We analyze the predictive performance of CSEM using a large statistical database. The prediction strength is quantified by comparing the CSEM interpretation to exploration drilling results for more than 100 wells in Norway. The comparison has been done by correlating inversion results for all surveys covering these wells with the well outcome, using a statistically driven anomaly detection workflow in order to avoid confirmation bias. The comparison is summarized by classifying the interpretations as true positives, true negatives, false positives or false negatives. We find that the CSEM interpretation correctly identified the true negative and positive cases for 79 % of the analyzed wells. We further show how integrated interpretation can provide more detailed predictions. This includes taking the sensitivity to the target into account, as well as integrating seismic data and rock physics parameters with the CSEM data in order to quantify the potential volumes in place. In some cases, we also see that the derived parameters are not compatible with hydrocarbon models, and prospects must be downgraded despite having clear CSEM anomalies associated with them. In addition to the statistics, our results are supported by several case examples.