The Training Data Project Wins Prestigious ICEAA 2025 Best Paper Award for Work on AI Data Labeling and Risk Reduction

Associated Press

23-06-2025

06/23/2025, Washington, D.C // PRODIGY: Feature Story //
David Cook, Co-Founder of The Training Data Project (Source: The Training Data Project)
The Training Data Project, a company focused on quantifying AI value and pioneering data labeling standards, has been awarded the 2025 Best Paper honor by the International Cost Estimating and Analysis Association (ICEAA) in the Management, EVM, Software & Agile category. The winning paper, 'Enabling Measurable Success in DoD AI Programs from Acquisition to Operations,' highlights the central role that training data and data labeling play in AI performance, accountability, and long-term program value.
The paper, co-authored by The Training Data Project co-founder David Cook, was selected from a competitive field of government and industry contributors. It outlines a practical methodology for quantifying the value and risk associated with AI systems in Department of Defense programs, beginning not at deployment, but at the foundation: the training data pipeline.
'It's an incredible honor to be recognized by the ICEAA, especially at a conference of cost estimators, a community I've never formally belonged to,' said Cook. 'But that's also the point. As AI continues to expand, its financial and operational value depends on something often overlooked: the integrity of the data we feed into it.'
At the core of The Training Data Project's mission is the belief that nothing moves in AI without quality data. Data labeling, the process of annotating and identifying data points to 'teach' AI models what to pay attention to, is the bridge between raw inputs and intelligent outcomes. When done incorrectly, the results can be not just ineffective, but dangerous.
'Bad training data is worse than no training data,' Cook added. 'Mistakes made early in the labeling process don't just vanish. They cascade. They replicate through the system like compound interest, and by the time you spot the failure, the only option might be to start over.'
To train an AI to recognize a stop sign, for example, it's not enough to feed it thousands of perfectly clear images. The model must also be exposed to a wide range of real-world variations including poor lighting, partial obstructions, weather damage, unusual angles, and visual interference. The more representative and well-labeled the training data, the better the AI can generalize and respond accurately in unpredictable, real-life conditions.
'Training data is not optional, it is foundational,' said Cook. 'Its importance spans all forms of AI. For Large Language Models, which depend on scale, diversity, and structure to function, it is absolutely crucial. Without standards and measurable quality in training data, organizations invite unquantifiable risk across the entire AI pipeline. Value in AI begins with value in the data.'
Founded in 2023 by Cook and CEO Noami DeVore, The Training Data Project helps government and enterprise organizations navigate the complex intersection of data labeling, AI governance, and risk reduction. The company's mission is structured around a framework it calls TRUST: Transparent, Reachable, Unbiased, Standards-based, and Traceable data practices.
Its work spans three primary pillars: defining quality and standards for training data, sharing best practices for cost-effective curation, and developing open source tools that support responsible AI deployment. From military applications to commercial AI systems, The Training Data Project offers a clear warning and a hopeful path forward. If organizations commit to data quality at the outset, they can unlock both innovation and measurable value while avoiding costly downstream failures.
Media Contact:
Name - Noami DeVore
Email - [email protected]