diff --git a/README.rst b/README.rst
index 6e9bfaa..0d62a4b 100644
--- a/README.rst
+++ b/README.rst
@@ -269,6 +269,34 @@ botocore debug logs to print to stderr but not to Cloudwatch:
     logger = logging.getLogger()
     logger.addHandler(watchtower.CloudWatchLogHandler())
 
+Resiliency to network outages
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+On startup, watchtower queries the configured log group, and creates it if it does not exist. A failure
+to query or create the log group for any reason will cause watchtower to crash and prevent your application from starting.
+This is intended, since otherwise watchtower cannot deliver your logs. If you
+require your application to be able to start even during a network connectivity outage or a CloudWatch Logs service outage,
+consider setting the ``create_log_group`` parameter to ``False`` in the handler configuration. With this in place,
+watchtower is able to start completely offline. If the configuration is otherwise correct, watchtower will resume sending
+logs when it regains a connection to the service.
+
+When your logs are ready to be sent, watchtower uses the following
+`boto3 retry configuration <https://boto3.amazonaws.com/v1/documentation/api/latest/guide/retries.html>`_:
+
+- Standard retry mode
+- 8 retry attempts with exponential backoff
+
+This means watchtower will attempt to send your log batch 8 times over the course of FIXME seconds. If all attempts fail,
+watchotwer will give up and print a warning to stderr that the log batch was lost. If you want to use a custom configuration
+(for example, to increase the number of retries), you can supply one to the handler as shown in the following example:
+
+.. code-block:: python
+
+    import boto3
+    from botocore.config import Config
+    client_config = Config(connect_timeout=5, read_timeout=60, retries={"max_attempts": 20, "mode": "standard"})
+    logs_client = boto3.client("logs", config=client_config)
+    watchtower.CloudWatchLogHandler(boto3_client=logs_client)
+
 Authors
 ~~~~~~~
 * Andrey Kislyuk