Skip to main content

Understanding workflow architecture

AI Workflows are built around a structured automation architecture consisting of triggers, actions, executions and moderation layers.

Understanding these components helps you build scalable and controlled automation pipelines inside Elovate.

Core workflow structure

Every AI Workflow consists of:

  1. A trigger

  2. One or multiple actions

  3. Workflow executions

  4. Optional moderation

  5. Synchronization

These components work together to automate catalog operations.

Workflow

A workflow is the complete automation pipeline.

It defines:

  • which products should be processed

  • which AI actions should run

  • how outputs should be reviewed

  • how results move through the pipeline

A workflow can contain multiple connected actions running sequentially.

Example workflow:

  1. Detect products missing attributes

  2. Extract missing attributes

  3. Generate optimized content

  4. Translate content

  5. Validate output

  6. Synchronize changes

Trigger

The trigger determines when products enter the workflow.

Triggers use rule based conditions to select matching products.

Examples:

  • Category equals Running Shoes

  • Description is empty

  • Status is enabled

  • Brand equals Adidas

Only products matching the configured rules are processed.

Trigger types

One-time trigger

Processes all currently matching products once.

Useful for:

  • bulk enrichment

  • catalog cleanup

  • migration projects

  • initial translations

Continuous trigger

Continuously monitors the catalog and processes newly matching products automatically.

Useful for:

  • ongoing enrichment

  • automatic translations

  • automated QA

  • continuous synchronization pipelines

Actions

Actions are the AI powered tasks executed after a trigger is activated.

Each workflow can contain multiple actions.

Actions execute sequentially, creating a waterfall effect.

This means one action can continue based on the output or approval state of a previous action.

Available action types

Current workflow actions include:

  • Attribute Extraction

  • Content Enrichment

  • Translation

  • Category Mapping

  • Content Validation

  • Quality Scoring

Additional actions may be added over time.

Sequential execution

Actions run in order from left to right inside the workflow.

Example:

  1. Attribute Extraction

  2. Content Enrichment

  3. Translation

In this example:

  • extracted attributes become available for enrichment

  • enriched content becomes available for translation

This allows workflows to build progressively richer product data.

Workflow execution

When a workflow runs, Elovate creates tasks for all matching products.

Each task represents a product being processed through the workflow pipeline.

Executions can contain:

  • running tasks

  • waiting tasks

  • moderated tasks

  • completed tasks

  • failed tasks

The workflow overview provides visibility into execution progress.

Moderation layer

AI Workflows support moderation and approval flows.

Depending on the action configuration:

  • editors can review generated output

  • moderators can approve content

  • AI reasoning can be inspected

  • confidence scores can be evaluated

Moderation helps maintain quality control before data is synchronized.

AI reasoning and confidence scoring

Certain actions provide:

  • AI reasoning

  • confidence scores

Reasoning explains why the AI generated specific output.

Confidence scores indicate how certain the AI is about the generated result.

This helps teams:

  • review uncertain outputs

  • improve trust

  • identify products requiring manual attention

Scope and test products

Each action contains scope settings.

These determine:

  • which products the action runs on

  • which products are used during testing

Before running large workflows, you can test prompts on a small set of products first.

This helps validate prompt quality and output behavior before full execution.

Synchronization

Generated output is not always synchronized immediately.

Depending on your moderation setup:

  • results may require approval

  • products may wait for synchronization

  • editors or moderators may review content first

After approval, synchronized data is pushed back into your connected platform.

Workflow statuses

Workflows and actions can contain multiple statuses.

Examples include:

  • Waiting

  • Running

  • Awaiting moderation

  • Synced

  • Completed

  • Error

These statuses help monitor execution progress and identify bottlenecks.

Workflow overview

The workflow overview provides centralized visibility into:

  • active workflows

  • task counts

  • execution progress

  • moderation queues

  • completed actions

  • errors

This makes it easier to manage AI operations across large catalogs.

Example workflow architecture

Example:
A webshop wants to enrich cat food products automatically.

Workflow structure:

  1. Trigger selects products in category "Dry Cat Food"

  2. Attribute Extraction extracts:

    • Flavor

    • Life stage

  3. Content Enrichment generates:

    • Shopping descriptions

    • SEO titles

  4. Translation converts content to German

  5. Moderation reviews generated output

  6. Approved content synchronizes to Magento

This entire process runs inside a single AI Workflow.

Why workflow architecture matters

The workflow architecture allows teams to:

  • automate repetitive processes

  • scale AI operations safely

  • build reusable pipelines

  • improve content consistency

  • reduce manual work

  • control AI quality through moderation

Instead of isolated AI jobs, workflows provide a structured automation system for managing catalog operations at scale.

Related Articles
What are AI Workflows
Understanding workflow actions
Understanding the workflow overview
Running workflow actions
Workflow statuses explained
Did this answer your question?