{"id":10118,"date":"2026-10-19T15:16:25","date_gmt":"2026-10-19T13:16:25","guid":{"rendered":"https:\/\/www.trusteddecisions.com\/?post_type=blog-new&#038;p=10118"},"modified":"2026-10-19T16:42:55","modified_gmt":"2026-10-19T14:42:55","slug":"human-in-the-loop","status":"publish","type":"blog-new","link":"https:\/\/www.trusteddecisions.com\/en\/blog\/human-in-the-loop\/","title":{"rendered":"Human in the Loop: Why AI Decisions Need Human Approval"},"content":{"rendered":"\n<style>\n.td-summary-box { --td-gold: #9E7D42; --td-gold-light: #C9A75F; --td-dark: #3C4B53; --td-text: #555555; background: #FFFFFF; border-left: 2px solid var(--td-gold); border-bottom: 2px solid var(--td-gold); border-radius: 12px; box-shadow: 0 15px 40px rgba(0,0,0,0.08); padding: 40px 44px 34px; margin: 40px 0; width: 100%; max-width: 100%; font-family: \"Open Sans\", -apple-system, \"Helvetica Neue\", Arial, sans-serif; box-sizing: border-box; }\n.td-summary-box h2.td-summary-title { display: block; font-family: \"Open Sans\", -apple-system, \"Helvetica Neue\", Arial, sans-serif; color: var(--td-dark); font-size: 1.35rem !important; font-weight: 700; text-transform: uppercase; letter-spacing: 0.5px; line-height: 1.3 !important; margin: 0 0 26px; padding: 0 0 12px; position: relative; }\n.td-summary-box h2.td-summary-title::after { content: \"\"; position: absolute; left: 0; bottom: 0; width: 70px; height: 3px; background-color: var(--td-gold); border-radius: 2px; }\n.td-summary-box ul { list-style: none !important; margin: 0 !important; padding: 0 !important; }\n.td-summary-box ul li { position: relative; padding: 0 0 0 38px !important; margin: 0 0 16px !important; color: var(--td-text); font-size: 1rem; line-height: 1.7; list-style: none !important; }\n.td-summary-box ul li:last-child { margin-bottom: 0 !important; }\n.td-summary-box ul li::before { content: \"\" !important; position: absolute; left: 0 !important; top: 7px !important; width: 25px !important; height: 15px !important; border: none !important; border-radius: 0 !important; transform: none !important; background: url('https:\/\/www.trusteddecisions.com\/wp-content\/uploads\/2025\/02\/Design-ohne-Titel-51.png') no-repeat left center \/ contain !important; }\n.td-summary-box ul li strong { color: var(--td-dark); }\n@media (max-width: 768px) { .td-summary-box { padding: 28px 20px 24px; margin: 30px 0; } .td-summary-box h2.td-summary-title { font-size: 1.15rem !important; } .td-summary-box ul li { font-size: 0.95rem; padding-left: 34px !important; } }\n@media (prefers-reduced-motion: reduce) { .td-summary-box * { transition: none !important; } }\n<\/style>\n<div class=\"td-summary-box\">\n  <h2 class=\"td-summary-title\">Key Takeaways<\/h2>\n  <ul>\n    <li><strong>Human in the Loop (HITL)<\/strong> describes AI systems in which people review, correct and approve results at defined points before they take effect.<\/li>\n    <li>Human approval is necessary wherever errors are costly, decisions must be justified or carry legal consequences.<\/li>\n    <li>The EU AI Act requires effective human oversight for high-risk AI. Following the Digital Omnibus, this obligation applies to Annex III systems from 2 December 2027.<\/li>\n    <li>HITL only becomes effective with clear approval criteria, competent reviewers, documented interventions and a data foundation that makes deviations visible.<\/li>\n    <li>In financial controlling this means: AI forecasts are proposals, while planning responsibility stays with people.<\/li>\n  <\/ul>\n<\/div>\n\n \n\n<h2 class=\"wp-block-heading\">What Does Human in the Loop Mean?<\/h2>\n\n \n\n<h3 class=\"wp-block-heading\">Definition and Origin of the Term<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Human in the Loop describes an approach in which people are firmly integrated into the workflow of an automated or AI-supported system. Artificial intelligence delivers proposals, classifications or forecasts. A person reviews these results, corrects them or approves them.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">This integration creates a feedback loop: every human correction flows back into the system as input and improves the ML models throughout their entire life cycle. HITL is therefore not a one-off check, but a lasting interplay between human and machine.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">The term originally comes from control engineering and simulation, where people operate flight simulators, for example, as part of the control loop. In the debate on artificial intelligence it has taken on a new meaning. Today it stands for decisions that an AI system should not make on its own.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">The distinction matters: Human in the Loop does not mean that someone briefly glances at the result at some point. It refers to defined control points with clear intervention rights.