diff --git a/tap_stripe/__init__.py b/tap_stripe/__init__.py
index 197952c6..a6f7cf10 100755
--- a/tap_stripe/__init__.py
+++ b/tap_stripe/__init__.py
@@ -284,6 +284,57 @@ def discover():
 
     return {'streams': streams}
 
+def value_at_breadcrumb(breadcrumb, rec):
+    if len(breadcrumb) == 1:
+        return rec.get(breadcrumb[0])
+    if rec.get(breadcrumb[0]):
+        return value_at_breadcrumb(breadcrumb[1:], rec[breadcrumb[0]])
+    return None
+
+
+def insert_at_breadcrumb(breadcrumb, value, rec):
+    if len(breadcrumb) == 1:
+        rec[breadcrumb[0]] = value
+    else:
+        if rec.get(breadcrumb[0]):
+            insert_at_breadcrumb(breadcrumb[1:], value, rec[breadcrumb[0]])
+        else:
+            rec[breadcrumb[0]] = {}
+            insert_at_breadcrumb(breadcrumb[1:], value, rec[breadcrumb[0]])
+
+
+def apply_whitelist(rec, stream_field_whitelist):
+    """The whitelist is a map from stream_name to a list of breadcrumbs that
+    indicates which nested fields should be persisted. There shouldn't be
+    any top-level fields in the whitelist since users can already remove
+    these fields via field selection. The shape of the whitelist is:
+    {
+       <stream_name>: [
+          [<field_1_breadcrumb>],
+          [<field_2_breadcrumb>],
+          ...
+          [<field_n_breadcrumb>]
+       ]
+    }
+    For now, the top level field should always be 'data' until we hear of
+    a need to extend this to other deeply nested objects
+    """
+    filtered_rec = {}
+
+    # Keep all the top level fields
+    for k, v in rec.items(): #pylint: disable=invalid-name
+        if not isinstance(v, (dict, list)):
+            filtered_rec[k] = v
+
+    for breadcrumb in stream_field_whitelist:
+        assert len(breadcrumb) > 1
+        assert breadcrumb[0] == 'data'
+        value_to_add = value_at_breadcrumb(breadcrumb, rec)
+        if value_to_add:
+            insert_at_breadcrumb(breadcrumb, value_to_add, filtered_rec)
+    return filtered_rec
+
+
 def reduce_foreign_keys(rec, stream_name):
     if stream_name == 'customers':
         rec['subscriptions'] = [s['id'] for s in rec.get('subscriptions', [])] or None
@@ -301,6 +352,7 @@ def reduce_foreign_keys(rec, stream_name):
                     rec['lines'][k] = [li.to_dict_recursive() for li in val]
     return rec
 
+# pylint: disable=too-many-locals
 def sync_stream(stream_name):
     """
     Sync each stream, looking for newly created records. Updates are captured by events stream.
@@ -308,6 +360,8 @@ def sync_stream(stream_name):
     LOGGER.info("Started syncing stream %s", stream_name)
 
     stream_metadata = metadata.to_map(Context.get_catalog_entry(stream_name)['metadata'])
+    stream_field_whitelist = json.loads(Context.config.get('whitelist_map', '{}')).get(stream_name)
+
     extraction_time = singer.utils.now()
     replication_key = metadata.get(stream_metadata, (), 'valid-replication-keys')[0]
     # Invoice Items bookmarks on `date`, but queries on `created`
@@ -361,6 +415,12 @@ def sync_stream(stream_name):
                                             Context.get_catalog_entry(stream_name)['schema'],
                                             stream_metadata)
 
+                # At this point, the record has been transformed and so
+                # any de-selected fields have been pruned. Now, prune off
+                # any fields that aren't present in the whitelist.
+                if stream_field_whitelist:
+                    rec = apply_whitelist(rec, stream_field_whitelist)
+
                 singer.write_record(stream_name,
                                     rec,
                                     time_extracted=extraction_time)