Falnor / Guides

What is human-in-the-loop automation, and where to put the person

Human-in-the-loop automation means software does the routine work while a person reviews or approves the steps that matter. It gives you the speed of automation without handing over decisions you cannot take back.

Book a discovery call

What human in the loop means

There are three common setups, and the names are easy to confuse.

Human in the loop: the system prepares the action and waits for a person to approve it before anything happens.
Human on the loop: the system acts on its own while a person monitors it and can step in or reverse a decision.
Human out of the loop: the system acts with no one watching. In business operations this is rare, and sensible only for low-stakes steps that are easy to undo.

When to use each one

Many systems use all three, depending on the step. The skill is deciding which step gets which.

Use in the loop for payments, changes to bank details, claim decisions, customer refunds and anything hard to undo.
Use on the loop for routine items the system has already proven it handles correctly, such as matching an invoice that agrees exactly with its purchase order.
Treat out of the loop as the exception, and only for steps where a mistake is cheap and reversible.

Examples from finance, claims and customer work

Supplier invoices

An invoice that matches its order within your limits can move on automatically. One with an unusual amount or a changed bank detail goes to a named approver with the evidence beside it. See accounts payable automation.

Claims

Checking that the documents are complete can run on its own. A large, unusual or disputed claim goes to a handler to decide. See claims processing automation.

Customer requests

A routine answer can be drafted from the account history. A refund or an exception waits for a person. See customer service automation.

How to design the approval step

Decide what triggers a stop

Use clear rules, such as an amount over a limit or a field that changed, rather than a feeling that something looks odd.

Show the evidence next to the decision

An approver who has to hunt for context will either slow down or approve without checking.

Name the approver

Send it to a person, not a shared inbox that nobody owns. Say who covers when they are away.

Record every approval

Keep what the system looked at, which rule applied and who said yes, so you can explain it later to a manager or an auditor.

Review and loosen over time

Look at what gets approved. If a type of item is always approved, it may be ready to run on its own.

Common mistakes

Asking for approval on everything, so approvers click through without reading.
Showing the decision without the evidence behind it.
Routing approvals to a team inbox with no owner.
Never revisiting the rules, so the people stay in the loop long after the system has earned trust.

Questions people ask

What is the difference between human in the loop and human on the loop?

With a person in the loop, the system waits for approval before it acts. With a person on the loop, it acts first and a person monitors and can step in.

Does human approval slow everything down?

Only the items that need it. Routine, well-understood items run on their own, and the unusual ones wait for a person with the evidence in front of them.

How do you decide what needs approval?

We map how the work really runs, then set rules for what counts as important, unusual or hard to undo. A system starts read-only and earns the right to act on a task once it is right at least 9 times in 10 on your real cases.

Is human oversight required by regulation?

Some rules call for it. The EU AI Act, for example, expects human oversight of high-risk AI systems. We are not a law firm, so check what applies to your business with your own adviser.