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Human in the Loop, Human on the Loop and Human out of the Loop<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Depending on the degree of involvement, three models can be distinguished:<\/p>\n\n \n\n<ul class=\"wp-block-list\">\n<li><strong>Human in the Loop:<\/strong> The system acts only after human approval. Example: an AI proposes a credit decision, and a loan officer makes the final call.<\/li>\n\n\n<li><strong>Human on the Loop:<\/strong> The system acts independently while a person monitors operations and can intervene at any time. Example: automated invoice checking with spot checks and a stop function.<\/li>\n\n\n<li><strong>Human out of the Loop:<\/strong> The system operates fully autonomously. People are only involved in design and subsequent evaluation.<\/li>\n<\/ul>\n\n \n\n<p class=\"wp-block-paragraph\">Which model fits is determined not by the technology but by the risk of the respective decision. Many companies combine the models: routine cases run on the loop, exceptions and borderline cases in the loop.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Key Terms in the International Debate<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Literature, standards and vendor documentation on this topic are predominantly in English, and several terms are now common in business practice. <strong>Human oversight<\/strong> refers to the overarching human supervision of AI systems, as named in the EU AI Act. <strong>Human control<\/strong> means the actual ability to stop or redirect an operation.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\"><strong>Human feedback<\/strong> and <strong>human input<\/strong> stand for the responses with which people train and improve models. In generative AI, the method known as reinforcement learning from human feedback has become well known. <strong>Human interaction<\/strong> describes the design of the interface, in other words how people work with machines and which interactions the process provides for.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">The discussion on decision making also covers the accuracy and reliability of decisions, errors and problems in datasets, and the role that developers and business users each take on. A common example: developers are responsible for the design of a system, while business users contribute the expertise for the respective context. This division of labour is part of every viable approach to AI in the enterprise.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Why AI Decisions Need Human Approval<\/h2>\n\n \n\n<p class=\"wp-block-paragraph\">AI systems work faster and more consistently than people in many tasks. Nevertheless, there are five reasons why human approval remains indispensable in critical processes.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Accountability Cannot Be Delegated to a Model<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">A model is not liable. For decisions on loans, staffing, prices or budgets, the company remains responsible, and with it a specific person. Approval makes this responsibility visible, verifiable and documentable.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Models Fail at Edge Cases<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">ML models learn from historical data. Situations that rarely or never occur in the training data are answered with the same apparent certainty as routine cases. Such edge cases include market disruptions, new products or unusual customer constellations. People recognise the missing context and can judge whether a result is plausible.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Bias and Black-Box Effects Otherwise Go Unnoticed<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Biased data leads to biased results. Without human control, a system reproduces this bias on a large scale. Especially with complex models whose decision logic resembles a black box, human review is often the only point at which anomalies are noticed.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Traceability Builds Trust<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Decisions that a person has reviewed and justified can be explained to customers, auditors and supervisory authorities. This strengthens the acceptance of AI within the company, including among its own employees.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Legal Requirements Demand Human Oversight<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">For high-risk AI, the EU AI Act makes effective human oversight mandatory. What this means in practice is explained in the section on the EU AI Act below.