diff --git a/.github/workflows/durations.yml b/.github/workflows/durations.yml
new file mode 100644
index 0000000000..d296d4c3e7
--- /dev/null
+++ b/.github/workflows/durations.yml
@@ -0,0 +1,63 @@
+name: Update Durations
+
+on:
+  # every Sunday at 00:00 UTC
+  # https://crontab.guru/#0_0_*_*_0
+  schedule:
+    - cron: '0 0 * * 0'
+
+  # https://docs.github.com/en/webhooks-and-events/webhooks/webhook-events-and-payloads#workflow_dispatch
+  workflow_dispatch:
+
+jobs:
+  update-durations:
+    runs-on: ubuntu-latest
+    permissions:
+      # necessary to open PR
+      # https://github.com/peter-evans/create-pull-request#action-inputs
+      contents: write
+      pull-requests: write
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: download recent artifacts
+        run: |
+          gh run list \
+            --branch main \
+            --workflow tests \
+            --limit 10 \
+            --json databaseId \
+            --jq '.[].databaseId' \
+          | xargs \
+            -n 1 \
+          gh run download \
+            --dir ${{ runner.temp }}/artifacts/ \
+            --pattern '*-all' \
+          || true
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
+
+      - uses: actions/setup-python@v4
+        with:
+          python-version: '3.11'
+
+      - name: combine recent durations from artifacts
+        run: python ./tools/durations/combine.py ${{ runner.temp }}/artifacts/
+
+      - name: create updated durations PR
+        uses: peter-evans/create-pull-request@v5
+        with:
+          push-to-fork: conda-bot/conda
+          token: ${{ secrets.DURATIONS_TOKEN }}
+          branch: update-durations
+          delete-branch: true
+          commit-message: Update test durations
+          author: Conda Bot <18747875+conda-bot@users.noreply.github.com>
+          committer: Conda Bot <18747875+conda-bot@users.noreply.github.com>
+          title: 🤖 Update test durations
+          body: |
+            Aggregate recent test durations for each test and update the durations file.
+
+            [durations.yml]: ${{ github.server_url }}/${{ github.repository }}/blob/main/.github/workflows/durations.yml
+
+            This PR was created automatically by the [`durations.yml`][durations.yml] workflow.
diff --git a/tools/durations/Linux.json b/tools/durations/Linux.json
new file mode 100644
index 0000000000..0967ef424b
--- /dev/null
+++ b/tools/durations/Linux.json
@@ -0,0 +1 @@
+{}
diff --git a/tools/durations/Windows.json b/tools/durations/Windows.json
new file mode 100644
index 0000000000..0967ef424b
--- /dev/null
+++ b/tools/durations/Windows.json
@@ -0,0 +1 @@
+{}
diff --git a/tools/durations/combine.py b/tools/durations/combine.py
new file mode 100644
index 0000000000..b42de633fe
--- /dev/null
+++ b/tools/durations/combine.py
@@ -0,0 +1,56 @@
+# Copyright (C) 2014 Anaconda, Inc
+# SPDX-License-Identifier: BSD-3-Clause
+"""
+Script to combine test durations from all runs.
+
+If the tests splits are looking uneven or the test suite has
+siginificantly changed, update ./tools/durations/${OS}.json in the root of the
+repository and pytest-split may work better.
+
+```
+$ gh run list --branch <branch>
+$ gh run download --dir ./artifacts/ <databaseId>
+$ python ./tools/durations/combine.py ./artifacts/
+$ git add ./tools/durations/
+$ git commit -m "Update test durations"
+$ git push
+```
+"""
+import json
+from pathlib import Path
+from statistics import fmean
+from sys import argv
+
+combined: dict[str, dict[str, list[float]]] = {}
+
+durations_dir = Path(__file__).parent
+artifacts_dir = Path(argv[-1]).expanduser().resolve()
+
+# aggregate all new durations
+for path in artifacts_dir.glob("**/*.json"):
+    os = path.stem
+    combined_os = combined.setdefault(os, {})
+
+    data = json.loads(path.read_text())
+    for key, value in data.items():
+        combined_os.setdefault(key, []).append(value)
+
+# aggregate new and old durations while discarding durations that no longer exist
+for path in durations_dir.glob("**/*.json"):
+    os = path.stem
+    combined_os = combined.setdefault(os, {})
+
+    data = json.loads(path.read_text())
+    for key in set(combined_os).intersection(durations_dir.glob("**/*.json")):
+        combined_os.setdefault(key, []).append(data[key])
+
+# write out averaged durations
+for os, combined_os in combined.items():
+    (durations_dir / f"{os}.json").write_text(
+        json.dumps(
+            {key: fmean(values) for key, values in combined_os.items()},
+            indent=4,
+            sort_keys=True,
+        )
+        + "\n"  # include trailing newline
+    )
diff --git a/tools/durations/macOS.json b/tools/durations/macOS.json
new file mode 100644
index 0000000000..0967ef424b
--- /dev/null
+++ b/tools/durations/macOS.json
@@ -0,0 +1 @@
+{}