Single requests are useful, but the payoff compounds when the agent builds whole workflows — jobs that run on a schedule, chain several steps, and plug into email or Slack. Metabase has the pieces: questions, alerts, dashboard subscriptions, and a full API underneath. Here are ten workflows worth setting up once and then forgetting.

1. The 8am Business Digest

Every weekday at 8am, leadership wants yesterday’s numbers: revenue, orders, signups. The agent builds the three questions, puts them on one dashboard, and schedules a dashboard subscription to email the team. Example: “Create a Morning numbers dashboard with revenue, new orders, and signups from yesterday, and email it to the management group every weekday at 8am.” Send the first run to yourself only — verify the numbers before looping in the CEO. Once it is trusted, the daily ritual needs zero human effort.

2. Stock-Out Watchdog

When inventory of any SKU drops below its reorder point, someone should know the same day. The agent writes a SQL question that joins inventory levels to reorder thresholds, then sets a Metabase alert that fires whenever that question returns rows, routed to purchasing. Example: “Build a low-stock alert: any SKU under its reorder point should email purchasing@ourcompany.com every morning.” Test it against a deliberately low SKU so you see the email format before it matters. This replaces a daily manual stock check with a five-minute setup.

3. Actuals vs Targets Scorecard

Revenue alone is trivia; revenue against target is management. Have the agent create a targets table — or upload your spreadsheet as a CSV and turn it into a model — then build a question comparing actuals to target per month with a variance column. Example: “Upload our 2026 targets CSV and build a dashboard comparing monthly actual revenue against target, with the variance as a bar chart.” Add a monthly subscription email and the review meeting writes itself. When targets change, update one table instead of editing forty questions.

4. Data Freshness Watchdog

Reports are dangerous when the pipeline stalls but the charts still render last week’s data. The agent builds a question that selects max(updated_at) from your key tables and sets an alert if it goes stale beyond a few hours. Example: “Alert me if the orders table hasn’t been updated in 6 hours.” It is a five-minute setup that catches broken pipelines before anyone presents stale numbers in a meeting. Point the same pattern at every table your dashboards depend on.

5. Nightly Re-Sync After the ETL Window

If your warehouse loads overnight, Metabase should re-scan afterwards so new columns appear and field values stay current. The agent schedules a cron job that calls Metabase’s sync and scan API endpoints right after your load finishes. Example: “Re-sync the warehouse database every night at 3am after the dbt run completes, and log the result to a file.” It will check how long a full scan takes so the job does not collide with your morning emails. Your analysts stop asking why last night’s table is missing from the query builder.

6. A Reporting Schema That Shields Production

Point Metabase straight at production and one bad dashboard can slow the real app. Have the agent create a dedicated read-only user with a statement timeout, or point the connection at a replica, and move your heavy questions onto it. Example: “Create a metabase_report database user with read-only access and a 30-second statement timeout, and re-point the production database connection at it.” It can also convert your slowest recurring questions into models that pre-aggregate nightly. Production stays fast; reporting stays possible.

7. Route Alerts into Slack

Email alerts die in inboxes; Slack alerts get acted on. The agent configures Metabase’s Slack integration with a bot token and channel, then points your existing alerts at the channel instead of an email list. Example: “Send the low-stock alert to the #purchasing Slack channel instead of email.” Expect some back-and-forth on Slack app permissions — that is normal, and the agent can read the OAuth error and fix the scope. Once wired, every workflow above gets louder and faster.

8. Per-Branch Weekly Packs from One Dashboard

Ten branches need the same dashboard filtered to their own numbers. Instead of ten copies, the agent builds one dashboard with a branch filter, then creates a scheduled subscription per branch with the filter pinned to that branch’s value. Example: “Set up weekly Friday emails of the branch dashboard, one per store, each filtered to that store.” If your Metabase version does not support pinned filter values on subscriptions, it will fall back to one dashboard per branch and tell you honestly which approach it took. Either way, branch managers stop asking head office for their numbers.

9. The Backup-Then-Upgrade Ritual

Upgrading Metabase is painless until it is not. Have the agent script the ritual: back up the application database, pull the new image, restart, then run a smoke test that hits three known questions and checks the version string. Example: “Back up the Metabase app DB, upgrade to the latest release, restart the container, and verify my three core dashboards still load.” Run it quarterly and upgrades stop being scary. Keep the backup script even if you never need it — that is the point.

10. Monthly Board-Pack Export

Every month somebody hand-screenshots dashboards into a deck. The agent automates the boring half: on the first of the month it exports the executive dashboard as PDF or PNG through the API and drops the files in a shared folder with the date in the name. Example: “On the 1st of each month, export the Executive dashboard to PDF and save it to /backups/board-pack/.” A human still writes the commentary — the agent just handles the copies. Over a year that is a full working week handed back.

Pick the two workflows that would save you the most hours this month and start there. Each one is a single conversation, and the agent will show its work as it goes. Metabase on OpenSysLab