“By rethinking how data is stored and accessed, moving from siloed third-party systems to user-centric data models, organizations can create more fluid, responsive web and mobile interactions that adapt to preferences in real-time,” says Osmar Olivo, VP of product management at Inrupt. “To maintain accuracy and performance, AI-driven experiences should be trained with diverse, real-world data while also incorporating user feedback mechanisms that allow individuals to correct, refine, and guide AI-generated insights by supplying their own preferences and metadata.”
Manish Rai, VP of product marketing at SnapLogic, predicts more than 80% of generative AI projects fail due to data connectivity, quality, and trust issues. “Success depends on tools that simplify agent development, make data AI-ready, and ensure reliability through observability, evaluation for accuracy, and policy enforcement.”
Rosaria Silipo, VP of data science evangelism at KNIME, notes many agentic applications have a human-in-the-loop step to check for correctness. “In other cases, special guardian AI agents focus on controlling the result; if the result is not satisfactory, they send it back and ask for an improved version.” For more data-related tasks, such as sentiment analysis, “genAI accuracy is compared to the accuracy of other classic machine learning models.”
