2026 Wiley Survey of 700 Practitioners: Data Curation Time Triples Success Rate in Visual and Physical AI Deployments

The Wiley-IEEE Spectrum white paper reports that data problems cause the majority of model failures in video, LiDAR, and sensor systems. 78% of teams record measurable production value while 74% still rate the domain underinvested. Annotation pipelines label entire datasets then discard large fractions before deployment, confirming curation volume over model scale as the decisive variable.

Prior text-centric scaling assumptions do not transfer. High-dimensional inputs require per-frame or per-point selection that current benchmarks under-measure. Teams shipping production systems already apply active selection and quality filtering at rates three times higher than non-shipping teams, exposing a repeatable operational pattern rather than isolated anecdotes.

The survey understates downstream effects on regulatory and safety validation. Physical AI systems that reach deployment carry forward data lineage requirements that text models largely avoid. Organizations that treat data work as residual capacity will face repeated retraining cycles and certification delays within the next 18 months.

Teams adopting structured data-centric workflows will separate further from architecture-focused peers. Production metrics will increasingly track curation hours per deployed model rather than parameter counts.