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">For decision-making processes in the company, all five reasons lead to the same conclusion: the question is not whether people are involved, but at which point, with which criteria and with what authority.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">How Does Human in the Loop Work in the AI Life Cycle?<\/h2>\n\n \n\n<style>\n.td-kpi-light { --td-gold: #9E7D42; --td-gold-light: #C9A75F; --td-dark: #3C4B53; --td-text: #555555; background: #FFFFFF; border-left: 2px solid var(--td-gold); border-bottom: 2px solid var(--td-gold); border-radius: 12px; box-shadow: 0 15px 40px rgba(0,0,0,0.08); padding: 44px 42px; margin: 40px 0; width: 100%; max-width: 100%; font-family: \"Open Sans\", -apple-system, \"Helvetica Neue\", Arial, sans-serif; box-sizing: border-box; }\n.td-kpi-light .td-kpi-light-header { display: flex; align-items: center; justify-content: center; gap: 12px; margin-bottom: 30px; padding-bottom: 14px; position: relative; text-align: center; }\n.td-kpi-light .td-kpi-light-header::after { content: \"\"; position: absolute; left: 50%; bottom: 0; transform: translateX(-50%); width: 80px; height: 3px; background-color: var(--td-gold); border-radius: 2px; }\n.td-kpi-light .td-kpi-light-icon { color: var(--td-gold); display: flex; }\n.td-kpi-light h3 { color: var(--td-dark); margin: 0; font-size: 1.2rem; font-weight: 700; text-transform: uppercase; letter-spacing: 0.5px; line-height: 1.35; }\n.td-kpi-light .td-kpi-light-grid { display: grid; grid-template-columns: repeat(6, 1fr); gap: 18px; }\n.td-kpi-light .td-kpi-light-card { grid-column: span 2; background: #FAF9F7; border: 1px solid #ECE8E0; border-radius: 10px; padding: 22px 22px 24px; box-sizing: border-box; transition: transform 0.25s ease, box-shadow 0.25s ease, border-color 0.25s ease; }\n.td-kpi-light .td-kpi-light-card:nth-child(4), .td-kpi-light .td-kpi-light-card:nth-child(5) { grid-column: span 3; }\n.td-kpi-light .td-kpi-light-card:hover { transform: translateY(-3px); box-shadow: 0 10px 24px rgba(158,125,66,0.12); border-color: var(--td-gold-light); }\n.td-kpi-light .td-kpi-light-card strong { display: block; color: var(--td-gold); font-size: 0.82rem; font-weight: 700; text-transform: uppercase; letter-spacing: 1px; margin-bottom: 10px; }\n.td-kpi-light .td-kpi-light-card p { margin: 0; color: var(--td-text); font-size: 0.9rem; line-height: 1.65; }\n@media (max-width: 768px) { .td-kpi-light { padding: 28px 20px; margin: 30px 0; } .td-kpi-light .td-kpi-light-header { justify-content: flex-start; text-align: left; } .td-kpi-light .td-kpi-light-header::after { left: 0; transform: none; } .td-kpi-light h3 { font-size: 1.05rem; } .td-kpi-light .td-kpi-light-grid { grid-template-columns: 1fr; } .td-kpi-light .td-kpi-light-card, .td-kpi-light .td-kpi-light-card:nth-child(4), .td-kpi-light .td-kpi-light-card:nth-child(5) { grid-column: span 1; } }\n@media (prefers-reduced-motion: reduce) { .td-kpi-light .td-kpi-light-card { transition: none; } .td-kpi-light .td-kpi-light-card:hover { transform: none; } }\n<\/style>\n<div class=\"td-kpi-light\">\n  <div class=\"td-kpi-light-header\">\n    <div class=\"td-kpi-light-icon\"><svg width=\"26\" height=\"26\" viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><line x1=\"18\" y1=\"20\" x2=\"18\" y2=\"10\"><\/line><line x1=\"12\" y1=\"20\" x2=\"12\" y2=\"4\"><\/line><line x1=\"6\" y1=\"20\" x2=\"6\" y2=\"14\"><\/line><\/svg><\/div>\n    <h3>Key Figures for Effective Human in the Loop Management<\/h3>\n  <\/div>\n  <div class=\"td-kpi-light-grid\">\n    <div class=\"td-kpi-light-card\"><strong>Intervention rate<\/strong><p>Share of cases that a person reviews instead of approving them automatically.<\/p><\/div>\n    <div class=\"td-kpi-light-card\"><strong>Override rate<\/strong><p>Share of reviewed cases in which the person changes or rejects the AI proposal.<\/p><\/div>\n    <div class=\"td-kpi-light-card\"><strong>Confidence threshold<\/strong><p>Model confidence value below which a case must go for approval.<\/p><\/div>\n    <div class=\"td-kpi-light-card\"><strong>Error rate after approval<\/strong><p>Share of approved results that later turn out to be wrong. Shows the quality of the review.<\/p><\/div>\n    <div class=\"td-kpi-light-card\"><strong>Approval lead time<\/strong><p>Time from AI proposal to human decision. Shows whether HITL slows the process down.<\/p><\/div>\n  <\/div>\n<\/div>\n\n \n\n<p class=\"wp-block-paragraph\">Human in the Loop applies at two stages: when a model is trained and during its ongoing operation. Both phases require different expertise and different roles.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Training: Annotation and Human Feedback<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Many machine learning models need data that people have annotated beforehand, for example receipts marked as correct or incorrect, or classified customer enquiries. Specialists also check whether training data is complete and representative.