looker-studio-bigquery skill by akillness

Looker Studio + BigQuery

Use this skill when the real deliverable is one trustworthy stakeholder-reporting packet on top of BigQuery.

looker-studio-bigquery owns the reporting layer for:

PM / ops KPI review boards
product funnel, retention, and rollout dashboards already modeled in BigQuery
marketing / GTM channel and revenue reporting
game / live-ops / business telemetry review boards
refresh, cost, audience, and export-handoff decisions for Looker Studio on curated data

Read these support docs first when needed:

When to use this skill

The user explicitly wants Looker Studio / Data Studio + BigQuery help.
The data already lives in BigQuery, or the dashboard clearly sits on curated BigQuery tables/views.
The main job is dashboard structure, refresh/cost shape, audience-specific delivery, or export-ready reporting.
The request is about KPI boards, PM/ops reviews, marketing/GTM reporting, product summary dashboards, or game/business telemetry dashboards.
The real risk is dashboard trust, refresh design, or mixed-audience sprawl rather than raw SQL interpretation alone.

When not to use this skill

The main job is explaining why a KPI moved or what the dataset means → use data-analysis
The main job is repeated anomaly hunting or reusable rule scans across metrics/events → use pattern-detection
The main job is telemetry reliability, alerting, or instrumentation coverage → use monitoring-observability
The main job is choosing a full BI / semantic platform → use survey
The main job is broad GCP/bootstrap/deploy work → route to the relevant infrastructure skill first

Instructions

Step 1: Pick one primary packet before talking about charts

Normalize the request into one packet.

looker_studio_bigquery_packet:
  primary_packet: dashboard-spec | slow-dashboard | refresh-shape | audience-split | exec-handoff
  audience: executive | pm-ops | product-analytics | marketing-gtm | game-liveops | mixed | unknown
  source_shape:

Pick exactly one primary_packet:

dashboard-spec — build or redesign a stakeholder dashboard
slow-dashboard — fix a dashboard that is slow, expensive, or brittle
refresh-shape — decide live vs scheduled vs snapshot vs BI Engine
audience-split — separate one overloaded report into audience-specific artifacts
exec-handoff — produce a thin dashboard plus export/sheet/deck handoff for commentary and approvals

If two packets seem plausible, choose the one that removes the biggest delivery risk first.

Step 2: Keep the dashboard thin on purpose

Before designing sections, answer these:

What review ritual or decision does this dashboard support?
What table/view is the source of truth?
Which metrics belong in BigQuery instead of report-level formulas?
Who owns freshness, trust checks, and caveats?

If the request still depends on raw fact tables, unstable joins, or fuzzy metric definitions, say so and push modeling work upstream before polishing the dashboard.

Step 3: Shape the packet, not a generic BI essay

Use the packet guidance in references/intake-packets-and-route-outs.md.

Minimum packet expectations:

dashboard-spec → page structure, KPI set, chart-to-question mapping, and BigQuery contract
slow-dashboard → bottleneck order, upstream precompute options, and one shortest-path fix
refresh-shape → live vs scheduled vs snapshot choice tied to the real review cadence
audience-split → artifact split by stakeholder ritual, trust need, and detail depth
exec-handoff → thin dashboard plus Connected Sheets / export / slide handoff plan

Do not answer with a generic chart buffet.

Step 4: Return a delivery brief

Use this structure:

Step 5: Route out honestly

If the user needs dataset reasoning or KPI explanation, route to data-analysis.
If the user needs repeated anomaly detection or reusable metric rules, route to pattern-detection.
If the user needs telemetry coverage, alerting, or instrumentation trust, route to monitoring-observability.
If the user needs a heavier BI or semantic-platform decision, route to survey.

Examples

Example 1: PM / ops dashboard build

Input

Build a leadership-ready PM review dashboard in Looker Studio on top of BigQuery weekly KPIs.

Output sketch

Packet: dashboard-spec
Dashboard shape: KPI scorecards + main trend + one drilldown + owner/freshness notes
Next artifact: dashboard spec

Example 2: Slow marketing board

Input

Our marketing funnel dashboard is too slow and expensive whenever people touch filters.

Output sketch

Packet: slow-dashboard or refresh-shape
Primary recommendation: precompute the repeated funnel logic in BigQuery before styling changes
Next artifact: refresh decision memo or SQL handoff list

Example 3: Mixed audience dashboard

Input

One board is trying to serve executives, PMs, growth, and live-ops. Should we keep patching it?

Output sketch

Packet: audience-split
Recommendation: split the artifact by decision ritual and trust needs
Next artifact: audience split brief

Example 4: Commentary-friendly handoff

Input

Leadership wants a weekly board, but they still annotate numbers in Sheets before the review.

Output sketch

Packet: exec-handoff
Recommendation: keep the dashboard thin and define the export / Connected Sheets handoff explicitly
Next artifact: export / Connected Sheets handoff

Best practices

Start from the current packet, not from chart types.
Keep heavy logic in BigQuery whenever possible.
Tie freshness to the real review ritual instead of pretending everything must be live.
Split dashboards by audience when one artifact keeps absorbing incompatible needs.
Name the owner, caveats, and trust contract in every serious dashboard handoff.
Prefer route-outs over stuffing analysis, anomaly hunting, or observability into this skill.

References