Skip to main content
Analytics turns the data already flowing through Mage into interactive dashboards for business reporting, operational insight, and team decision-making. Use it to explain revenue trends, customer growth, product adoption, campaign performance, data quality, model behavior, warehouse costs, and any other metric your pipelines produce. Instead of exporting data to another tool before anyone can understand it, teams can create and refresh dashboards directly where the data is prepared. Analysts, data engineers, operators, and business stakeholders can all work from the same trusted source. Analytics is different from monitoring. Monitoring tells you whether your pipelines, schedules, logs, alerts, and infrastructure are healthy. Analytics tells you what the data means and what your team should do next.
Mage Pro Analytics dashboard with revenue, engagement, channel, regional sales, customer segment, acquisition funnel, and product performance charts.

​
Access Analytics

Select Analytics in the left navigation, or go directly to:
This is the Analytics workspace. From there, you can create dashboards, open existing dashboards, add charts, edit charts with AI, refresh data, and arrange the dashboard layout.

​
What Analytics helps you do

Analytics is built for fast, practical answers: Dashboards can include time series, comparisons, distributions, rankings, tables, heatmaps, funnels, scorecards, and other visual summaries. Start with the business question you want answered, and Mage helps turn the available data into a useful view.

​
Create a dashboard

  1. Open /apps/analytics/dashboards.
  2. Click New dashboard.
  3. Name the dashboard for the audience or decision it supports, such as Executive revenue, Customer health, Product adoption, or Data quality.
  4. Open the dashboard.
  5. Click Add chart to create the first visualization.
Keep dashboards focused. A good dashboard usually answers one type of question for one audience: executive reporting, growth analysis, customer operations, data quality, model review, or cost tracking.

​
Add charts with AI

Click Add chart, then describe the outcome you want. AI Sidekick uses the dashboard and the data you are working with to create the chart and save it to the dashboard. Useful prompts describe the metric, segment, time range, and decision:
You can also create a chart while reviewing pipeline results. Click the chart action from a supported result or preview, describe what you want to see, and Sidekick keeps the chart connected to the right data. For an existing chart, use Edit with AI to ask for changes:
  1. Change the metric, grouping, or time period.
  2. Add a comparison line or segment split.
  3. Rename the chart or clarify labels.
  4. Turn a trend into a ranking, scorecard, table, or heatmap.
  5. Simplify a noisy chart for an executive audience.

​
Choose the right data

Analytics can visualize the data your Mage project already works with: That means Analytics can cover both business data and workflow activity. For example, one dashboard can show revenue by customer segment, failed records by source, pipeline run duration, block run failures, and model quality by training run. When you build charts from pipeline or block outputs across multiple runs, Sidekick checks the output grain before choosing the chart. Run-summary outputs are useful for operational trends such as run volume, checked time, row counts, failures, and block duration. Business-date charts, such as deals by close date or invoices by paid date, need rows that represent the business records and a matching business date field. If the available output only contains run summaries or nested saved state, Sidekick should state that coverage before saving the chart. For pipeline and block output datasets, Analytics chart creation uses the real saved output data for the selected run or partitions. JSON-backed outputs use readable data.json backing data. Chart-backed datasets do not use sample_data, sample=True output reads, or sample_data.json preview artifacts to create or refresh charts. For large datasets, ask for the summary you need rather than every raw row. For example, ask for “weekly active users by plan for the last 12 weeks” or “top 20 accounts by expansion revenue” so the dashboard stays fast and readable. When a chart starts from a data preview, Analytics uses the preview source and query plan to build a chart-owned dataset snapshot; it does not use only the rows currently visible in the preview table. Analytics stores a bounded chart-ready snapshot for each saved chart. If Sidekick warns that a time-series or ordered chart may have reached its dataset preview limit, narrow the request, group the data into coarser buckets, or ask Sidekick to state the covered date or category range before using the chart for decisions.

