Every Marketing Source and Patient Outcome in One Dashboard You Can Just Ask

To see how marketing was doing, someone logged into Meta, then Google Ads, then Search Console, then GA4, exported from each, and stitched a spreadsheet, which still could not connect a campaign or a search query to a patient who actually booked and got treated. Paid sat apart from organic; web behavior sat apart from outcomes. The owner could not answer a simple question, "how many new patients did Google bring last month, and what treatment did they book?", without making it a project. As spend spread across more channels, the reporting got slower and the picture got blurrier.

Five dashboards is not a dashboard.

The governing rule: every marketing source plus the patient outcomes from Dentally land in one model, served two ways, a live custom dashboard and a plain-English LLM interface, so the team sees paid, organic, web, and patient outcomes together and gets answers without exporting anything.

Two decisions shaped it. A custom dashboard rather than an off-the-shelf BI tool, because the group wanted a branded, purpose-built marketing view built around patient outcomes, not a generic chart gallery. And a governed LLM interface: it answers by querying the defined set of metrics in the model, not by writing raw SQL against raw tables, so a question asked in plain English always returns a number that matches the dashboard.

The privacy boundary is explicit: only marketing-relevant fields come from Dentally (bookings, treatment value, source), never clinical detail, and the LLM answers over the marketing model, not patient records.

What we deliberately did not do: no revenue feedback to the ad platforms (that is the attribution engine's job, a separate build), no clinical data in the dashboard or within the LLM's reach, and no raw text-to-SQL free-for-all.

Managed connectors (Airbyte) pull Meta Ads, Google Ads, Search Console, and GA4 (GA4 through its native BigQuery export); a Python connector pulls bookings and treatment outcomes from Dentally. Everything lands in BigQuery, where dbt builds one marketing model that joins spend, organic search, and on-site behavior to booked patients and treatment value, with a defined set of metrics. A custom TypeScript dashboard on Cloud Run serves the live view across all five sources. An LLM interface, Claude Opus 4.8 for analytical questions and Claude Haiku 4.5 for simple lookups, turns plain-English questions into governed queries over those defined metrics and returns the answer with a chart, so "how many new patients did Google bring last month, and what treatment did they book?" is one question, not a spreadsheet.

The group now has one live dashboard across all five sources, with roughly eight hours a week of manual reporting eliminated. For the first time marketing is tied to booked patients and treatment value, so the owner sees which channels, paid and organic alike, actually produce patients, in one place. And anyone can ask the data a question in plain English and trust the answer, because it is computed from the same defined metrics the dashboard uses.

The unified marketing model is the foundation. It pairs naturally with the patient-attribution engine when the group wants to move from seeing performance to optimizing spend and feeding value back to the platforms. The same model and the same natural-language interface extend to new locations and to operational KPIs, so one command center grows with the group rather than being rebuilt per practice.