diff --git a/src/mozanalysis/experiment.py b/src/mozanalysis/experiment.py
index 85968c1d..15e8028b 100644
--- a/src/mozanalysis/experiment.py
+++ b/src/mozanalysis/experiment.py
@@ -1458,12 +1458,20 @@ class AnalysisWindow:
 
     @start.validator
     def _validate_start(self, attribute, value):
-        assert (value >= 0 and self.end >= 0) or (value < 0 and self.end < 0)
+        assert (value >= 0 and self.end >= 0) or (value < 0 and self.end < 0), (
+            f"AnalysisWindow `start` invalid: {value} and {self.end} must share a sign",
+            "(either both positive/zero, or both negative)",
+        )
 
     @end.validator
     def _validate_end(self, attribute, value):
-        assert value >= self.start
-        assert (value >= 0 and self.start >= 0) or (value < 0 and self.start < 0)
+        assert value >= self.start, (
+            f"AnalysisWindow `end` invalid: {value} must be >= {self.start}"
+        )
+        assert (value >= 0 and self.start >= 0) or (value < 0 and self.start < 0), (
+            f"AnalysisWindow `end` invalid: {value} and {self.start} must share a sign",
+            "(either both positive/zero, or both negative)",
+        )
 
 
 @attr.s(frozen=True, slots=True)
diff --git a/tests/test_experiment.py b/tests/test_experiment.py
index 5f47d5b6..dddd39c9 100644
--- a/tests/test_experiment.py
+++ b/tests/test_experiment.py
@@ -25,6 +25,12 @@
 from mozanalysis.segments import Segment, SegmentDataSource
 
 
+@pytest.mark.parametrize(("start", "end", "invalid"), [(-1, 1, "start"), (3, 1, "end")])
+def test_analysis_window_validation(start, end, invalid):
+    with pytest.raises(AssertionError, match=f"AnalysisWindow `{invalid}` invalid"):
+        AnalysisWindow(start, end)
+
+
 def test_time_limits_validates():
     # Mainly check that the validation is running at all
     # No need to specify the same checks twice(?)