Workflow
    catalog analysisseoai suggestionshuman reviewrisk gateauditrevertkilledShopify AI Product Optimizer

    Shopify AI Product Optimizer catalog review loop (killed experiment)

    Historical process model for the killed Shopify app experiment: deterministic catalog evidence identified field-level gaps, merchant context bounded generated alternatives, risk and human review controlled permission to act, and snapshots plus change logs kept applied work reversible.

    Part of project

    Workflow

    Shopify AI Product Optimizer catalog review loop (killed experiment)

    11 steps
    1. Load the merchant context artifact: provider, model, brand voice, target audience, primary language, automation mode, and catalog criteria
      action
    2. Fetch the current Shopify product and preserve title, description, SEO fields, handle, type, vendor, tags, options, variants, and a revertable snapshot
      action
    3. Run weighted deterministic checks and write the SEO score, passed checks, failed checks, and field-level issue artifact before any model call
      subprocess
    4. Decide whether evidence-backed issues remain; stop without generation when the product passes the relevant checks
      decision
    5. For each affected field, combine current product facts, failed-check evidence, merchant context, and field constraints into a structured suggestion and rationale
      subprocess
    6. Classify the proposal by mutation risk: low for SEO metadata/tags, medium for visible title/description, and high for URL handles
      decision
    7. Human operator reviews current value, suggested value, rationale, risk, and brand fit; approve a subset, reject it, rewrite it, or stop the run
      decision
    8. If automation is enabled, decide whether trigger, criteria, billing cap, mode, and field allowlist permit action; otherwise preserve the suggestion as pending review
      decision
    9. Queue only approved or explicitly allowed suggestion IDs for Shopify mutation; retain rejected, disallowed, failed, and pending outcomes with their exact status
      subprocess
    10. Write before/apply evidence to the ChangeLog, verify the Shopify result, and use the saved snapshot plus conflict check to produce revert proof when restoration is needed
      action
    11. Killed-state boundary: retain the source, fixture screenshots, workflow model, validation, dry-run, and apply provenance without representing this Shopify automation as currently operating
      state