diff --git a/import_flow_events.py b/import_flow_events.py
index f671737..fa05eed 100644
--- a/import_flow_events.py
+++ b/import_flow_events.py
@@ -79,7 +79,8 @@
       utm_content VARCHAR(40) ENCODE lzo,
       utm_medium VARCHAR(40) ENCODE lzo,
       utm_source VARCHAR(40) ENCODE lzo,
-      utm_term VARCHAR(40) ENCODE lzo
+      utm_term VARCHAR(40) ENCODE lzo,
+      export_date DATE NOT NULL ENCODE lzo
     );
 """
 Q_CREATE_EVENTS_TABLE = """
@@ -89,23 +90,30 @@
       -- but redshift doesn't support that.
       flow_time BIGINT NOT NULL ENCODE lzo,
       flow_id VARCHAR(64) NOT NULL DISTKEY ENCODE lzo,
-      type VARCHAR(64) NOT NULL ENCODE lzo
+      type VARCHAR(64) NOT NULL ENCODE lzo,
+      export_date DATE NOT NULL ENCODE lzo
     );
 """
 
 Q_CHECK_FOR_DAY = """
     SELECT timestamp FROM flow_events
-    WHERE timestamp::DATE = '{day}'::DATE
+    WHERE timestamp::DATE >= '{day}'::DATE - '1 day'::INTERVAL
+      AND timestamp::DATE <= '{day}'::DATE + '1 day'::INTERVAL
+      AND export_date = '{day}'::DATE
     LIMIT 1;
 """
 
-Q_CLEAR_DAY_METADATA = """
-    DELETE FROM flow_metadata
-    WHERE begin_time::DATE = '{day}'::DATE;
-"""
 Q_CLEAR_DAY_EVENTS = """
     DELETE FROM flow_events
-    WHERE timestamp::DATE = '{day}'::DATE;
+    WHERE timestamp::DATE >= '{day}'::DATE - '1 day'::INTERVAL
+      AND timestamp::DATE <= '{day}'::DATE + '1 day'::INTERVAL
+      AND export_date = '{day}'::DATE;
+"""
+Q_CLEAR_DAY_METADATA = """
+    DELETE FROM flow_metadata
+    WHERE begin_time::DATE >= '{day}'::DATE - '1 day'::INTERVAL
+      AND begin_time::DATE <= '{day}'::DATE + '1 day'::INTERVAL
+      AND export_date = '{day}'::DATE;
 """
 
 Q_COPY_CSV = """
@@ -148,7 +156,8 @@
       utm_content,
       utm_medium,
       utm_source,
-      utm_term
+      utm_term,
+      export_date
     )
     SELECT
       flow_id,
@@ -164,7 +173,8 @@
       utm_content,
       utm_medium,
       utm_source,
-      utm_term
+      utm_term,
+      '{day}'::DATE
     FROM temporary_raw_flow_data
     WHERE type LIKE 'flow%begin';
 """
@@ -173,7 +183,9 @@
     SET duration = durations.flow_time
     FROM (
       SELECT flow_id, MAX(flow_time) AS flow_time
-      FROM temporary_raw_flow_data
+      FROM flow_events
+      WHERE "timestamp" >= '{day}'::DATE
+        AND "timestamp" <= '{day}'::DATE + '1 day'::INTERVAL
       GROUP BY flow_id
     ) AS durations
     WHERE flow_metadata.flow_id = durations.flow_id
@@ -185,8 +197,10 @@
     SET completed = TRUE
     FROM (
       SELECT flow_id
-      FROM temporary_raw_flow_data
+      FROM flow_events
       WHERE type = 'flow.complete'
+        AND "timestamp" >= '{day}'::DATE
+        AND "timestamp" <= '{day}'::DATE + '1 day'::INTERVAL
     ) AS complete
     WHERE flow_metadata.flow_id = complete.flow_id
     AND flow_metadata.begin_time >= '{day}'::DATE - '1 day'::INTERVAL
@@ -197,8 +211,10 @@
     SET new_account = TRUE
     FROM (
       SELECT flow_id
-      FROM temporary_raw_flow_data
+      FROM flow_events
       WHERE type = 'account.created'
+        AND "timestamp" >= '{day}'::DATE
+        AND "timestamp" <= '{day}'::DATE + '1 day'::INTERVAL
     ) AS created
     WHERE flow_metadata.flow_id = created.flow_id
     AND flow_metadata.begin_time >= '{day}'::DATE - '1 day'::INTERVAL
@@ -242,13 +258,15 @@
       timestamp,
       flow_time,
       flow_id,
-      type
+      type,
+      export_date
     )
     SELECT
       'epoch'::TIMESTAMP + timestamp * '1 second'::INTERVAL,
       flow_time,
       flow_id,
-      type
+      type,
+      '{day}'::DATE
     FROM temporary_raw_flow_data;
 """
 
@@ -287,8 +305,8 @@ def import_events(force_reload=False):
             # Create the temporary table
             db.run(Q_CREATE_CSV_TABLE)
             # Clear any existing data for the day, to avoid duplicates.
-            db.run(Q_CLEAR_DAY_METADATA.format(day=day))
             db.run(Q_CLEAR_DAY_EVENTS.format(day=day))
+            db.run(Q_CLEAR_DAY_METADATA.format(day=day))
             s3path = EVENTS_FILE_URL.format(day=day)
             # Copy data from s3 into redshift
             db.run(Q_COPY_CSV.format(
@@ -296,13 +314,13 @@ def import_events(force_reload=False):
                 **CONFIG
             ))
             # Populate the flow_metadata table
-            db.run(Q_INSERT_METADATA)
+            db.run(Q_INSERT_METADATA.format(day=day))
             db.run(Q_UPDATE_DURATION.format(day=day))
             db.run(Q_UPDATE_COMPLETED.format(day=day))
             db.run(Q_UPDATE_NEW_ACCOUNT.format(day=day))
             db.run(Q_UPDATE_METRICS_CONTEXT.format(day=day))
             # Populate the flow_events table
-            db.run(Q_INSERT_EVENTS)
+            db.run(Q_INSERT_EVENTS.format(day=day))
             # Print the timestamps for sanity-checking.
             print "  MIN TIMESTAMP", db.one("SELECT MIN(timestamp) FROM temporary_raw_flow_data")
             print "  MAX TIMESTAMP", db.one("SELECT MAX(timestamp) FROM temporary_raw_flow_data")