Audience-Driven Dashboards Across Three BI Platforms

You're designing three dashboards on three BI platforms from one governed data set for Sophie Whitcombe, the CEO of a boutique hotel group across southern England.

The discipline skills: audience-driven dashboard design as producing fundamentally different artifacts rather than the same dashboard filtered three ways; tool selection across three BI platforms — Metabase, Tableau, Apache Superset — driven by audience workflow; cross-platform metric verification when one governed definition is referenced by three consumers; cross-dialect SQL audit between BigQuery and DuckDB; BigQuery cost as a design dimension where refresh frequency and pre-aggregation per audience matter as much as which charts ship; and a maintenance-strategy document that names retirement criteria for platforms that no longer earn their place.

The AI-direction lesson: AI generates one dashboard for all three audiences, treats three platforms as breadth rather than a strategic decision, transplants one platform's idioms into another, defaults to full-table scans against cloud data, and does not surface dialect differences unless directed. You name the audience-driven design principle per audience, record platform-per-audience reasoning, design verification across three handoffs, and surface the maintenance-strategy question Sophie owns. You also encode the directing discipline as infrastructure — a dashboard-design-review skill that catches AI's audience-blind defaults without re-prompting per session.

Your Role

You're the analyst Sophie reaches out to because she has three groups of people who need different things from the same data and an Excel dump that satisfies none of them. The deliverables are three working dashboards on three platforms, a cross-platform verification spec, a maintenance-strategy document, and a reusable design-review skill that lives on after the project closes.

The infrastructure floor — project memory, AGENTS.md, the profile-dataset skill, the pre-commit hook, MCP scopes for DuckDB and BigQuery, the dbt semantic layer, the path-scoped rules under experiments/, the connectivity architecture artifact — runs in the background. What's new is that your judgment now spans three audiences in parallel rather than five phases in sequence: per-audience information architecture, platform fit, refresh policy, and a single consistency floor across all three. You're the keeper of one number meaning one thing across three different visual treatments.

What's New

Last time you owned the full experiment lifecycle for Coral Coast Rentals — the business-first chain, test-type and alpha per metric, the phase-scoped connectivity architecture, the governance framework draft, and three audience-calibrated reports from one experiment.

This time the three audiences are live dashboards rather than written reports. The first contact is a meeting transcript — Sophie has framed the three-audience problem in her own words, but the per-audience information architectures are yours to design.

Three things are genuinely new. Three fundamentally different dashboards from one governed data set — the board's monthly portfolio view with GOPPAR alongside RevPAR, the GMs' daily per-property operations dashboards with threshold alerts, and the revenue manager's SQL-Lab-backed self-service for booking pace and channel-mix exploration. Apache Superset enters as the third BI platform, connected to BigQuery. And the dashboard-design-review skill is authored before any dashboard ships — encoding the audience-driven rubric so AI's audience-blind defaults are caught by infrastructure.

The hard part is the moment AI's three first-draft dashboards each fall back to the same template. You catch the audience mismatch per dashboard and rebuild each information architecture to fit.

Tools

• Established and carried forward: Claude Code, Codex CLI, BigQuery and DuckDB (with their MCP servers), dbt Core (read-only), Metabase, Tableau Desktop, Python with pandas, Jupyter, Git and GitHub with pre-commit, the root and experiments/ CLAUDE.md and AGENTS.md, the profile-dataset skill, the infrastructure/connectivity.md artifact, and the auto-memory from P17 (curated at project start).
• Apache Superset — new; the third BI platform, installed in Unit 4 from the reference install guide, connected to BigQuery, used for the revenue manager's self-service tool.
• Superset API — new; the dashboard-authoring connection scope.
• dashboard-design-review skill — new; SKILL.md format with a scoped description encoding the audience-driven design rubric.
• platforms/strategy.md — new artifact register; the maintenance-strategy document naming which platform serves which audience and the retirement criteria.

Materials

• Sophie's meeting transcript — the entry artifact. The three-audience problem is framed in her own words; the per-audience information architectures and the platform-per-audience map are yours to design.
• Synthetic Hedgerow Hotels dataset in BigQuery — three years of property-management system data across five properties, around 500,000 reservations. Tables for reservations, properties, housekeeping, maintenance, F&B revenue, three guest-satisfaction sources on different scales, competitor rates, channel bookings, and rate history.
• Carried-forward dbt project — dbt-hedgerow/ loaded against Hedgerow's BigQuery dataset with the hospitality metric definitions (occupancy_rate, adr, revpar, goppar, daily_revenue, customer_satisfaction_normalised, channel_mix_share) and per-property views.
• Starter directories — experiments/ (path-scoped memory carrying forward from P16-P17), infrastructure/connectivity.md (P17 file with an empty P18 section), platforms/strategy.md (empty template), audience-profiles/ (three empty briefs — board.md, operations.md, revenue-manager.md), and skills/dashboard-design-review/SKILL.md (empty skeleton).
• Reference primers at reference/ — audience-driven dashboard design, BigQuery cost discipline, the three-platform paradigm comparison, cross-dialect SQL verification, dashboard maintenance strategy, the hospitality metric set, and the Apache Superset install guide.
• Project CLAUDE.md and AGENTS.md — context and tech stack populated; the curation notes, audience profile briefs, platform-per-audience map, dashboard summaries, refresh-and-cost policy, verification spec, audit notes, the new skill, and the maintenance-strategy document are placeholders you fill in.

Download the materials zip (https://learntodriveai.dev/materials/analytics/p-18/materials.zip) and unzip it to ~/dev/analytics/p-18.