diff --git a/docs/python/file_formats.rst b/docs/python/file_formats.rst
index 5c8751e..4511b12 100644
--- a/docs/python/file_formats.rst
+++ b/docs/python/file_formats.rst
@@ -3,6 +3,8 @@ File formats
 
 We read and write a lot of CSV and JSON files. Their format should be consistent.
 
+.. _format-json:
+
 JSON
 ----
 
diff --git a/docs/python/performance.rst b/docs/python/performance.rst
index f6da387..fa8b3f3 100644
--- a/docs/python/performance.rst
+++ b/docs/python/performance.rst
@@ -9,8 +9,27 @@ Performance
 
    :ref:`Django performance<django-performance>`
 
-CPU profiling
--------------
+Software performance depends on many choices: language (like :doc:`../rust/index` versus Python), framework (like `FastAPI versus Django <https://www.techempower.com/benchmarks/>`__), architecture (e.g. map-reduce), networking (e.g. batch requests), etc. Many choices are costly to change at a later date (e.g. full rewrite).
+
+Profiling
+---------
+
+Use profiling to:
+
+-  Identify slow dependencies, in case faster alternatives can be easily swapped in
+-  Find major hotspots, like a loop that runs in exponential time instead of quadratic time
+-  Find minor hotspots, if changing language, etc. is too costly
+
+Once a hotspot is found, the solution might be to:
+
+-  Call it once, via refactoring: for example, traversing JSON once for all CoVE calculations, instead of once for each calculation, in `lib-cove <https://github.com/OpenDataServices/lib-cove/issues/65>`__
+-  Call it less, via batching: for example, :ref:`reducing the number of SQL queries in Django projects<django-performance>`
+-  Cache the results: for example, `caching mergers in Kingfisher Process <https://github.com/open-contracting/kingfisher-process/blob/c4b05204faf08d00ed7914a41c2fd0770e0f6b3e/process/processors/compiler.py#L52>`__
+-  Process in parallel: for example, distributing work to multiple threads, like we do with :doc:`../services/rabbitmq`
+-  Replace it entirely: for example, using the :ref:`orjson<format-json>` package instead of the ``json`` library
+
+CPU
+~~~
 
 For example:
 
@@ -22,8 +41,8 @@ For example:
 
 To see where a running program is spending its time, use `py-spy top <https://github.com/benfred/py-spy>`__.
 
-Memory profiling
-----------------
+Memory
+~~~~~~
 
 For example:
 
@@ -33,7 +52,6 @@ For example:
    time mprof run libcoveoc4ids data.json
    mprof plot
 
-
 Reference
 ---------