diff --git a/litellm/integrations/focus/transformer.py b/litellm/integrations/focus/transformer.py
index b7d28e3dbb9..b3f321f9773 100644
--- a/litellm/integrations/focus/transformer.py
+++ b/litellm/integrations/focus/transformer.py
@@ -42,6 +42,12 @@ def _build_tags_expr(available_keys: list[str]) -> pl.Expr:
     )
 
 
+def _non_blank_string_expr(column_name: str) -> pl.Expr:
+    """Return a stripped string column, treating null/blank values as null."""
+    value = pl.col(column_name).cast(pl.String).str.strip_chars()
+    return pl.when(value.is_not_null() & (value != "")).then(value).otherwise(None)
+
+
 class FocusTransformer:
     """Transforms LiteLLM DB rows into Focus-compatible schema."""
 
@@ -81,6 +87,12 @@ class FocusTransformer:
 
         none_str = pl.lit(None, dtype=pl.Utf8)
         none_dec = pl.lit(None, dtype=pl.Decimal(18, 6))
+        service_name = pl.coalesce(
+            _non_blank_string_expr("model_group"),
+            _non_blank_string_expr("model"),
+            _non_blank_string_expr("custom_llm_provider"),
+            pl.lit("litellm-proxy"),
+        )
 
         return frame.select(
             dec(pl.col("spend").fill_null(0.0)).alias("BilledCost"),
@@ -122,7 +134,7 @@ class FocusTransformer:
             pl.col("model").cast(pl.String).alias("ResourceType"),
             pl.lit("AI and Machine Learning").alias("ServiceCategory"),
             pl.lit("Generative AI").alias("ServiceSubcategory"),
-            pl.col("model_group").cast(pl.String).alias("ServiceName"),
+            service_name.alias("ServiceName"),
             pl.col("team_id").cast(pl.String).alias("SubAccountId"),
             pl.col("team_alias").cast(pl.String).alias("SubAccountName"),
             none_str.alias("SubAccountType"),
diff --git a/tests/test_litellm/integrations/focus/test_transformer.py b/tests/test_litellm/integrations/focus/test_transformer.py
new file mode 100644
index 00000000000..c7b14b39ede
--- /dev/null
+++ b/tests/test_litellm/integrations/focus/test_transformer.py
@@ -0,0 +1,51 @@
+"""Tests for FocusTransformer normalization."""
+
+from __future__ import annotations
+
+import polars as pl
+
+from litellm.integrations.focus.transformer import FocusTransformer
+
+
+def _usage_frame(rows: list[dict]) -> pl.DataFrame:
+    defaults = {
+        "date": "2026-05-17",
+        "user_id": "user-1",
+        "api_key": "virtual-key-id",
+        "api_key_alias": "test-key",
+        "model": "gpt-4.1",
+        "model_group": "gpt-4.1",
+        "custom_llm_provider": "openai",
+        "spend": 0.1,
+        "team_id": "team-1",
+        "team_alias": "Test Team",
+        "user_email": "user@example.com",
+    }
+    return pl.DataFrame([{**defaults, **row} for row in rows])
+
+
+def test_should_use_model_group_as_service_name_when_present():
+    result = FocusTransformer().transform(
+        _usage_frame([{"model": "gpt-4.1", "model_group": "production-gpt"}])
+    )
+
+    assert result["ServiceName"].to_list() == ["production-gpt"]
+
+
+def test_should_fallback_service_name_when_model_group_is_blank():
+    result = FocusTransformer().transform(
+        _usage_frame(
+            [
+                {"model": "mcp-vector-store", "model_group": ""},
+                {"model": "", "model_group": "  ", "custom_llm_provider": "proxy"},
+                {"model": None, "model_group": None, "custom_llm_provider": None},
+            ]
+        )
+    )
+
+    assert result["ServiceName"].to_list() == [
+        "mcp-vector-store",
+        "proxy",
+        "litellm-proxy",
+    ]
+    assert all(service_name.strip() for service_name in result["ServiceName"])