AI Transform Suggestions
Generate a ready-to-run sequence of transform operations from your table metadata. Review each one, then apply the set to your pipeline.
AI transform suggestions read your table's metadata and return an ordered set of operations you can apply to your pipeline. Each suggestion names the operation, explains why it was suggested, and states whether it targets data quality or feature engineering. Nothing is applied until you approve the set.
What It Uses to Decide
Suggestions are generated from your table's governed metadata, not from your rows:
- Column names and data types from the connected integration
- Column descriptions generated and validated in AI Column Insights
- Data quality statistics per column from DQ profiling: null counts, distinct counts, and type compliance
- Operations already in your pipeline, in order, so suggestions build on what you have rather than repeating it
- Only the columns you selected. Suggestions can reference those columns and the new ones they create, nothing else.
This is why suggestion quality tracks your AIR Score. A table with validated column descriptions produces operations aimed at what your data actually measures. A table with no descriptions produces generic cleanup.
The panel shows the source table's DQ score and AIR score before you generate, so you can judge suggestion quality in advance.
The Three Modes
Choose what the suggestion set should focus on.
Focuses on cleaning, null handling, type fixes, and deduplication.
Use this when your source table has known quality problems: blank values in key columns, dates stored as text, or duplicate rows from a repeated sync. Suggestions target the dimensions that move your DQ score.
No description field is shown in this mode. The suggestions come from the data quality statistics alone.
Engineers new columns, aggregations, and derived metrics.
Use this when the table is already clean and you want analysis-ready fields: a revenue tier derived from an amount column, a delivery duration from two timestamps, or a per-segment aggregate.
You can describe the features you want in 20 to 500 characters, or leave the field blank and let the AI infer the most useful flow from your table metadata.
Improves data quality first, then adds features.
Use this on a raw table you intend to take all the way to analysis. The returned set is ordered so cleaning operations come before the operations that depend on clean input.
The description field works the same as in Add Features: fill it in to steer the feature work, or leave it blank.
What a Suggestion Contains
A set contains 5 to 15 operations. Each suggestion in it has four parts:
| Part | What it tells you |
|---|---|
| Order | Its position in the suggested sequence, starting at 1. Operations are meant to run in this order. |
| Purpose | Either data quality or feature engineering, so you can see at a glance what each step is for. |
| Description | Why this operation was suggested, what it does, and the expected data quality or feature benefit. |
| Operation | The operation and its full configuration, ready to run. This is what gets added to your pipeline when you apply the set. |
Every suggestion is validated against the operation schema before it reaches you. A suggestion that does not resolve to a real operation with valid parameters is rejected server-side, so the set you review is always runnable.
How to Generate Suggestions
Open Transform Setup
Open a transformation with a source integration and base table selected. The suggestion panel reads the table you have already chosen and loads its metadata.
Choose what the suggestions should do
Select Improve Data Quality, Add Features, or Improve Data Quality + Add Features.
Describe the features you want
For Add Features and the combined mode, describe what you want to engineer. The field accepts 20 to 500 characters.
Leave it blank to let the AI infer the best transform flow from your table metadata alone.
Select a model and context mode
Choose an AI provider available on your plan. Then choose a context mode, which controls what is sent to the provider: whether real table and column names are used or anonymized, whether data quality statistics and frequent values are included, and whether your user role is shared.
See Privacy and Context Modes for what each mode sends.
Generate
Generation consumes one analysis credit from your workspace pool. The panel shows progress while it prepares table context and builds the suggestion set.
If your workspace is out of analysis credits, generation is blocked before the request runs. Purchase an add-on pack to continue.
Review and apply
Read each suggested operation and its rationale. Applying the set appends the operations to your pipeline in the suggested order.
Applying does not run anything. Use Save & Preview on each operation to see its effect on your data and its DQ delta before you run or schedule the pipeline.
What Suggestions Will Not Do
Suggestions work within the table you selected. Join and Concat are never suggested: combining tables depends on knowing how your data actually relates, and that is a decision you make rather than one inferred from column names.
Add a join or concat yourself from the operations list, then generate again. The AI reads it as an applied step and suggests what comes next, so the relationship stays yours to define and the suggestions build on top of it.
Everything else is available, including Code Editor. When no combination of no-code operations reaches your goal, a suggestion can contain a Python script, which you review in the same panel as any other step.
Reviewing and Applying a Set
The suggested operations appear in the left rail in their intended order. Select any one to see its full configuration in the panel: the operation, the columns it touches, and every parameter it will run with.
Applying adds the whole set to your pipeline. Individual suggestions cannot be excluded.
Steps in a set are written to depend on each other. A step that groups by a revenue band needs the earlier step that created the band. Because the set is built as one sequence, it is applied as one sequence.
If the set is not what you want, use Start Over and generate a different one rather than applying it and deleting steps afterwards.
Two details worth knowing while you review:
- If a suggestion exactly matches an operation already in your pipeline, it stays visible in the list but is not applied again. The count in the success message can be lower than the number of suggestions you reviewed.
- A set that ends in Edit Columns keeps only the columns that operation lists. Check that list includes the columns created by earlier steps, otherwise they are dropped from the output.
Applying does not run anything. Use Save & Preview on each operation to see its effect and its DQ delta before you run or schedule the pipeline.
Taking It Further
Once a set is applied, you can keep building in either direction: add operations yourself from the standard operations list, or generate another set.
Generating again after applying continues the pipeline. Your applied operations are sent as context, so the next set starts from the state your data is now in rather than repeating work. A first set that cleans and types a raw table is a reasonable place to stop, review, then ask for the feature engineering separately.
Each generation costs one analysis credit, including Start Over and every regeneration.
Generating again without applying works differently: the AI treats the current suggestions as a draft to improve and returns a full replacement set, renumbered from 1, rather than additional steps.
Starting Over and Exiting
Two controls manage a suggestion set in progress:
- Start Over clears the current stored suggestions so you can generate a fresh set, for example after changing the mode or the feature description. Generating again consumes another analysis credit.
- Exit suggestions leaves the flow and discards any stored suggestions that have not been applied.
Neither action affects operations already applied to your pipeline.
Plan Availability
AI transform suggestions are available on every plan, including the 14-day evaluation.
Two plan limits apply:
- Provider access follows your plan tier, the same as elsewhere in Edilitics. A provider not included in your plan cannot be selected for suggestions.
- Column limits follow your plan's transform column cap. A request covering more columns than your plan allows is blocked with an upgrade prompt.
Suggestions draw on the same analysis credit pool as AskEdi and dashboard Summarise.
Frequently Asked Questions
Related Docs
No-Code Operations
The 25 operations a suggestion set can draw from, with configuration reference for each.
AI Column Insights
Generate and validate the column descriptions that determine suggestion quality.
AI Readiness (AIR) Score
How column documentation and data quality combine into the score shown before you generate.
Analysis Credits
How the shared credit pool works across AskEdi, transform suggestions, and dashboard Summarise.
Need help? Email support@edilitics.com with your workspace, job ID, and context. We reply within one business day.
Last updated on
Hybrid Pipelines
Combine no-code operations and Python Polars scripts in a single pipeline. Each step passes its output to the next regardless of type.
Scheduling
Schedule Edilitics Transform pipelines to run once, daily, or on custom cron expressions. Pause, resume, or edit schedules anytime with timezone-awareness.