Skip to main content

Proposals vs workflows: which one to use

When to let the AI propose a direct change in the AI queue, and when to set up an AI workflow instead.

Through MCP, AI changes reach Elovate in two ways: as a proposal in the AI queue, or as the result of an AI workflow. Both are reviewed by a person before anything is synced. Which one you use depends on the size and type of the job.


Proposals: small, targeted changes

Use a proposal when you know which items need to change and why.

  • A better meta title for 10 specific products.

  • Correcting a wrong fact in one description.

  • Adding products to a category.

  • Translating the option labels of one attribute.

  • A new FAQ page or blog post.

The assistant writes the new value itself, adds a reason, and an administrator approves or declines each proposal in the AI queue.


Workflows: the same job on many items

Use a workflow when the same kind of change applies to hundreds or thousands of items, or should keep running for new products.

  • Meta descriptions for every product that doesn't have one.

  • Extracting attributes from descriptions across a whole category.

  • Translating all descriptions to a new store view.

The assistant sets up the workflow and its cards, and can test them on a few products. A person starts the workflow in Elovate. Each result passes an editor and a moderator before it is synced.


At a glance

Proposal

Workflow

Best for

A handful to a few hundred named items

Large scopes and recurring work

Who writes the text

The assistant, in the conversation

The workflow's AI action, using a fixed prompt

Review

Administrator in the AI queue

Editor, then moderator

Repeatable

No, one-off

Yes, also for new products

Chaining steps

No

Yes: extraction → enrichment → translation

Started by

Created directly by the assistant

A person in Elovate


Tip: many items with the same text

Catalogues often repeat texts, for example every colour or size of the same product. Instead of rewriting or translating each item separately, the assistant can group items that share a text, write or translate that text once, and propose it for the whole group. Only items with a unique text get their own proposal. This is faster and keeps texts consistent.

Did this answer your question?