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">With active learning, the model specifically selects the cases in which it is most uncertain and presents only these to a person. This creates a strong learning effect with limited effort.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Operation: Validation, Edge Cases and Feedback Loop<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">During operation, people review the results before they take effect. A threshold principle has proven effective: if a model&#8217;s confidence is above a defined value, the case continues automatically. Below it, the case goes to a person for review.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">Every intervention is documented and feeds into the next training round. This feedback loop gradually increases the accuracy and reliability of the system. At the same time, it shows where errors accumulate and where processes or data sources need to be improved.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">To keep the loop manageable, it needs key figures. How often do reviewers intervene? How often do they overrule the model? How long does an approval take? Only these values show whether HITL is working or merely costing time.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Benefits and Limits of HITL<\/h2>\n\n \n\n<h3 class=\"wp-block-heading\">Accuracy, Bias Control and Traceability<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Human in the Loop combines the speed of AI with human judgement. The benefits show in three areas:<\/p>\n\n \n\n<ul class=\"wp-block-list\">\n<li><strong>Accuracy:<\/strong> Human review catches errors before they have consequences, and the feedback improves the models permanently.<\/li>\n\n\n<li><strong>Bias control:<\/strong> Reviewers detect biases that were not visible in test data and can take corrective action.<\/li>\n\n\n<li><strong>Traceability:<\/strong> Documented approvals create transparency and support ethical decision making as well as accountability towards third parties.<\/li>\n<\/ul>\n\n \n\n<h3 class=\"wp-block-heading\">Scalability, Costs and Automation Bias<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">HITL also has limits that decision-makers should be aware of. Every manual review costs time and staff. Anyone who has every case approved loses the efficiency gain of automation.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">On top of this comes automation bias: people tend to follow machine suggestions, especially under time pressure. An approval that is confirmed in 98 percent of cases is often no longer a genuine review. People also make mistakes and do not always judge consistently.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">The solution lies in risk-based design: human review where it has an effect, automation where the risk is low.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Fields of Application from Medicine to Finance<\/h2>\n\n \n\n<style>\n.td-table-light { --td-gold: #9E7D42; --td-dark: #3C4B53; --td-text: #555555; width: 100%; max-width: 100%; margin: 40px 0; overflow-x: auto; -webkit-overflow-scrolling: touch; background: #FFFFFF; border-left: 2px solid var(--td-gold); border-bottom: 2px solid var(--td-gold); border-radius: 12px; box-shadow: 0 15px 40px rgba(0,0,0,0.08); box-sizing: border-box; }\n.td-table-light table { width: 100%; min-width: 620px; border-collapse: collapse; font-family: \"Open Sans\", -apple-system, \"Helvetica Neue\", Arial, sans-serif; text-align: left; margin: 0; border: none; }\n.td-table-light th { background: #F4F3F0; color: var(--td-dark); padding: 18px 22px; font-size: 0.78rem; font-weight: 700; text-transform: uppercase; letter-spacing: 1px; line-height: 1.4; vertical-align: bottom; border: none; border-bottom: 2px solid var(--td-gold); }\n.td-table-light td { padding: 16px 22px; color: var(--td-text); font-size: 0.9rem; line-height: 1.65; vertical-align: top; border: none; border-bottom: 1px solid #ECE8E0; }\n.td-table-light tbody tr { transition: background-color 0.25s ease; }\n.td-table-light tbody tr:nth-child(even) { background: #FAF9F7; }\n.td-table-light tbody tr:hover { background: #F4EFE4; }\n.td-table-light tbody tr:last-child td { border-bottom: none; }\n.td-table-light td strong { color: var(--td-dark); font-weight: 700; }\n@media (max-width: 768px) { .td-table-light { margin: 30px 0; } .td-table-light table { min-width: 580px; } .td-table-light th { padding: 14px 16px; font-size: 0.72rem; } .td-table-light td { padding: 14px 16px; font-size: 0.85rem; } }\n@media (prefers-reduced-motion: reduce) { .td-table-light tbody tr { transition: none; } }\n<\/style>\n<div class=\"td-table-light\" tabindex=\"0\" role=\"region\" aria-label=\"Human