<p>A substantial proportion of cognitive impairment and dementia remains undiagnosed, posing a major global health challenge. However, there is little empirical evidence to guide which case-finding strategies most effectively balance diagnostic reach and cost-effectiveness. To address this, we developed a digital twin of Singapore’s older adult population (<i>n</i> = 753,905), simulated using a deep learning model trained on a real-world cohort and scaled to match national demographic structures. Our simulations show that without active case finding, 79.5% of cases are undiagnosed. A targeted strategy of screening adults aged ≥75 years with worries about cognitive decline achieved an optimal balance, reducing the undiagnosed rate to 48.3% at a low cost per new diagnosis (SGD$25.3). The most cost-effective strategy, however, depends on the willingness to pay, with highly targeted approaches (for example, age ≥85 years) favoured at lower thresholds, and broader strategies (for example, known hypertension) favoured at higher ones. Furthermore, lower screening test specificity markedly increases downstream health system costs. Our findings demonstrate the power of digital twin simulations to inform resource allocation, by providing a tiered framework for policies on case finding in cognitive impairment: highly targeted strategies can be adopted in resource-limited settings, with systematic expansion to broader approaches as the willingness to pay permits. Although this study models the immediate resource requirements for case finding, evaluation of the long-term benefits and harms of these strategies will be an important next step for future research.</p>