How a Mammoth pipeline works: add and preview tasks

What a pipeline is and how tasks run

Each task is a separate, reversible operation applied to the rows of your View. Tasks run in order from top to bottom, and the output of task N is the input to task N+1. The first item in the Pipeline panel is always Source, the raw data coming in. The Source card appears as soon as you add your first task to a fresh View.

RAW DATA
   ↓
[Task 1: Remove Duplicate Rows]
   ↓
[Task 2: Conditional Filter: keep rows where Status = "Active"]
   ↓
[Task 3: Edit Cells: standardize Region column values]
   ↓
[Task 4: Join: merge with budget data from another dataset]
   ↓
CLEAN OUTPUT

Tasks are stored as instructions, not copies of the data. You can reorder, edit or delete any task without losing your source data. See Edit, reorder and copy pipeline tasks.

Pipeline tasks:

• run automatically every time new data flows through the View;
• apply to all the data, not only the rows you can see;
• run in the order they appear in the Pipeline panel, top to bottom;
• are part of automation, so they run when you schedule a pipeline;
• keep a complete history in the Activity Log.

All 30+ transformation functions, plus Data Check and Checkpoint, are pipeline tasks.

Pipeline tasks versus Display Changes

There are two separate ways to change a View:

There is no "Rename Column" pipeline task. To rename a column, double-click its header or choose Rename Column in its 3-dot menu. That creates a Display Change, and the new name appears in pipeline task dropdowns and exports. See Rename, format and sort columns.

Open the Transform menu

1. Open any View.
2. Click the Describe your transformation box in the action bar at the top of the View, or the blue + button next to it (tooltip "Add a task").
3. The Transform menu opens on AI Powered, which lists five functions. A left panel lists the seven categories.

Type in the Describe your transformation box to filter the functions to those matching your keywords. The Ask in plain words result always appears and carries your typed text into its prompt.

Once the pipeline has tasks, you can also click a + button in the Pipeline panel: below the last task card, or between two cards to insert a task at that position. It opens the same Transform menu. Cards in an error state have no insert button.

For what each function does and which page covers it, see Every function in the Transform menu.

Describe a transformation in plain English

Some functions build the task from a plain-English description:

• Generative AI writes new column values with a large language model (classification, enrichment, translation, summaries). See Use AI on your columns.
• Ask in plain words turns a plain-language request into a SQL query and runs it on your data. See Write SQL with AI.
• Conditional Filter and Extract Text have an AI generation bar: describe the rule and Mammoth builds it.
• Bulk Replace uses AI to suggest groups of similar values.

Examples:

• "Remove all rows where the customer name contains a test email address" (Conditional Filter)
• "Extract the numeric dollar value from the Notes column" (Extract Text)
• "Classify this product description as Electronics, Apparel, Home, or Food" (Generative AI)

AI-generated logic produces the same reusable, auditable tasks as manual ones. You can inspect, edit or override it before applying.

Add a task to the pipeline

Every function in the Transform menu follows the same four steps.

1. Open the Transform menu (see above).
2. Click a category to browse its functions, or type in the search bar. Click the function you want.
3. Fill in the configuration panel that opens at the bottom of the View: column selectors, operators, values, and where to write the result.
4. Click Apply. With Draft mode off, the task runs at once and the grid updates. With Draft mode on, choose Add to Pipeline: the task is queued and runs when you click Apply now.

The new task card shows the function name, its key settings and the row count after it runs.

While a configuration panel is open, the pipeline shows a "New task will be inserted here" placeholder card. It marks where the new task will appear when you click Apply.

Preview a task before you apply it

Most functions show a live preview of their result in a window above the panel, before you click Apply. The preview never changes your data.

The preview opens once the settings are complete. Until then it reads "Complete the settings to see a preview." It updates as you change the settings ("Updating…").