in the Loop by field of application\">\n  <table>\n    <thead>\n      <tr><th>Field of application<\/th><th>Point of human intervention<\/th><th>Risk without approval<\/th><\/tr>\n    <\/thead>\n    <tbody>\n      <tr><td><strong>Medical diagnostics<\/strong><\/td><td>Physician confirms or rejects the AI finding<\/td><td>Misdiagnosis and liability risk<\/td><\/tr>\n      <tr><td><strong>Lending<\/strong><\/td><td>Loan officer decides on the AI proposal<\/td><td>Wrong decisions and discrimination, high-risk AI under the AI Act<\/td><\/tr>\n      <tr><td><strong>Fraud detection<\/strong><\/td><td>Analyst reviews the suspected case before blocking an account<\/td><td>Wrongly blocked customers and reputational damage<\/td><\/tr>\n      <tr><td><strong>Customer service<\/strong><\/td><td>Employees review the AI reply before sending<\/td><td>Incorrect commitments and loss of trust<\/td><\/tr>\n      <tr><td><strong>Recruitment<\/strong><\/td><td>Recruiters make the selection after AI pre-sorting<\/td><td>Discrimination, high-risk AI under the AI Act<\/td><\/tr>\n      <tr><td><strong>Controlling and planning<\/strong><\/td><td>Controller approves the AI forecast<\/td><td>Mismanagement of budgets and liquidity<\/td><\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n \n\n<p class=\"wp-block-paragraph\">Human in the Loop is found wherever AI-supported decisions have significant consequences. The use cases differ mainly in the point at which a person intervenes.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">In <strong>medical diagnostics<\/strong>, AI flags conspicuous findings in imaging data, and the physician makes the diagnosis. In <strong>finance<\/strong>, models propose credit decisions or report suspicious transactions, and analysts verify the cases before accounts are blocked or loans are declined.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">In <strong>autonomous mobility<\/strong>, a person takes over when the system cannot reliably assess a situation. In <strong>customer service<\/strong>, AI assistants draft replies that employees check before sending. In <strong>human resources<\/strong>, systems pre-sort applications, and people make the selection.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">What all these examples have in common: the model accelerates, the person decides. Collaboration is clearly divided, and monitoring of the system does not end at go-live.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Human in the Loop and the EU AI Act<\/h2>\n\n \n\n<h3 class=\"wp-block-heading\">What Article 14 Requires for Human Oversight<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">The <a href=\"https:\/\/eur-lex.europa.eu\/eli\/reg\/2024\/1689\/oj\" target=\"_blank\" rel=\"noopener noreferrer\">EU AI Act (Regulation (EU) 2024\/1689)<\/a> makes human oversight mandatory for high-risk AI. Article 14 requires such systems to be designed so that natural persons can effectively oversee them while they are in use. The aim is to prevent or minimise risks to health, safety and fundamental rights.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">The persons exercising oversight must understand the capabilities and limitations of the system, interpret its output correctly and be able to consciously decide against a recommendation. They must also be able to stop the system or disregard its output. The regulation explicitly names the risk of relying too heavily on machine output.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">High-risk AI includes systems in areas such as employment and recruitment, creditworthiness assessment, critical infrastructure and education. For companies, this means that anyone using AI in these fields needs a demonstrable HITL concept.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">New Deadlines Under the Digital Omnibus<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">With the <a href=\"https:\/\/eur-lex.europa.eu\/eli\/reg\/2026\/1744\/oj\" target=\"_blank\" rel=\"noopener noreferrer\">Digital Omnibus on AI (Regulation (EU) 2026\/1744)<\/a>, the EU has postponed the application of the high-risk obligations. The regulation was published in the Official Journal on 24 July 2026. For stand-alone high-risk systems under Annex III, the obligations now apply from 2 December 2027, and for AI in regulated products under Annex I from 2 August 2028.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">The postponement is a breathing space, not an all-clear. The obligation to ensure AI literacy among staff under Article 4 has applied since February 2025. And an effective oversight concept cannot be introduced shortly before the deadline, because it affects processes, roles and data.