Generate custom courses on any topic — with hands-on practice, AI guidance, and visuals built in.
Already have an account?
A VP Growth asks for a weekly exec KPI dashboard for the SaaS growth review. The hidden risk is that everyone means something different by revenue, churn, and activation, and the first AI-built dashboard will faithfully encode those contradictions.
Define success as a dashboard that supports named decisions, not a dashboard that contains all the KPIs. In this meeting, decisions usually include whether pipeline coverage is sufficient, whether churn is concentrated in a segment, and whether activation changes warrant a product rollout pause.
Define non-goals just as explicitly. You are not building a forecasting model, not building a rep performance leaderboard, and not creating a single view that works for execs, managers, and ICs simultaneously.
If you do not make those boundaries real in the brief, the downstream failure is yours. The AI will generate charts that look plausible, and you will own the consequences when an exec makes a decision on a mis-specified metric.
A usable dashboard brief pins down five things the AI cannot safely infer.
Start with audience. Execs need decision thresholds and exceptions. Operators need diagnostic breakdowns. If you collapse them, the AI will default to generic summaries and you will deliver something that nobody can use.
Set cadence and refresh timing. Weekly means a fixed close time, for example Monday 06:00 UTC, not a rolling last 7 days window. Rolling windows silently change denominators and make week over week comparisons invalid.
Specify grain as the unit of analysis, for example account by calendar week over 52 weeks with 18,000 account-week rows. Grain determines what can be aggregated without double counting.
List dimensions you will slice by, for example plan tier, acquisition channel, geo region, and cohort month. Dimensions are not decoration. They are the hypothesis space you are willing to defend.
Lock a metric dictionary. A definition includes numerator, denominator, inclusion rules, exclusion rules, and time alignment. If churn is logo churn for paid accounts excluding refunds, write that. If it is revenue churn including downgrades, write that instead. The AI cannot reconcile competing business meanings.
Look at an annotated one-page brief and notice what is specified versus assumed.
AI is useful here as a structured interviewer. You use it to generate intake questions and a draft brief, then you verify each field against your source of truth, which is the stakeholder decision and the underlying tables.
A scoping prompt that works is explicit about what you need filled in and what must be testable.
Prompt
You are helping scope a weekly exec dashboard for a SaaS growth review.
Dataset: subscriptions, invoices, product_events. Time range: last 52 weeks.
Return a one-page dashboard brief with:
- audience and decisions supported
- cadence and close time
- grain and required joins
- dimensions and required filters
- metric dictionary for MRR, gross revenue churn, logo churn, CAC, activation rate
For each metric include numerator, denominator, window, inclusion/exclusion rules.
Add acceptance tests that can be run as queries (e.g., totals reconcile to finance MRR).
Ask 8 clarification questions where definitions are ambiguous.
The acceptance tests are where you protect yourself. If the brief says MRR must reconcile to finance within 0.5 percent at month close, that becomes a gate. Without gates, you will argue about screenshots instead of definitions.
Practice using a structured template that outputs a brief plus acceptance tests.
Three traps cause most dashboard rework.
Vanity KPIs appear when a stakeholder asks for signups without specifying activation or retention. If the decision is growth quality, you must anchor the KPI to a behavior, for example activated within 7 days, not a top of funnel count.
Mismatched grains happen when someone wants CAC by channel and churn by account on the same tile without defining the join path. If CAC is campaign-month and churn is account-week, the AI will pick a join that multiplies rows. You must decide the primary grain and which metrics are displayed as separate tiles versus combined.
One dashboard for everyone fails because audiences require different defaults. Execs need top-level with alerting thresholds. Managers need drill-down. You cannot satisfy both with one set of filters and one narrative.
Diagnose flawed requests by naming the grain conflicts and missing metric definitions.