Fivetran spend review board with pause and downgrade actions

Build me an app for reviewing what our Fivetran pipelines actually cost. The audience is our data platform lead and the finance partner who signs off the Fivetran bill. Fivetran's billing screen tells us what we used but never why a particular sync got expensive, so this app has to connect billed row volume back to the specific connectors and tables producing it, and let us act on what we find without leaving the screen.

Important context on where the numbers come from. The Fivetran REST API does not expose billing or monthly active row figures at all. Those come from the free Fivetran Platform Connector, which syncs usage and sync logs into our warehouse. So every cost and volume number in this app is read from Snowflake with Execute SQL Statement, and everything about connector configuration and state comes from the Fivetran API. The Platform Connector schema is usually named FIVETRAN_LOG or FIVETRAN_PLATFORM depending on when it was set up, and the row counts live in the incremental MAR table with one row per connector, schema, table, and measured date. Make the database and schema names a setting in the app rather than hardcoding them, and run a discovery query on first load so a wrong schema name produces a clear message instead of a broken board.

The main view is a spend board: one row per connection, ranked by monthly active rows for the current month. Query Snowflake with Execute SQL Statement to aggregate rows per connector for the current month and the previous three months, then call Fivetran List Connections and join on connector id to bring in the connector name, source service type, group and destination, paused state, sync frequency, and last sync status. Columns: connector, source type, destination, rows this month, rows last month, month over month change as a percent, sync frequency, paused or active, and last sync outcome. Default sort is rows this month descending, and I want to re-sort by month over month change to surface runaway connectors.

Above the table put two callouts that make the two failure modes impossible to miss. Dormant: connectors that are active and syncing on a schedule but have produced no rows, or near zero rows, for the last 60 days. These still cost money, because Fivetran bills per connector with a minimum charge on every active connector, so an idle connector is not a free connector. Runaway: connectors whose row count has grown more than a threshold, default 50 percent, month over month. Each callout shows a count and filters the board when clicked. Both thresholds should be editable in settings.

Clicking a row opens a connector detail view. It shows the monthly row history as a simple chart from the last 6 to 12 months of warehouse data, the full connector record from Get Connection including schedule, sync frequency, setup state, last sync time and any failure info, and a table level breakdown of which tables inside this connector produced the most rows, from the same warehouse data grouped by schema and table. Next to that breakdown, show which tables and columns are actually enabled using Get Connection Schema Config, so a table driving high volume can be matched to the switch that turns it off.

Every board row and the detail view have an Audit this connector button that starts a background agent. The agent reads that connector's monthly row history from Snowflake including the per schema and per table breakdown, calls Get Connection and Get Connection Schema Config for its schedule and its enabled tables and columns, and works out which specific tables and columns are driving the row count and whether the volume looks like real growth, a schema change, a re-sync, or a table nobody queries. It then writes a recommendation back into the app naming specific tables to block, columns to disable, a sync frequency to stretch, or a connector to pause, each with the row impact of doing it. Store each recommendation against the connector id with a timestamp, so the board can show a last audited date and the detail view shows the latest recommendation plus its history. Show progress while the agent runs and surface the finished recommendation without needing a manual refresh.

From the board and the detail view I need to act directly. A pause and resume control that calls Modify Connection to set the paused state. A sync frequency control that calls Modify Connection to stretch the schedule to a longer interval, for example from 15 minutes to 6 or 24 hours, with the common options in a dropdown. A per table switch in the detail view that calls Modify Table Config to disable a table nobody queries. Confirm every action first with a short summary of exactly what will change, then record it in the app with who did it, when, which connector or table, and what changed, shown as an activity log on the detail view. Refresh the affected row after the call so the board reflects the new state immediately.

Add an Export for finance action on the board that appends the current review to a Google Sheet using Append Values. One row per connector with the review date, connector name, source type, destination, rows this month, rows last month, month over month change, sync frequency, paused state, the latest agent recommendation summary, and any action taken since the last export. Let me pick the spreadsheet and tab and remember that choice. Always append and never overwrite, so the sheet becomes a running history the finance partner can chart across months.

Keep it dense and readable, this is a working board rather than a marketing dashboard. Format large numbers so they scan (1.2M rather than 1204322), colour the month over month change by direction, and visually distinguish dormant and paused rows. Loading the board should be one Snowflake query plus one Fivetran list call, cached briefly, with a manual refresh button. If the Platform Connector data is missing or looks stale, say so at the top of the board with what to check, rather than showing an empty table.