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Human in the Loop in Financial Controlling and Corporate Planning<\/h2>\n\n \n\n<h3 class=\"wp-block-heading\">AI Forecasts with an Approval Step<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">In financial controlling, AI is making its way in primarily through forecasts. Models of <a href=\"https:\/\/www.trusteddecisions.com\/en\/blog\/predictive-analysis-future-oriented-corporate-strategies\/\">predictive analysis<\/a> recognise patterns in sales, cost and liquidity data faster than any manual evaluation. Yet the decision on a forecast, a budget or a measure remains a management decision.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">A practical HITL model for <a href=\"https:\/\/www.trusteddecisions.com\/en\/blog\/forecast-planning-and-customer-analysis-how-to-plan-successfully\/\">forecast planning<\/a> looks like this: the AI creates a proposal for each planning object. Deviations above a defined tolerance compared with the previous year or the last plan are flagged and presented to the responsible controller. The controller confirms, adjusts or rejects the proposal and documents the reasoning in the system. This keeps every planning figure assigned to a responsible person.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Data Foundation and Planning Architecture for Control Points<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">A control point is only as good as the data it is based on. Reviewers need not only the AI&#8217;s result but also how it was derived: drivers, comparative values, data source and data status. Without this transparency, approval becomes a formality.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">This results in three requirements for the architecture. First, a consistent data foundation backed by reliable <a href=\"https:\/\/www.trusteddecisions.com\/en\/blog\/data-quality-management-as-a-control-loop-kpis-operating-model-roi\/\">data quality management<\/a>, so that deviations are genuine signals and not data errors. Second, planning systems that store versions, comments and approvals in an audit-proof way, as solutions such as Corporate Planner make possible. Third, dashboards, for example in Qlik, that make intervention rates and deviations visible.<\/p>\n\n \n\n<p class=\"wp-block-paragraph\">Only when these three levels work together does Human in the Loop turn from a principle into a manageable process.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Designing Approval Properly: Four Principles for Practice<\/h2>\n\n \n\n<style>\n.td-premium-cta-split { display: flex; flex-direction: column; background-color: #FFFFFF; border-radius: 12px; overflow: hidden; width: 100%; max-width: 100%; margin: 60px 0; box-shadow: 0 15px 40px rgba(0,0,0,0.08); border-left: 2px solid #9E7D42; border-bottom: 2px solid #9E7D42; box-sizing: border-box; font-family: \"Open Sans\", -apple-system, sans-serif; }\n.td-premium-cta-visual { width: 100%; height: 300px; background-image: url('https:\/\/www.trusteddecisions.com\/wp-content\/uploads\/2026\/10\/human-in-the-loop-titelbild.webp'); background-size: cover; background-position: center; position: relative; overflow: hidden; }\n.td-premium-cta-visual::after { content: \"\"; position: absolute; top: 0; left: 0; width: 100%; height: 100%; background: linear-gradient(180deg, rgba(60, 75, 83, 0.2) 0%, rgba(255, 255, 255, 0.3) 100%); z-index: 1; }\n.td-premium-cta-content { padding: 50px 40px; display: flex; flex-direction: column; justify-content: center; align-items: center; background-color: #FFFFFF; text-align: center; }\n.td-premium-cta-logo { max-width: 200px; height: auto; margin-bottom: 25px; display: block; }\n.td-premium-cta-title { color: #3C4B53; font-weight: 700; font-size: 1.8rem; line-height: 1.3; margin-bottom: 20px; text-transform: uppercase; letter-spacing: 0.5px; position: relative; padding-bottom: 12px; }\n.td-premium-cta-title::after { content: \"\"; position: absolute; bottom: 0; left: 50%; transform: translateX(-50%); width: 80px; height: 3px; background-color: #9E7D42; border-radius: 2px; }\n.td-premium-cta-text { color: #555555; font-size: 1rem; line-height: 1.7; margin-bottom: 30px; max-width: 650px; }\n.td-premium-cta-text strong { color: #3C4B53; }\n.td-premium-cta-button { display: inline-block; background-color: #9E7D42; color: #FFFFFF !important; font-weight: 700; font-size: 1rem; text-transform: uppercase; letter-spacing: 1px; text-decoration: none; padding: 16px 35px; border-radius: 50px; border: 2px solid #9E7D42; transition: transform 0.3s ease, background-color 0.3s ease; box-shadow: 0 8px 20px rgba(158,125,66,0.15); }\n.td-premium-cta-button:hover { background-color: #3C4B53; border-color: #3C4B53; transform: translateY(-3px) scale(1.02); }\n@media (max-width: 768px) { .td-premium-cta-split { margin: 40px 0; } .td-premium-cta-visual { height: 200px; } .td-premium-cta-content { padding: 