Abstract
Accelerators have emerged as critical intermediaries in the entrepreneurial ecosystem, yet little is known about how the professional composition of their evaluation panels affects the predictive validity of startup selection decisions. Does adding other professional backgrounds to evaluation panels that are typically only composed of investors introduce noise that obscures venture quality, or does it provide valid signals that enhance prediction? Drawing on 91,423 evaluations of 19,152 early-stage startups submitted to a large global accelerator between 2016 and 2024, we examine whether professionally diverse panels predict venture success more accurately than homogeneous panels. Leveraging stratified random assignment of evaluators to startups, we find that diverse panels significantly improve funding and survival prediction for high-scoring ventures. Investors assign systematically lower recommendation scores than entrepreneurs and executives, reflecting more conservative screening thresholds. Analysis of written feedback shows that all evaluators rely on shared baseline standards but diverge in their emphasis on market opportunity, team strength, competitive positioning, and execution, leading diverse panels to include a broader and more complementary set of evaluative perspectives. We further show that feedback diversity is a key mechanism linking panel composition to predictive accuracy, as panels that draw on more varied evaluative dimensions produce recommendation scores that more strongly predict long-term outcomes. Our findings demonstrate that professional diversity is signal, not noise, capturing complementary screening heuristics that enhance predictive accuracy in high-uncertainty selection contexts.