The net effects of algorithmic and high-frequency traders mask considerable heterogeneity in how they impact institutional transaction costs. Using regulatory data, we analyse the heterogeneity across individual trading accounts. We develop a method to identify subsets of traders causally related to higher institutional transaction costs and estimate that they add ten basis points to the cost of executing large institutional orders. Their effects are counteracted by traders that systematically decrease these costs. We find that fast traders and those with high order-to-trade ratios are no more likely to increase costs than others. Traders that increase costs are more active in small stocks.