AI-assisted mapping with Simba

Etlworks ships two AI-assisted paths for building a mapping. The newer and more flexible path is conversational: open Simba chat from any mapping editor — flat, nested, or X12 — describe what you want, and iterate until the mapping looks right. The older path is the Create Mapping button in the editor: a one-shot dialog with an optional prompt, available only when both source and destination are flat. This article covers both.

When should I use AI-assisted mapping?

• First-pass mapping with many fields — Simba gets the obvious matches done in seconds, and you fine-tune by hand or by asking Simba to refine.
• Renamed or restructured columns — Simba handles casing and naming variations (customerId ↔ customer_id) and can interpret hints like "exclude audit columns".
• Nested or X12 mapping — the conversational path through Simba chat works against all three editor types; the Create Mapping button does not.
• Iterating on a draft — ask Simba to change the mapping, then ask again. The conversation thread is the editing trail.

AI-assisted mapping through Simba chat (since 9.6.x)

This is the conversational path. It works in all three editors: flat mapping, nested mapping, and the X12 message designer.

Workflow

1. Open a transformation, click Mapping, and open the mapping editor.
2. Open Simba chat — the Simba icon in the app shell. Simba sees that you have a mapping editor open and uses the editor's current state as context.
3. Ask Simba to propose a mapping. Plain-English requests work:
   • "Suggest a mapping for these fields"
   • "Map the source to the destination but exclude any audit columns like created_at and updated_by"
   • "For each source field, find the closest match in the destination by name. Don't add new destination fields."
   • "Build a nested mapping that puts every customer's orders in an orders array on the destination."
   • "Convert the source field names from camelCase to snake_case in the destination."
4. Simba responds with a proposed mapping in chat. Review what it suggests.
5. Iterate. Ask Simba to change things:
   • "Use customer_id, not cust_id, for the join column."
   • "Drop the email mapping — we mask emails downstream."
   • "Add a calculated field that concatenates first_name and last_name into full_name."
6. Once you're happy, tell Simba to apply: "Apply this mapping" (or any phrasing — "go ahead", "do it", "save these changes" all work).
7. Simba writes the proposed mapping into the editor's draft. The mapping table updates.
8. Review the draft in the editor — edit, delete, or add rows by hand as needed.
9. Click Save on the mapping to persist. The mapping is not saved until you click Save explicitly — the Simba-applied draft can be discarded if you change your mind.

What Simba sees

When you ask Simba about the mapping, it reads:

• The list of source fields with names, types, and nullability flags.
• The list of destination fields with names, types, and nullability flags.
• The nesting level for each field (flat / nested level / X12 segment loop).
• The current state of the editor — what's already mapped, what's unmapped.
• Your chat turns — the conversation history during this thread.

Simba does not see raw data samples or mappings from other flows in your instance.

How "apply this mapping" works

When you ask Simba to apply, it calls an internal tool that writes the proposed mapping rows into the open editor's draft state — the same buffer the editor uses when you add rows by hand. No mapping is persisted at this stage. You see the proposed rows in the editor, can edit any of them, and only when you click the editor's Save button does the mapping become part of the flow.

AI-assisted mapping through the Create Mapping button

This is the older, simpler path. It runs as a one-shot dialog inside the mapping editor — no chat back-and-forth — and is available only when both the source and the destination are flat.

Workflow

1. Open the mapping editor.
2. Click Create Mapping in the editor toolbar.
3. In the dialog, select the Flat view and the Match destination fields mode. This is the only combination that invokes Simba.
4. Optionally enter a hint in the Optional prompt for AI box — e.g., "exclude created_at and updated_by".
5. Click Create Mapping. Simba proposes the mapping in a single call and the editor populates with the result.
6. Review and save.

For details, see Create Flat Mapping Automatically in Working with the Mapping Editor.

Other Create Mapping modes (no AI)

The Create Mapping dialog also exposes these non-AI modes:

Chat vs. button — which should I use?

Limits and costs

• New AI-assisted mapping consumes from the AI token allowance as Simba chat — see the AI usage dashboard for current consumption. Old does not.
• Schemas with hundreds of fields work but cost more tokens. For very large schemas, pre-filter with a hint such as "only fields starting with cust_".
• Simba does not see raw data — only the schema metadata. If you need value-driven mapping decisions (e.g., "this column always contains email addresses"), call those out explicitly in your chat turn.
• Programmatic access to the same mapping tools is available through the AI Agent API and the MCP server — the mapping_apply_operations tool is the same one Simba uses internally when you say "apply".

Related articles

• Working with the Mapping Editor — the editor where both paths run.
• Mapping Fields Between Source and Destination — the field-level mapping reference.
• Nested mapping — for JSON / XML / EDI mapping work.
• Simba — Etlworks AI Agent — what Simba is and what else it can do.