Request a call

Human-in-the-loop (HITL)

Definition

Human-in-the-loop is a workflow design where a person reviews, corrects or approves an AI system's output at set points before it reaches a customer or a system of record.

Updated

Also called human review.

How it works

The workflow pauses and hands a task to a person, with everything they need to decide in one view: the AI's draft or extracted values, the source it used, and options to approve, edit or reject. Their decision resumes the workflow.

Where the pause goes is the real design decision. Common triggers:

  • Low confidence, such as an invoice field the extractor scored below a set threshold.
  • High stakes: refunds over a set amount, anything sent to a regulator, contract changes.
  • Novelty: a request type the system has not seen before.
  • Sampling: a random share of routine items, so quality is still measured when nothing is flagged.

In n8n, a workflow can post an approval request to Slack or Gmail and wait for the reply before it continues, and its human-in-the-loop for tools setting holds an AI agent's tool call until a reviewer approves it. Approvals can also live in your CRM or helpdesk as a task.

How businesses use it

Human review is how most companies start with AI in a live process. A finance clerk approves extracted invoice lines before they post to the ledger. A sales rep edits the follow-up email an agent drafted after a call, then sends it.

Each correction is also data. Logged edits show where the model is weak and become test cases for LLM evaluation. Once accuracy on a category holds up for several weeks, you can raise the share of items that skip review.

Where it goes wrong

Review fatigue is the usual failure. If a person approves 200 items a day and nearly all of them are fine, they start clicking approve without reading. Keep queues short by routing only what needs judgment, and show the reviewer exactly what the model was unsure about.

The other failure is a queue with no owner. Name the person or role who reviews each queue and set a response time, or items sit for days.

And if every item needs a person anyway, ask whether the AI step is earning its cost. A well-built form or a plain rule may do the job for less.

Related terms

  • AI guardrails

    AI guardrails are checks placed around a language model that block, rewrite or escalate inputs and outputs that break your rules on safety, accuracy, privacy or scope.

  • AI agent

    An AI agent is a language model that chooses and calls tools in a loop, reading each result, until it finishes a task or hands off to a person.

  • Intelligent document processing

    Intelligent document processing uses OCR and AI models to read documents such as invoices, contracts and forms, extract the fields you need, check them and post them into your systems.

  • LLM evaluation

    LLM evaluation scores a language model system's outputs against a fixed set of real test cases, so you can tell whether a change made it better or worse.

  • n8n

    n8n is a workflow automation platform, self-hosted or cloud, that connects apps and APIs in visual workflows and supports custom code and AI agent steps.

All 30 definitions are in the AI and search glossary.

Find out where this fits in your own systems

Bring one process or page to a 30-minute call.

Request a 30-minute scoping call

We reply within 1 working day and suggest times that suit you.