35px 20px; } .td-premium-cta-title { font-size: 1.4rem; } .td-premium-cta-button { width: 100%; box-sizing: border-box; } }\n<\/style>\n<div class=\"td-premium-cta-split\">\n  <div class=\"td-premium-cta-visual\">&nbsp;<\/div>\n  <div class=\"td-premium-cta-content\">\n    <img decoding=\"async\" class=\"td-premium-cta-logo\" src=\"https:\/\/www.trusteddecisions.com\/wp-content\/uploads\/2023\/11\/cropped-TD-Logo-Transparent_250x76.png\" alt=\"Trusted Decisions Logo\" \/> \n    <span class=\"td-premium-cta-title\">Steer AI Decisions Safely<\/span>\n    <p class=\"td-premium-cta-text\"><strong>Control by design:<\/strong> We combine AI-supported forecasts and plans with clear approval processes, reliable data and transparent dashboards in Qlik and Corporate Planner.<\/p>\n    <a class=\"td-premium-cta-button\" href=\"https:\/\/kontakt.trusteddecisions.com\/kontakt\/page_ghq1xw\/\">Schedule a Call Now<\/a>\n  <\/div>\n<\/div>\n\n \n\n<p class=\"wp-block-paragraph\">How can HITL be introduced so that it protects without slowing things down? Four principles have proven effective.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Define Control Points Before Automating<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Before go-live, define which decisions require approval and from which threshold. The basis is a risk assessment for each process step: what does an error cost, how quickly is it noticed, who is affected?<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Make Approval Criteria Readable for Machines and People<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Reviewers need clear criteria, not gut decisions. For every proposal, the system should show why it arrived at this result and how confident it is.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Embed Feedback Loops as a Routine<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Corrections must be recorded in a structured way and evaluated regularly. Only then does the model learn, and only then does the organisation recognise where processes or data need to be improved.<\/p>\n\n \n\n<h3 class=\"wp-block-heading\">Clarify Responsibility and Intervention Rights<\/h3>\n\n \n\n<p class=\"wp-block-paragraph\">Define who may approve, who escalates and who can stop a system. Reviewers need time, training and explicit backing to contradict a model. A reviewer who merely confirms under time pressure is a sham control.<\/p>\n\n \n\n<h2 class=\"wp-block-heading\">Frequently Asked Questions About Human in the Loop<\/h2>\n\n \n\n<style>\n.td-faq-box { --td-gold: #9E7D42; --td-gold-light: #C9A75F; --td-dark: #3C4B53; --td-text: #555555; background: #FFFFFF; border-radius: 12px; box-shadow: 0 15px 40px rgba(0,0,0,0.08); margin: 30px 0 40px; width: 100%; max-width: 100%; font-family: \"Open Sans\", -apple-system, \"Helvetica Neue\", Arial, sans-serif; box-sizing: border-box; overflow: hidden; }\n.td-faq-box details { border-bottom: 1px solid #ECE8E0; }\n.td-faq-box details:last-child { border-bottom: none; }\n.td-faq-box summary { list-style: none; cursor: pointer; display: flex; align-items: center; justify-content: space-between; gap: 18px; padding: 22px 30px; transition: background-color 0.25s ease; }\n.td-faq-box summary::-webkit-details-marker { display: none; }\n.td-faq-box summary:hover { background: #FAF9F7; }\n.td-faq-box summary:focus-visible { outline: 2px solid var(--td-gold); outline-offset: -2px; }\n.td-faq-box summary h3 { margin: 0 !important; padding: 0; font-family: inherit; color: var(--td-dark); font-size: 1.05rem !important; font-weight: 700; line-height: 1.45 !important; }\n.td-faq-box summary::after { content: \"+\"; flex: 0 0 auto; width: 30px; height: 30px; border-radius: 50%; border: 2px solid var(--td-gold); color: var(--td-gold); font-size: 1.2rem; font-weight: 700; line-height: 26px; text-align: center; box-sizing: border-box; transition: transform 0.25s ease; }\n.td-faq-box details[open] summary::after { content: \"\\2212\"; }\n.td-faq-box details[open] summary { background: #FAF9F7; }\n.td-faq-box .td-faq-boxnswer { padding: 0 30px 24px; background: #FAF9F7; }\n.td-faq-box .td-faq-boxnswer p { margin: 0; color: var(--td-text); font-size: 0.98rem; line-height: 1.7; }\n.td-faq-box .td-faq-boxnswer a { color: var(--td-gold); }\n@media (max-width: 768px) { .td-faq-box { margin: 24px 0 30px; } .td-faq-box summary { padding: 18px 18px; } .td-faq-box summary h3 { font-size: 0.98rem !important; } .td-faq-box .td-faq-boxnswer { padding: 0 18px 20px; } }\n@media (prefers-reduced-motion: reduce) { .td-faq-box summary, .td-faq-box summary::after { transition: none; } }\n<\/style>\n<div class=\"td-faq-box\">\n  <details>\n    <summary><h3>What Is the Human-in-the-Loop Concept?