses about 'why' — but the why needs qualitative or experimental work.
Synthesise across tools
Pull findings from different layers and ask AI to reconcile them into one picture.
Keep definitions straight
Confirm AI is using your canonical metric definitions, not its own guess.
Real prompts, and what good output looks like
I want to know why our week-2 retention dropped. Which parts of my analytics stack should I use, and in what order?
1) Product analytics / warehouse — confirm the drop is real and segment it (which cohort, platform, acquisition source). This is the 'what'.
2) Once localised, qualitative — session replays, support tickets, or a quick survey of the affected cohort to get at 'why'.
3) Experimentation — if you form a fixable hypothesis, test the fix.
AI (me) can help at each step, but 'why retention dropped' is not answerable from the quantitative layer alone.
A prompt you can reuse
Fill in the highlighted parts and paste it into your AI tool of choice. Edit the output — it's a starting point, not a finished answer.
Help me use my analytics stack to answer a question. Map it to the right layers and flag what each can and can't tell me. My question: [what I want to know] Tools I have: [e.g. product analytics, a warehouse + SQL, qualitative tools, an experimentation platform] Please: 1. Break my question into what-happened vs. why parts. 2. Tell me which tool/layer answers each, and in what order. 3. Flag explicitly where the quantitative layer can show correlation but not cause, so I know when to switch to qualitative or experiments. 4. Note any metric I should pin to a canonical definition before querying.
What AI gets wrong here
Mistaking 'what' for 'why'
Natural-language analytics make it easy to get a confident answer that only describes what happened.
Definition drift
AI may compute 'active users' or 'retention' with its own definition, not yours.
Tool sprawl answers
AI may answer from whatever's easiest rather than the right layer.
AI makes every layer of the stack faster to query, but it can't supply causation or motivation — and it won't volunteer that limit unless you ask. The PM skill it doesn't replace is knowing which question belongs to which tool, and recognising when a fluent 'what happened' needs a 'why' that only qualitative or experimental work can give.