​
Organize dashboards for the team

Each chart appears as a card on the dashboard. You can:
  1. Rename charts so viewers immediately understand the metric.
  2. Edit charts with AI as the question changes.
  3. Duplicate a chart to test another view without losing the original.
  4. Drag charts into a new order to create a clear story from top to bottom.
  5. Resize charts so the most important metrics get more space.
  6. Group related charts together so trends, drivers, and details are easy to scan.
  7. Remove charts that are no longer useful.
  8. Refresh charts when the underlying data has changed.
Use the top of the dashboard for the metrics that matter most. Put detailed breakdowns, tables, and diagnostics after the summary so viewers can move from signal to detail. Dashboard layouts are flexible. Drag a chart to rearrange the page, make headline metrics larger, give detailed tables more room, or place related charts side by side. This makes it easy to turn a collection of charts into a polished dashboard for a meeting, executive review, customer health check, or recurring operational workflow.

​
Refresh data

Analytics dashboards are meant to stay useful after the first version. When the source data changes, refresh the chart or dashboard so viewers see the latest available results. Charts with loaded data stay visible and dim while refreshing, so you can view the previous results until the new data is ready. A moving indicator marks the refresh; with reduced motion enabled, charts dim without animation. Refresh is useful for recurring reporting:
  1. Daily revenue and usage snapshots.
  2. Weekly customer health reviews.
  3. Data quality checks after each pipeline run.
  4. Model evaluation after new training or scoring jobs.
  5. Warehouse cost reviews at the end of a billing period.
If a refresh cannot complete, Mage keeps the previous dashboard view available so your team does not lose the last known good result.

​
Choose pipeline output partitions

For a chart backed by pipeline block outputs, open the chart’s Data source action to choose which partitions it reads when you refresh: Latest partitions and All partitions select partitions again on each explicit chart or dashboard Refresh, while keeping the chart’s saved query and transformations. Reloading the page alone keeps the saved chart snapshot. Partitions are ordered by run completion time, newest first. Retries for the same execution partition count as one partition. These settings apply only to charts backed by pipeline block outputs; other data sources keep their existing refresh behavior. Existing charts also keep their saved source behavior until you change this setting. Choose Latest partitions when each partition contains a new batch of records. If each run produces a full snapshot, combining partitions can repeat records; use a count of 1 to chart the latest snapshot. Selection is limited to 1,000 partitions and 10,000 scanned runs. If selection exceeds a safety limit, or a selected output is missing or unreadable, refresh fails explicitly instead of silently showing an incomplete selection. Narrow the partition selection or restore the missing output, then refresh again. All partitions does not remove the chart dataset’s separate row limits. You can also ask Sidekick to change a chart’s partition selection:
The saved source field is source_ref.partition_selection. Its mode is fixed, last_n, or all; last_n also requires a positive integer count, for example:

​
Executive dashboard

Use this for a high-level business review. Start with scorecards and simple trends, then add a few breakdowns that explain why the numbers changed. Example charts:
  1. Monthly recurring revenue.
  2. New, expansion, contraction, and churned revenue.
  3. Net revenue retention by segment.
  4. Top growing accounts.
  5. Accounts at risk.

​
Product dashboard

Use this for adoption, activation, and engagement analysis. Example charts:
  1. Weekly active users by plan.
  2. Activation funnel by signup cohort.
  3. Feature adoption by account tier.
  4. Retention by cohort.
  5. Most common user paths or events.

​
Data quality dashboard

Use this for operational trust in your pipelines and datasets. Example charts:
  1. Failed records by source.
  2. Validation errors by type.
  3. Freshness by table or pipeline.
  4. Duplicate records over time.
  5. Missing values by field.

​
Model quality dashboard

Use this for ML and AI workflows where stakeholders need to see model behavior over time. Example charts:
  1. Accuracy, recall, and precision by evaluation date.
  2. Prediction volume by class.
  3. Drift by feature.
  4. Confusion matrix summary.
  5. Experiment comparison by metric.

​
Best practices

  1. Start with the decision the dashboard should support.
  2. Use plain-language chart titles, such as Revenue by plan or Failed records by source.
  3. Group charts by story: summary, drivers, segments, and diagnostics.
  4. Keep raw-detail tables secondary unless the dashboard is for debugging or operations.
  5. Use AI edits to iterate quickly, then keep the clearest version.
  6. Create separate dashboards for separate audiences.
  7. Refresh dashboards before recurring business reviews.

​
Troubleshooting