To work out the result quickly, the preview reads a sample of the View. A line under it says how much it read. It reads "Uses the first" followed by the sample size and the View's row count, or "Uses all" followed by the row count when it read every row. With a sample, counts in the preview describe the sample, not the whole View, and the summary says so.

Turn the preview on or off

The panel header has a Preview control with an eye icon:

You can move, resize, minimize, expand and close the window. Keyboard users can move or resize it with the arrow keys; Home puts it back to its default place and size. Double-click the title bar to expand it.

What each preview tab shows

Result is always there. The other tabs appear when they apply to the function.

Functions that preview on request or not at all

• Fill Missing Values with AI fill takes time, so its preview waits for you: "This function takes time to run. Click Run preview to see its result." Click Run preview.
• Send to Dataset, Bulk Replace, the Export step, Data Check and Checkpoint Alert have no live preview.
• Generative AI has its own Preview button instead.

Apply several tasks at once with Draft mode

Draft mode controls when tasks run. Turn it on from More Pipeline options (...) at the top of the Pipeline panel.

With Draft mode on, the pipeline header shows Draft mode ON. Tasks you add or change are queued and don't run. The function panel's split button defaults to Add to Pipeline. The grid keeps showing the state from the last time the pipeline actually ran.

Review, apply or discard pending changes

When changes are queued, a bar at the bottom of the panel reads "Pipeline has Pending Updates" with three options:

The Draft mode toggle is disabled while the pipeline is in an error state. Apply now stays disabled until the failing tasks are fixed.

If another user already has unpublished changes on the pipeline, an "Adding to another user's draft" warning explains that your task joins their draft. It doesn't go live until the draft is published. Choose Add anyway to continue.

Use a parameter as a value in a task

Some value boxes accept a project parameter instead of a typed value. When the parameter's value changes, the step uses the new value on its next run. You manage parameters in Parameters and snippets.

Conditions (Add Condition and Conditional Filter), Data Check thresholds and the SQL mode of Math Function (Insert Parameter) can use parameters too.

To bind a value:

1. Click the small chip at the left of the value box. It reads Val for a typed value.
2. Choose Parameter ("Bind to a project parameter"). The chip changes to #.
3. Click Select a parameter and pick one. Only parameters of a matching type are listed.
4. The box now shows the parameter's name. Click its X to remove the binding.

You can't apply the step while the box is set to Parameter with nothing picked ("Select a parameter to continue"). To go back to a typed value, click the chip and choose Value ("A fixed value").

When someone changes a parameter's value, every pipeline that uses it is flagged as stale. Mammoth doesn't rerun them on its own: rerun from the parameter library or the stale banner. The rerun starts from the first task that uses the parameter.

Put tasks in the best order for speed

Apply filter tasks early to cut the row count before costly steps like joins and pivots. For best performance, order a pipeline like this:

1. Validate early: Data Check tasks near the top.
2. Filter early: Conditional Filter to reduce row volume.
3. Clean columns: Remove or Convert Column Type.
4. Standardize: Bulk Replace, Text Formatting.
5. Enrich: Join, Lookup, Add Column.
6. Calculate: Math Function, Window Function.
7. Format: Label & Insert Values, Combine Columns.
8. Govern: Checkpoint before export.

For large datasets, use Draft mode to batch-test changes before applying. Use an automation to keep source datasets current rather than re-uploading by hand. See Speed up pipelines and connectors and What you can automate.

Troubleshooting the task preview

For a task that fails after you apply it, see Fix a failing step.

Related

• Edit, reorder and copy pipeline tasks: suspend, copy, search and restore earlier versions.
• Quick start: from a file to a shared dashboard: build a first pipeline end to end.
• Rename, format and sort columns: change how data looks without adding tasks.
• Column types, IDs, currency and percentages: the warnings some tasks show about column meaning.
• Pause sends while you edit: hold dashboards and exports while you change the pipeline.
• Join two datasets on a shared column: combine data from two or more datasets.