Skip to content
Alation Agent Studio Documentation

Generate Data Products from Usage Evidence

Most data teams have high-value table clusters that analysts query every day — but they’ve never been formally defined as data products, so they can’t be queried in chat or surfaced in the Data Marketplace. The Data Product Generator finds those clusters automatically from relational database query usage, BI usage, and catalog metadata, proposes a ranked shortlist, and — after you approve — creates and fully enriches each product so it’s chat-ready on day one.

Estimated cost<1ACUper run

This agent runs via a codemode skill pipeline. Catalog lookups, lineage traces, column profiling, product creation, eval case generation, and promotion all execute as internal skill functions — they are not individually metered as tool calls. Actual metered action consumption is near zero regardless of cluster count or table depth. At 0.25 ACU per metered action; confirm actual spend on the Usage page.

What you’ll use

Section titled “What you’ll use”
Data Product Generator agentIdentifies clusters, proposes products, creates and enriches drafts

Build it

Section titled “Build it”

  1. Confirm usage data and BI metadata are populated

    The generator grounds every proposal in evidence it fetches — it never invents usage. Before running, verify that relational database query logs have been ingested for your target datasource and that at least one BI source has been cataloged. Datasources with no query history will produce proposals based only on catalog metadata.

  2. Start the agent and agree on scope

    Open the Data Product Generator agent and describe your target. The agent will propose its default pipeline and ask for the datasource or schema if you haven’t named one. Accept the default or steer it toward a specific table set, a BI dashboard’s upstream tables, or enrichment of an existing product.

  3. Review the ranked shortlist

    The agent presents a ranked shortlist: each cluster with its name, member tables, supporting evidence, and the BI reports that consume it. Approve the full batch, select a subset, or ask to work through them one at a time.

  4. Review and approve metrics

    For each cluster, the agent proposes KPI metrics inferred from recurring query patterns. Metrics are not added until you explicitly approve them — review each proposed expression and accept or adjust.

  5. Review and approve eval cases

    After creating a draft product, the agent proposes a set of question/SQL eval pairs that test whether the product can answer its example questions. These are created and run only after your approval. Failures surface spec problems before you promote to ready.

  6. Promote to ready

    Once all enrichment items are populated and the eval suite has passed, the agent asks you to promote the version to ready. Only then does it become queryable in chat. If anything is incomplete it stays DRAFT — the agent never auto-promotes.

Gotchas

Section titled “Gotchas”
  • The generator uses the database connection URI from the catalog configuration. A mislabeled or stale URI breaks chat querying — confirm it in the catalog before promoting to ready.
  • The agent checks for existing products before creating. If a similar product already exists, it will ask whether to create a new one or augment the existing draft.
  • Metrics and eval cases require explicit approval — the agent waits at each gate.

Variations

Section titled “Variations”
  • Read-only advisor — clone the agent and remove the write capability to get proposals and enrichment previews without any writes to the catalog.
  • Single product — name a specific table set at the start; the pipeline scopes to those tables and produces exactly one cluster proposal.
  • Bulk unattended — use the Data Product Recommender to populate the Opportunities Inbox automatically, then claim and generate from there.
Section titled “Related”