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>Human in the Loop is a concept in which people validate, correct or approve AI results at defined points. Human feedback flows back into the system and continuously improves the models.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>What Is the Difference Between Human in the Loop and Human on the Loop?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>With Human in the Loop, the system acts only after human approval. With Human on the Loop, it acts independently while a person monitors operations and intervenes when necessary.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>Does HITL Improve the Accuracy of ML Systems?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>Yes. Human review catches errors during ongoing operation, and the documented corrections serve as additional training data. As a result, the accuracy and adaptability of the models increase over time.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>What Role Do People Play in Training AI Models?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>People annotate training data, check it for completeness and bias, and evaluate model output. In generative AI, such evaluations flow directly into training via human feedback.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>Is ChatGPT Human in the Loop?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>In training, yes: language models such as ChatGPT were refined, among other methods, with reinforcement learning from human feedback, in which people rate responses. In use, a Human in the Loop process only emerges when users review the output before it is used.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>Is Agentic AI Human in the Loop?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>Not automatically. AI agents carry out tasks independently in several steps. Whether a person is involved depends on the configuration. For critical actions such as payments, contract changes or customer communication, approval steps should be built in permanently.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>Where Is Human in the Loop Used in Practice?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>Typical fields of application are medical diagnostics, credit decisions and fraud detection in finance, autonomous mobility, customer service and recruitment. In financial controlling, it mainly concerns AI-supported forecasts and plans.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>Does HITL Meet the Requirements of the EU AI Act?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>A well-designed HITL concept is the central building block for meeting the human oversight obligation under Article 14. However, it must be effective: the persons involved need competence, information and the authority to stop the system.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>Does Human in the Loop Slow Down AI Processes?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>Only where every case is reviewed. With risk-based thresholds, routine cases run automatically and only uncertain or high-impact cases go for approval. This largely preserves the efficiency gain.<\/p><\/div>\n  <\/details>\n  <details>\n    <summary><h3>How Can Sham Control Be Prevented?<\/h3><\/summary>\n    <div class=\"td-faq-boxnswer\"><p>Through measurable key figures such as intervention and override rates, sufficient time for reviews and training. If the confirmation rate remains close to 100 percent over time, the criteria and the reviewers&#8217; working conditions should be questioned.<\/p><\/div>\n  <\/details>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>Key Takeaways Human in the Loop (HITL) describes AI systems in which people review, correct and approve results at defined<span class=\"excerpt-hellip\"> [\u2026]<\/span><\/p>\n","protected":false},"author":2,"featured_media":10119,"parent":0,"template":"","format":"standard","meta":[],"blog-tag":[],"blog-type":[135],"class_list":["post-10118","blog-new","type-blog-new","status-publish","format-standard","has-post-thumbnail","hentry","blog-type-ki-technologie"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/blog-new\/10118","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/blog-new"}],"about":[{"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/types\/blog-new"}],"author":[{"embeddable":true,"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/users\/2"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/media\/10119"}],"wp:attachment":[{"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/media?parent=10118"}],"wp:term":[{"taxonomy":"blog-tag","embeddable":true,"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/blog-tag?post=10118"},{"taxonomy":"blog-type","embeddable":true,"href":"https:\/\/www.trusteddecisions.com\/en\/wp-json\/wp\/v2\/blog-type?post=10118"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}