Analysis ·
AI in business: why you need to keep humans in the loop
Original title, in French: “IA en entreprise : pourquoi il faut mettre l’humain dans la boucle”
- Publication
- ZDNet France (France)
- Author
- Ritoban Mukherjee
- Published on
- 30 September 2026 · read in the page structured data
Summary
The article explores the "human-in-the-loop" (HITL) model, in which decisions made by AI systems are subject to human validation before execution, notably in healthcare, finance and legal work. Several experts (Akash Thakur, Daniel Gamber of Cambrion, Asim Husain of Alterion, Eric Vaughan of IgniteTech) explain the limits of model confidence scores, the four types of response available when an error occurs, and the importance of an "audit trail" required by the European AI Act. Concrete examples are cited: the Nominal platform for accounts payable management, Innovaccer Gravity for medical care providers, and IgniteTech's MyPersonas and Eloquens AI products, which achieve 95% "grounding accuracy". The article also mentions an FTC settlement with DoNotPay and an algorithmic bias detected by the IRS.
Summary based on the full text of the article.
Read the article on ZDNet France →Figures cited
-
193,000 USD
Amount of the FTC settlement with DoNotPay
United States, February 2025
« En février 2025, aux Etats-Unis, la FTC (Federal Trade Commission, qui contrôle la consommation et le contrôle des pratiques anticoncurrentielles) a finalisé un accord à hauteur de 193 000 dollars contre DoNotPay, la plateforme qui se présentait comme le premier avocat robot au monde. »
Source: FTC
-
95 %
Grounding accuracy of IgniteTech's product
IgniteTech's MyPersonas and Eloquens AI products
« Notre précision d'ancrage se situe autour de 95 %, nous y avons travaillé très dur. »
-
—
Audit frequency of Black taxpayers compared with non-Black taxpayers
United States, 2024 GAO report on the IRS audit selection system
« selon un rapport de 2024 du Government Accountability Office (aux Etats-Unis) consacré au système automatisé de sélection des contrôles fiscaux (service des impôts), les contribuables noirs faisaient l'objet de contrôles trois à cinq fois plus souvent que les contribuables non noirs »
Source: Government Accountability Office
Classification
- Country
- United States
- Company size
- All sizes
- Industries
- Banking, insurance, asset management, finance departments · Healthcare, pharmaceuticals, social care · Consulting, legal, accounting, business services · All industries, or industry not specified
- Companies
- Alterion (supplier), DoNotPay (cited), Innovaccer Gravity (supplier), IgniteTech (supplier), Cambrion (supplier), Nominal (supplier), Google (cited)
- Keywords
- human oversight, human-in-the-loop, ai compliance, algorithmic bias, ai audit, ai governance
Entry written by Marsanne Studio from the cited article, with the help of an AI model (claude-sonnet-5). The rights to the article and its illustrations belong to ZDNet France and its author. Entry translated from French by Claude Sonnet 5.5.