diff --git a/cloudbuild_CI.yaml b/cloudbuild_CI.yaml
index 4375decde..c7bbf2ab3 100644
--- a/cloudbuild_CI.yaml
+++ b/cloudbuild_CI.yaml
@@ -50,6 +50,7 @@ steps:
       - '--run_unit_tests'
       - '--run_presubmit_tests'
       - '--run_preprocessor_tests'
+      - '--run_bq_to_vcf_tests'
     id: 'test-gcp-variant-transforms-docker'
       # By default the script uses a GS bucket of gcp-variant-transforms-test
       # project. For other projects we should either use the following option
diff --git a/docs/bigquery_to_vcf.md b/docs/bigquery_to_vcf.md
index b258736d8..d01d4ffac 100644
--- a/docs/bigquery_to_vcf.md
+++ b/docs/bigquery_to_vcf.md
@@ -89,6 +89,11 @@ In addition, the following optional flags are provided:
   You may change this flag if you have a dataset that is very dense and variants
   in each shard cannot be sorted in memory.
 
+For large datasets, using
+[Cloud Dataflow Shuffle](https://cloud.google.com/dataflow/service/dataflow-service-desc#cloud-dataflow-shuffle)
+can speed up the pipeline, by specifying the parameter:
+`--experiments shuffle_mode=service`.
+
 ### Running from github
 
 In addition to using the docker image, you may run the pipeline directly from