diff --git a/backend/migration.py b/backend/migration.py
index f92ac22..071d38d 100644
--- a/backend/migration.py
+++ b/backend/migration.py
@@ -185,9 +185,12 @@ def load_csv_data(data_dir: Path) -> pd.DataFrame:
     return pd.DataFrame()
 
 
-def migrate_data(df: pd.DataFrame, geocode: bool = False):
+def migrate_data(df: pd.DataFrame, geocode: bool = False, geocode_cache: dict = None):
     """Migrate DataFrame data to database."""
 
+    if geocode_cache is None:
+        geocode_cache = {}
+
     # Clean URN column - convert to integer, drop invalid values
     df = df.copy()
     df["urn"] = pd.to_numeric(df["urn"], errors="coerce")
@@ -200,13 +203,25 @@ def migrate_data(df: pd.DataFrame, geocode: bool = False):
     )
     print(f"\nMigrating {len(school_data)} unique schools...")
 
-    # Geocode if requested
-    geocoded = {}
+    # Geocode postcodes that aren't already in the cache
+    geocoded = dict(geocode_cache)  # start with preserved coordinates
     if geocode and "postcode" in df.columns:
-        print("\nGeocoding postcodes...")
-        postcodes = df["postcode"].dropna().unique().tolist()
-        geocoded = geocode_postcodes_bulk(postcodes)
-        print(f"  Successfully geocoded {len(geocoded)} postcodes")
+        cached_postcodes = {
+            str(row.get("postcode", "")).strip().upper()
+            for _, row in school_data.iterrows()
+            if int(float(str(row.get("urn", 0) or 0))) in geocode_cache
+        }
+        postcodes_needed = [
+            p for p in df["postcode"].dropna().unique()
+            if str(p).strip().upper() not in cached_postcodes
+        ]
+        if postcodes_needed:
+            print(f"\nGeocoding {len(postcodes_needed)} postcodes ({len(geocode_cache)} restored from cache)...")
+            fresh = geocode_postcodes_bulk(postcodes_needed)
+            geocoded.update(fresh)
+            print(f"  Successfully geocoded {len(fresh)} new postcodes")
+        else:
+            print(f"\nAll {len(geocode_cache)} postcodes restored from cache, skipping geocoding.")
 
     with get_db_session() as db:
         # Create schools
@@ -396,6 +411,24 @@ def run_full_migration(geocode: bool = False) -> bool:
     Returns True if successful, False if no data found.
     Raises exception on error.
     """
+    # Preserve existing geocoding so a reimport doesn't throw away coordinates
+    # that took a long time to compute.
+    geocode_cache: dict[int, tuple[float, float]] = {}
+    inspector = __import__("sqlalchemy").inspect(engine)
+    if "schools" in inspector.get_table_names():
+        try:
+            with get_db_session() as db:
+                rows = db.execute(
+                    __import__("sqlalchemy").text(
+                        "SELECT urn, latitude, longitude FROM schools "
+                        "WHERE latitude IS NOT NULL AND longitude IS NOT NULL"
+                    )
+                ).fetchall()
+                geocode_cache = {r.urn: (r.latitude, r.longitude) for r in rows}
+                print(f"  Saved {len(geocode_cache)} existing geocoded coordinates.")
+        except Exception as e:
+            print(f"  Warning: could not save geocode cache: {e}")
+
     print("Dropping existing tables...")
     Base.metadata.drop_all(bind=engine)
 
@@ -409,5 +442,5 @@ def run_full_migration(geocode: bool = False) -> bool:
         print("Warning: No CSV data found to migrate!")
         return False
 
-    migrate_data(df, geocode=geocode)
+    migrate_data(df, geocode=geocode, geocode_cache=geocode_cache)
     return True