diff --git a/polars/polars-lazy/polars-plan/src/dsl/function_expr/fill_null.rs b/polars/polars-lazy/polars-plan/src/dsl/function_expr/fill_null.rs
index cb9c39991249..0ba60976ab59 100644
--- a/polars/polars-lazy/polars-plan/src/dsl/function_expr/fill_null.rs
+++ b/polars/polars-lazy/polars-plan/src/dsl/function_expr/fill_null.rs
@@ -4,20 +4,16 @@ pub(super) fn fill_null(s: &[Series], super_type: &DataType) -> PolarsResult<Ser
     let array = &s[0];
     let fill_value = &s[1];
 
-    if matches!(super_type, DataType::Unknown) {
-        return Err(PolarsError::SchemaMisMatch(
-            format!(
-                "Cannot 'fill_null' a 'Series' of dtype: '{}' with an argument of dtype: '{}'",
-                array.dtype(),
-                fill_value.dtype()
-            )
-            .into(),
-        ));
+    let (array, fill_value) = if matches!(super_type, DataType::Unknown) {
+        let fill_value = fill_value.cast(array.dtype()).map_err(|_|{
+                let msg = "'fill_null' supertype could not be determined. Set the correct literal value or ensure the type of the expression is known .";
+            PolarsError::SchemaMisMatch(msg.into())
+        })?;
+        (array.clone(), fill_value)
+    } else {
+        (array.cast(super_type)?, fill_value.cast(super_type)?)
     };
 
-    let array = array.cast(super_type)?;
-    let fill_value = fill_value.cast(super_type)?;
-
     if !array.null_count() == 0 {
         Ok(array)
     } else {
diff --git a/py-polars/tests/unit/test_functions.py b/py-polars/tests/unit/test_functions.py
index bae197aa7609..1632176a03f7 100644
--- a/py-polars/tests/unit/test_functions.py
+++ b/py-polars/tests/unit/test_functions.py
@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+import typing
+
 import numpy as np
 import pytest
 
@@ -289,3 +291,29 @@ def test_overflow_diff() -> None:
     assert df.select(pl.col("a").cast(pl.UInt64).diff()).to_dict(False) == {
         "a": [None, -10, 20]
     }
+
+
+@typing.no_type_check
+def test_fill_null_unknown_output_type() -> None:
+    df = pl.DataFrame(
+        {
+            "a": [
+                None,
+                2,
+                3,
+                4,
+                5,
+            ]
+        }
+    )
+    assert df.with_columns(
+        np.exp(pl.col("a")).fill_null(pl.lit(1, pl.Float64))
+    ).to_dict(False) == {
+        "a": [
+            1.0,
+            7.38905609893065,
+            20.085536923187668,
+            54.598150033144236,
+            148.4131591025766,
+        ]
+    }