Skip to content
Alation Agent Studio Documentation

Populate the Data Product Opportunities Inbox

The Alation Opportunities Inbox lets data stewards discover and claim data product candidates without running an agent themselves. This recipe uses the Data Product Recommender to fill that inbox automatically: it scans active datasources for high-value table clusters, validates them against BI usage and catalog metadata, and writes recommendation cards in a single unattended run. Stewards then review at their own pace and claim the ones they want to build.

Estimated cost<1ACUper run

This agent runs via a codemode skill pipeline. Inbox checks, datasource scans, lineage traces, evidence fetches, cluster synthesis, and recommendation writes all execute as internal skill functions — they are not individually metered as tool calls. Actual metered action consumption is near zero regardless of datasource scope or cluster count. 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 Recommender agentIdentifies clusters and writes recommendation cards to the Opportunities InboxScheduled flow (optional)Run the recommender on a recurring schedule

Build it

Section titled “Build it”

  1. Confirm usage data and BI metadata are populated

    The recommender grounds every card in evidence it fetches — it never invents usage. Verify that relational database query logs have been ingested for your target datasources and that at least one BI source has been cataloged.

  2. Run the agent

    Open the Data Product Recommender agent and describe the scope. Examples: “scan all active datasources for data product candidates” or “recommend data products for the SALES schema.” The agent runs the full pipeline without stopping to ask for input.

  3. Review the results

    The agent reports how many cards were inserted, how many already existed, and how many were skipped. Navigate to the Alation Opportunities Inbox to see the new cards with their confidence labels, evidence summaries, and suggested example questions.

  4. Have stewards claim and generate

    Stewards review the inbox, dismiss irrelevant candidates, and claim the ones they want. Claiming a card hands it to the Data Product Generator for supervised creation and enrichment. The recommendation card pre-populates the member tables and evidence — so the generation session starts with a richer proposal than starting from scratch.

  5. (Optional) Schedule it

    Wrap the agent in a scheduled flow to keep the inbox fresh automatically — weekly is a reasonable cadence for most data platforms.

Gotchas

Section titled “Gotchas”
  • The recommender never creates data products. If you want products created in the same session, use the Data Product Generator instead.
  • Cards from datasources with no query history produce bare skeleton products when claimed and need manual enrichment.
  • Candidates below the confidence threshold are silently skipped — the reported skipped count is the signal to check datasource metadata quality.
  • Gaps in BI coverage are noted in the evidence on each card — a steward can add the missing context when they claim it.

Variations

Section titled “Variations”
  • Scoped inbox fill — provide a specific datasource to restrict the scan to one source.
  • Read-only audit — run the agent with the message “read the current recommendations inbox and summarize what’s there” to audit inbox state without adding new cards.
  • Recommender + generator in sequence — run the recommender first to identify the most impactful clusters, then run the generator targeting the top-ranked candidates.
Section titled “Related”