Skip to main content

Enrichment

Enrichment improves store product content with AI-generated text, images, and classification tags. The service supports one-time actions, reusable jobs, provider configuration, and usage reporting.

Dashboard Areas​

AreaPurpose
InstantSelect products and enrichment outputs for an immediate run.
JobsDefine reusable or scheduled enrichment rules, select a prompt, and inspect executions.
PromptsDefine reusable instructions, output language, and editable product fields.
AI ProvidersConfigure a merchant-owned OpenAI-compatible text and image provider.
UsageReview token, image, estimated cost, provider, status, and error history.

Instant Enrichment​

Use Instant when an operator needs a controlled, reviewable change for selected products. Choose the product set, provider, and outputs before starting the run. Review generated content before it becomes part of a production catalog workflow.

Typical outputs include:

  • Product descriptions and summaries.
  • Search or merchandising text.
  • Product images when an image model is configured.
  • Classification and hyperpersonalization tags.

Enrichment Prompts​

Create multiple prompts under Store > Enrichment > Prompts. Each prompt has a name, instructions, optional output language (the store language is used when blank), active status, and an explicit set of editable fields. A job selects one prompt and reads its current active configuration on every run. Editing a prompt changes later runs; deactivating it pauses active jobs that use it. Existing jobs without a prompt retain the default instruction and field selection. Prompt selection requires text enrichment.

A prompt can classify products or change only selected fields: title, description, tags, attributes, category, categories, otherDetails, metadata, variants, sku, externalId, productUrl, price, currency, status, inStock, inventoryQuantity, imageUrl, and imageUrls. Price, inventory, status, identifiers, and image URLs affect commerce, so select them only when the job explicitly needs those changes. The service validates selected output types and values before saving; it reserves enrichment audit metadata. Generated images remain a separate job option.

Enrichment Jobs​

Jobs are appropriate for repeated or larger workflows. A job records its trigger, status, processed count, failed count, skipped count, error, and execution times.

Use jobs when:

  • New products should be enriched after synchronization.
  • A catalog segment needs a consistent content policy.
  • Products missing selected fields should be processed in batches.
  • Teams need execution history instead of one-time manual actions.

AI Providers​

Register an active merchant-owned OpenAI-compatible text and image endpoint. Realm runtime AI providers are configured separately and are not available for merchant Enrichment jobs.

Provider configuration can include:

  • Provider name and description.
  • Base URL.
  • Text and image model identifiers.
  • Input, output, and image pricing metadata.
  • Default-provider selection.

Store provider secrets securely and grant access only to administrators who operate AI integrations.

Usage and Cost Review​

The Usage page reports input tokens, output tokens, generated images, estimated cost by currency, provider, trigger, product, run status, and recent failures. Compare consumption with the configured merchant limits before scheduling large jobs.

Usage Scenarios​

  • Add descriptions to imported products that only contain names and SKUs.
  • Generate consistent merchandising text for a new category.
  • Create tags that feed Hyperpersonalization campaigns.
  • Compare managed and customer-provided models on a pilot product set.
  • Detect repeated provider errors before a scheduled job processes the full catalog.
  • Track cost by provider during a model migration.

Operating Checklist​

  • Start with representative pilot products.
  • Keep the original merchant data available for rollback.
  • Review generated claims, prices, and regulated language.
  • Set a default provider deliberately.
  • Confirm provider rate limits before scheduling jobs.
  • Monitor failed and skipped counts after every large run.
  • Review usage and estimated cost regularly.