Data-driven optimization and scale-up of a nickel metallaphotoredox alkenylation

<jats:p>Highly substituted olefins are ubiquitous structural motifs in pharmaceuticals, natural products, and functional materials, yet their modular synthesis from abundant feedstocks remains challenging. Here, we report a nickel metallaphotoredox cross-coupling between hydroboration-derived trialkylboranes and alkenyl triflates, providing direct access to highly substituted olefins from simple olefins and ketones. Reaction development was enabled through a two-stage machine-learning workflow in which Bayesian optimization rapidly identified productive reaction conditions in batch, while transfer learning accelerated translation to continuous flow by exploiting shared chemical information between the two reactor platforms. The multitask model revealed a strong correlation between batch and flow (ρ = 0.80) and improved predictive accuracy relative to independently trained models, reducing leave-one-out prediction errors by up to 14%. The resulting methodology enables efficient generation of otherwise challenging-to-access primary alkyl radicals and accommodates a broad range of olefin- and ketone-derived coupling partners, including substrates bearing ethers, esters, enamides, boronic esters, and strained ring systems. Mechanistic studies support a dual nickel/photoredox catalytic radical manifold and confirm the enhanced performance of an unusual catalyst composition identified through Bayesian optimization, highlighting how data-driven optimization can uncover productive catalytic regimes that depart from conventional catalyst-design heuristics.</jats:p>