Human in the loop (HITL) is a way of running an AI system where people review, approve, or correct its work at set points. In customer service, it usually means a person approves certain actions, reviews samples of conversations, or takes over when the AI agent escalates.
Much of the design work is choosing where people step in. Review on every step can cancel out the time the AI saves, and no review on high-stakes steps lets errors reach customers.
Four forms of human in the loop
Approval before an action. The agent prepares the action, and a person confirms it before it runs. Typical cases are a large refund, a hardship arrangement, or a change of payout account.
Review after the fact. People check a sample of finished conversations for accuracy, tone, and compliance. Quality assurance teams already work this way with human agents.
Escalation. The agent hands the conversation to a person when a rule or signal says so. Examples are a complaint, a vulnerable customer, or a request outside its scope. A good handoff carries the context, so the customer does not repeat themselves.
Feedback into the system. Reviewers' corrections become new test cases, rule changes, or knowledge updates.
Day-to-day teamwork between AI agents and people is covered under human-AI collaboration. HITL is the oversight part of that relationship.
Human oversight in the EU AI Act
Article 14 of the EU AI Act, Regulation (EU) 2024/1689, covers high-risk AI systems. They must be designed so that natural persons can oversee them effectively while in use. The measures must be "commensurate with the risks, level of autonomy and context of use".
Article 14(4) lists what the people assigned to oversight must be enabled to do. They should understand the system's capacities and limits, stay aware of automation bias, and interpret its output correctly. They should also be able to disregard, override, or reverse the output, and to stop the system in a safe state.
Article 26(2) adds that deployers assign oversight to people with "the necessary competence, training and authority". Annex III lists the high-risk uses. For banks and insurers, it names creditworthiness and credit scoring, and risk assessment and pricing in life and health insurance.
GDPR Article 22 and automated decisions
GDPR Article 22 gives people the right not to be subject to a decision "based solely on automated processing". The right covers decisions with legal or similarly significant effects. Such decisions are allowed when necessary for a contract, authorized by law, or based on explicit consent.
In the contract and consent cases, the controller must offer at least "the right to obtain human intervention". People must also be able to express their view and contest the decision. In a contact center, that can touch declining a payment plan, rejecting a claim, or closing an account.
Making oversight work
Watch for automation bias. NIST's Generative AI Profile (July 2024) describes it as "excessive deference to automated systems". A reviewer who approves hundreds of suggestions a day may stop reading them closely. It helps to track how often reviewers change what the AI proposed.
Give reviewers the evidence. A reviewer needs the conversation, the data the agent used, and the rule that applied. A record in the audit trail can supply all three.
Write the triggers as rules. Escalation thresholds written as explicit rules behave the same way on every call, and an auditor can read them.
Frequently asked questions
What is an example of human in the loop in customer service?
A collections agent agrees on a payment plan with a customer within limits set by rules. Plans outside those limits, or customers who mention hardship, go to a person who decides. Supervisors also review a weekly sample of calls.
What is the difference between human in the loop and human on the loop?
In common usage, human in the loop means a person takes part in a decision, for example by approving it. Human on the loop means a person monitors the system and can step in without approving each action. Many contact centers use both, for different steps.
Does the EU AI Act require human oversight for customer service AI?
Article 14 applies to high-risk AI systems. Annex III lists credit decisions and life and health insurance pricing among them, and customer service is not listed as such. An agent that feeds those decisions may be covered, so a legal review per use case is worth doing. Separately, Article 50 requires people to be told when they are talking to an AI system, unless that is obvious.
Does GDPR Article 22 ban automated decisions?
No. It sets a default right not to be subject to solely automated decisions with legal or similarly significant effects. It then allows such decisions in three cases. In the contract and consent cases, people keep the right to human intervention, to give their view, and to contest the decision. Answering a routine question usually falls outside it.
How we approach this at Zowie
Zowie's AI agents work on voice, chat, and email, with over 100 million conversations a year across our clients. The language model speaks. A separate, deterministic Decision Engine returns every amount, rate, offer, disclosure, and escalation from your rules, with an audit log. To map where your team should step in, you are welcome to book a demo.
Explore: Human-AI collaboration, Intelligent handoff (AI to human), AI compliance & governance, AI audit trail, AI agent evaluation, Quality assurance