A media startup wants to add machine learning-based image tagging to its application but has no in-house ML expertise and does not want to hire specialized data scientists or manage GPU clusters. Which AWS Well-Architected Framework design principle for the Performance Efficiency pillar best supports this goal?