Dynamical decoupling (DD) is a key technique for suppressing decoherence and preserving the performance of quantum algorithms. We introduce a measurement-informed DD (MDD) protocol that determines control unitary gates from partial measurements of noisy subsystems, with measurement overhead scaling linearly with the number of subsystems. We prove that, under local energy relaxation and dephasing noise, MDD achieves the maximum entanglement fidelity attainable by any DD scheme based on bang–bang operations to first order in evolution time. On the IBM Eagle processor, MDD enhanced the success probability of a $14$-qubit quantum Fourier transform by more than two orders of magnitude compared with standard and higher-order DD sequences, including XX, XY, and UDD protocols, and improved the accuracy of $56$-qubit sample-based quantum diagonalization for N${}_2$ compared with XX-pulse DD. These results establish MDD as a scalable and effective approach for suppressing decoherence in large-scale quantum algorithms.