AbstractThis paper uses a novel dataset on investments by the European Bank for Reconstruction and Development to quantify a (sizeable) trade‐off between the impact and financial objectives of a large lender. The unique feature of this dataset is ex ante records of impact. These are made at the early stages of work on each transaction alongside probability‐of‐default scores. Impact scores are further updated at the final approval stage with around 55 percent of transaction concepts translating into signed deals. We show that this approach delivers a simultaneous selection of debt investments on the quality of credit and impact with a sizable trade‐off between pursuing commercial and development objectives. For commercially riskier investments, impact characteristics have a greater bearing on the probability of an investment going ahead. We further use machine‐learning analysis to show that the impact of some investments is strengthened prior to project approval.