Data defects in credit portfolios inflate expected‑loss models and regulatory capital, yet conventional data‑quality dashboards express this risk in abstract percentages that obscure monetary stakes. The goal of the article is to develop a data‑quality matrix that converts every data imperfection into an explicit euro cost. This paper presents the Loss‑Aware Data Quality Matrix (LADQM), a lightweight matrix that prices each missing or corrupted value by multiplying three ingredients: a loan’s expected credit loss, the SHAP‑based relevance of the feature, and the drop in model discrimination when that feature is masked. The matrix yields euro-denominated metrics by aggregation. A 300‑line open‑source pipeline implements LADQM in a model-agnostic way; experiments instantiate it with XGBoost, producing cell‑level heat maps and portfolio summaries of realised and avoidable loss. Experiments on three public datasets – a 149k‑row U.S. mortgage book, a 614‑row consumer‑loan sample and a 500‑row micro‑loan file – reveal how a monetary lens overturns naïve completeness rankings: the “dirtiest” micro‑loan file hides only €0.1 m in potential loss, while the “clean” mortgage data conceal €23 m. Fixing just three costliest columns in the mortgage book would cut potential loss by 27%, twice the benefit of targeting the most frequent blanks. Predictive performance remains stable: pruning three leaky features lowers AUC by 0.001 yet removes a four‑point inflation of recall and F1, averting capital under‑provisioning. Thus, LADQM re‑casts data quality from compliance chore to quantifiable risk lever. Limitations include exogenous remediation costs, focus on retail portfolios, and a batch implementation that must be refactored for streaming before real‑time deployment. Nevertheless, LADQM arms regulators, risk managers and data engineers with a common euro‑denominated language for prioritising the fixes that matter most.