Reactive intermediates are key to halting and promoting chemical transformations, however due to their elusive nature, they are seldom harnessed for reaction design. Herein, we describe studies aimed at stabilizing reactive intermediates in the N-heterocyclic carbene (NHC) catalytic cycle, which enabled fully shutting down the known benzoin coupling pathway, while rerouting its intermediates toward deuteration. The reversible nature of NHC catalysis and the selective stabilization of reaction intermediates facilitated clean hydrogen-deuterium exchange reactions of aromatic aldehydes by D<sub>2</sub>O, even for challenging electron withdrawing substrates. The addition of catalytic amounts of phenyl boronic acid was used to further stabilize highly reactive intermediates and mitigate the formation of benzoin coupling by-products. The mechanistic understanding at the foundation of this work resulted in unprecedented mild conditions with base and catalyst loadings as low as 0.1 mol%, and a scalable deuteration reaction applicable to a broad substrate scope with outstanding functional group tolerance. More importantly, adopting this approach enabled the construction of a machine-learning derived guideline for identifying the appropriate catalyst and conditions for different substrates based on a logistic regression classification model. Experimental studies combined with machine learning and computational methods shed light on the non-trivial mechanistic underpinnings of this reaction.