diff --git a/scripts/bench.sh b/scripts/bench.sh
new file mode 100755
index 000000000..ec10ffc48
--- /dev/null
+++ b/scripts/bench.sh
@@ -0,0 +1,12 @@
+#! /usr/bin/env bash
+
+# Usage: bench.sh
+#
+# Runs random:bench and outputs the results in a CSV file with the name
+# "<git branch>-<git commit>.csv".
+
+set -euo pipefail
+
+OUTPUT="$(git branch --show-current)-$(git rev-parse --short HEAD).csv"
+stack bench random:bench --ba "--csv=${OUTPUT} --small"
+echo "Results are in $OUTPUT"
diff --git a/scripts/compare.py b/scripts/compare.py
new file mode 100755
index 000000000..c9cd63759
--- /dev/null
+++ b/scripts/compare.py
@@ -0,0 +1,59 @@
+#! /usr/bin/env nix-shell
+#! nix-shell -i python3 -p python3 python37Packages.pandas
+
+# Usage: compare.py <reference.csv> <results.csv>
+#
+# Compares the benchmark results in the second argument with those in the first
+# argument. Outputs a comparison of those rows where the results differ
+# significantly.
+
+import pandas as pd
+
+# Threshold for the relative difference, as a percentage
+SIGNIFICANT = 20
+
+if __name__ == "__main__":
+    from sys import argv
+
+    ref = pd.read_csv(argv[1])
+    res = pd.read_csv(argv[2])
+
+    # v1.1 benchmarks call it 'randomR', newer ones 'uniformR'. They are
+    # directly comparable. We normalise to 'uniformR' here.
+    ref["Name"] = ref["Name"].str.replace("randomR", "uniformR")
+    res["Name"] = res["Name"].str.replace("randomR", "uniformR")
+    comp = pd.merge(
+        ref,
+        res,
+        on="Name",
+        how="outer",
+        validate="one_to_one",
+        suffixes=("_ref", "_res"),
+    )
+
+    diff_col = "Diff_rel"
+    comp[diff_col] = (comp["Mean_ref"] - comp["Mean_res"]) / comp["Mean_res"]
+
+    cols = ["Name", "Mean_ref", "Mean_res", diff_col]
+    print_opts = {
+        "index": False,
+        "columns": cols,
+        "formatters": {diff_col: "{:,.0%}".format},
+    }
+    significant_percent = SIGNIFICANT / 100
+
+    slower = comp[comp[diff_col] < -significant_percent]
+    if slower.empty:
+        print("SLOWER: none")
+    else:
+        print("SLOWER")
+        print(slower.to_string(**print_opts))
+
+    print()
+
+    faster = comp[comp[diff_col] > significant_percent]
+    if faster.empty:
+        print("FASTER: none")
+    else:
+        print("\nFASTER")
+        print(faster.to_string(**print_opts))