PM Mapped
Working with AI ANALYTICS IN PRACTICE · Building Dashboards

Building Dashboards with AI

AI speeds up dashboard building — drafting the queries behind each tile and suggesting visualisations — but a dashboard is a decision tool, not a data display, and deciding what earns a place on it is editorial judgment AI can't do for you.

← Back to the Building Dashboards tool
1How to use AI for this

A step-by-step way to work

1

Start from the decision, not the data

Tell AI what decision or question the dashboard serves; that constrains what belongs on it.

2

Draft the queries per tile

Have it write the SQL behind each metric (and verify it, as with any AI SQL).

3

Get visualisation suggestions

Ask which chart type fits each metric and why — AI is reasonable on chart-to-data fit.

4

Write the plain-English caption

Have it draft the 'what this tells you / what to do if it moves' note for each tile.

5

Cut ruthlessly yourself

AI will happily fill a dashboard; you decide what to leave off so the signal stays clear.

2Worked examples

Real prompts, and what good output looks like

Choosing the right chart
Your prompt

I'm showing weekly active users, retention by cohort, and the share of revenue by plan tier. What chart type for each, and why?

What good output looks like

Weekly active users → line chart: shows trend over time at a glance.

Retention by cohort → cohort/triangle heatmap: designed to show how each signup cohort decays over weeks.

Revenue share by plan tier → stacked bar over time (not a pie): a pie hides change; stacked bars show both share and trend.

Avoid a pie for the third — you'll want to see the mix move week to week.

Solid guidance, and the 'not a pie' steer is the kind of small correctness AI is good at. The call on whether these three even belong together is still yours.
3Copy-paste template

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.

Reusable prompt
Help me build a dashboard. Start from the decision it serves, then the tiles.

This dashboard helps [who] decide [what decision/question].
Candidate metrics I'm considering:
[list them]
My data:
[tables and key columns]

Please:
1. Tell me which candidate metrics actually serve that decision — and which to cut as noise/vanity.
2. For the keepers, draft the SQL behind each (explain it so I can verify).
3. Suggest the right chart type for each and why.
4. Draft a one-line 'what this tells you / what to do if it moves' caption per tile.
4Common pitfalls

What AI gets wrong here

Cluttered, vanity-heavy dashboards

AI will build whatever metrics you name, including ones that don't drive a decision.

Do this instead: Make it justify each tile against the dashboard's decision; cut anything that's just 'nice to see'.

Unverified tile queries

Every AI-written query can carry the usual SQL traps (wrong joins, double-counts).

Do this instead: Verify each query as you would any AI SQL; sanity-check the number on each tile.

Pretty over clear

AI may suggest flashy charts (pies, gauges) that obscure the data.

Do this instead: Favour the chart that shows change and comparison clearly; reject decoration.
The judgment that stays yours

AI is a real time-saver for the build — queries and chart choices — but the discipline of a good dashboard is subtraction: keeping only what informs a decision. That editorial call, and the responsibility for what a viewer will conclude, stays with you. A clean dashboard of the right three metrics beats an AI-assembled wall of twenty.