Define what the filter should change
Start with a question a teammate can answer from the dashboard. In a fictional trial-activity report, the team might want to see activity during a selected date range. Decide whether that means account creation time, event time, or another timestamp. A widget labeled Date is ambiguous when different cards measure different moments.
Write a small expected-result sheet before editing: the selected period, the records included, and the cards expected to change. Use harmless data with an empty period as well as a populated one. This guide's proposed exercise is not a query result from a live customer database.
Add the widget and connect the intended cards
Open the dashboard's edit mode, add a filter in the intended location, and select its type. For each relevant card, map the widget to the field it should constrain. Inspect automatic connections rather than assuming a date filter selected the right timestamp everywhere. Dashboard-level and card-level widgets have different connection scopes.
Use a mapping table to explain the choice. A card counting signups can use account creation time, while a card showing activity uses event time only if that difference is intentional and visible to the reader. If the goal is a consistent signup cohort, a mixture of those meanings would need a different question design.
| Card | Intended field or parameter | Check before saving |
|---|---|---|
| Accounts created | Account creation timestamp | Known accounts match the selected period |
| Activity events | Event timestamp | Card label explains that this is event time |
| Native SQL activity count | Mapped date field filter | Variable and dashboard connection both exist |
| All-time reference | Deliberately unconnected | Label explains why the number stays fixed |
Prepare SQL questions before connecting them
Metabase supports field filters and basic variables in native SQL questions. A field filter maps to a database field and lets Metabase build the predicate. A basic variable supplies a value and uses different SQL syntax. For flexible date ranges on an actual database field, review the field-filter option before choosing a basic Date variable.
The original illustration below assumes a table named trial_activity with an occurred_at field. In the variables panel, set activity_period to Field filter and map it to that field. Adapt the schema and SQL dialect to your database; this is a teaching query, not an executed result. Save the question, then connect the dashboard widget to its parameter.
SELECT COUNT(*) AS activity_count
FROM trial_activity
WHERE {{activity_period}}Verify results and the dashboard's actual refresh
Compare an ordinary period, an empty period, and records close to the chosen boundary. Note the timestamp meaning and time-zone assumptions used by the question. Check every intended card, including the SQL card, against the expected-result sheet. A changing headline total alone does not demonstrate that the whole dashboard uses the intended subset.
Inspect whether the dashboard applies filter changes automatically or expects an Apply action. Keep filter settings separate from the period printed in a screenshot or report. For a handoff, include the selected values, question definition, and observation date so the next person can reproduce the same view.
Why does one Metabase card ignore the filter?
Inspect its connection, available field, SQL variable type, and widget scope. For a native query, check field mapping and any table alias as well as the dashboard connection. Compare a single known record before rewriting the entire dashboard. If a card is intentionally all-time, label that choice so readers do not mistake a fixed reference for a broken filter.
Does a required customer filter restrict data access?
Treat access as a separate configuration. Metabase distinguishes permissions on data from permissions on collections containing questions and dashboards. Review the appropriate authorization and data-isolation approach for customer-facing use. A visible customer selector or required default is a way to choose results, not evidence that an unrelated customer cannot reach other data.
Sources and editorial method
Official sources linked throughout this guide were checked on October 29, 2026. Product editions, interfaces, plans, and limits can change. The examples and evaluation exercises are editorial proposals, not reports of hands-on product testing or customer results. Confirm the current documentation for your account before implementing a workflow.