Devon Corporation — a board-facing BI dashboard

1The problem

Devon Corporation's nine retirement-living locations each had their own reporting, no standard definitions, and no view of the portfolio as a whole. The brief: build the first portfolio-level executive dashboard — something a board could read in 30 seconds and walk away from with a specific place to focus attention and capital, not a data dump.

2The approach

Before touching Tableau, I modelled the data properly. Source tabs (locations, units, resales, contracts, NPS, monthly metrics) went through a staging → intermediate → marts layer, deliberately not flattened into one join — the source tables had different grains and didn't reconcile with each other by design, so forcing a single flat table would have quietly fabricated relationships that weren't real. Each mart answers one question at one grain, connected in Tableau via relationships rather than physical joins.

Data lineage diagram: five layers from source workbook tabs through staging, intermediate, marts, to the six board-facing dashboard views.

Source → staging → intermediate → marts → dashboard. Every number on the dashboard traces back to a source tab. Scroll the diagram sideways to read it all.

The finding that mattered — one site driving the whole portfolio's decline — didn't come from eyeballing a chart. I ran three independent tests (a control chart, trend regression, and an ANOVA) rather than asserting a pattern from a trend line, and adopted a standing rule partway through: before any claim goes on a slide, check it against all nine sites and for materiality. That rule caught and corrected four over-claims during the build — including a "3x the next site" figure that was actually 2.25x once checked properly.

Rejected hypothesis, kept anyway An early read blamed the decline on falling satisfaction scores. Checking the timeline showed occupancy had already turned down a full quarter before the satisfaction drop — the causal story didn't hold, so it was cut. A test you ran and dropped is stronger evidence of rigour than a clean narrative, so it's documented rather than hidden.

3The outcome

100%

of the portfolio's net occupancy decline traced to a single site.

$15.8Munrealised value concentrated there

33.8%of turnaround variance explained by site alone

The dashboard's visual grammar does one job: neutral everywhere, with a single emphasis colour reserved for the hero numbers and the one site that needed attention. Red (or in this fictionalised version, the same idea in a neutral accent) doesn't decorate — it points. The recommendation that followed was specific and dollar-quantified, not "investigate further": diagnose the service and resale drivers at that one site, because clearing its vacancy alone would release close to $10M in capital.

Interactive — hover a site for its full scorecard.

4What I'd take into a team

This is the clearest example I have of BI work held to the same bar as an analysis — every number on the dashboard is defensible in the room, because the modelling and the statistics were done before the first chart was built, not after.

TableauDimensional modelling SQLStatistical testing (ANOVA, regression) Data governanceExecutive storytelling

Provenance A work sample, not a client deliverable. Devon Corporation and the location names are fictionalised and the underlying figures synthetic — the modelling approach, the statistical method and the findings process are exactly what I did.