Skip to content
AI

Human-in-the-loop

Human-in-the-loop means an action proposed by AI is applied only after a person approves it, so the last word on critical steps stays with the team.

Human-in-the-loop is a rule for dividing work between AI and people: the AI proposes the action, the person decides, and the action is applied after that. Preparation and analysis go to the machine, and the step that carries responsibility stays with a person. For a store owner this means the AI does not change a budget, a customer record or a rule on its own.

Approval does not have to weigh the same at every step. Adding a note and changing an ad budget do not carry the same risk. A well-built system says up front which actions go for approval, lets you edit a proposal before approving it, and writes down what was done. That way the team keeps its speed and can see later what changed and why.

Why It Matters

  • Even if the AI makes a wrong proposal, a person sees it before the action is applied.
  • For actions that are hard to undo, such as budgets and customer records, control stays with the team.
  • Every approved action is logged, so the team can trace what was done, when and with whose approval.

How It Works

  1. The AI makes a proposalA stage change, note, assignment or budget proposal is prepared. The action has not been applied yet.
  2. Proposals are collected in one placePending proposals show up in a single place, and the person reviews them all from there.
  3. The person decidesThey approve the proposal as it is, reject it, or correct its text or value first.
  4. The action is applied and loggedOnly an approved action is applied. What was done is written to the log.

Example

At Liva Butik, Copilot proposes raising the daily budget of an Instagram ad. The budget proposal reaches Selin Kaya as a suggestion only; the budget does not change until she approves. Selin finds the suggested amount too high, lowers it and approves. The same day Copilot also proposes assigning a task for Mert K.'s sales opportunity. Selin assigns it to Burak Demir and approves. Both actions are logged. Had she rejected a proposal, nothing would have changed.

Not to Be Confused With

Human-in-the-loop does not mean every message is typed by hand. AI Employees carry part of the conversation with customers; approval is a gate for critical actions. Fully autonomous marketing is a vision, and today critical actions run with human approval.

In Creato

In Creato, critical actions run with human approval today.

  • You approve or reject Copilot's stage change, note, assignment and task proposals in the Onay Masası (approval desk), and you can edit a proposal before approving.
  • Meta budget suggestions are only suggestions; the budget does not change until you approve.
  • Rules and frequently asked questions extracted from your website are not used by an AI Employee until you approve them.
  • In an MCP connection, you choose the tool's approval level.
Explore Copilot

Frequently Asked Questions

Does human approval slow the AI down?

It slows the steps that need approval a little, and that is a deliberate choice. Not every action has to go for approval; routine steps move along inside the conversation. Critical actions are collected in one place, so the team reviews what is waiting in order. The price of the slowness is that a wrong change is noticed before it is applied.

When will fully autonomous operation arrive?

No date is shared for that; until then the decision stays with the team. That is why approval steps are designed up front: a proposal can be edited, and a rejected proposal changes nothing. For the wider picture, see the agentic AI page.

Can I look back at an action I approved?

Yes, applied actions are kept in a log. Who approved what and when stays in the record, and you can trace where a change came from. A rejected proposal changes nothing because it was never applied. Copilot presents its proposals on this logic.

Related Terms