diff --git a/tests/integration/test_nlp_query_integration.py b/tests/integration/test_nlp_query_integration.py
index a7ee9d87..16c4543e 100644
--- a/tests/integration/test_nlp_query_integration.py
+++ b/tests/integration/test_nlp_query_integration.py
@@ -362,9 +362,12 @@ class TestNLPQueryServiceIntegration:
             error=None
         )
         
-        custom_processor.client.return_value.request = AsyncMock(
+        # Mock flow context to return prompt service responses
+        mock_prompt_service = AsyncMock()
+        mock_prompt_service.request = AsyncMock(
             side_effect=[phase1_response, phase2_response]
         )
+        flow.side_effect = lambda service_name: mock_prompt_service if service_name == "prompt-request" else flow_response if service_name == "response" else AsyncMock()
         
         # Act
         await custom_processor.on_message(msg, consumer, flow)
@@ -374,7 +377,7 @@ class TestNLPQueryServiceIntegration:
         assert custom_processor.graphql_generation_template == "custom-graphql-generator"
         
         # Verify the calls were made
-        assert custom_processor.client.return_value.request.call_count == 2
+        assert mock_prompt_service.request.call_count == 2
 
     @pytest.mark.asyncio 
     async def test_large_schema_set_integration(self, integration_processor):
@@ -466,30 +469,36 @@ class TestNLPQueryServiceIntegration:
             messages.append(msg)
             flows.append(flow)
         
-        # Mock responses for all requests
-        mock_responses = []
-        for i in range(10):  # 2 calls per request (phase1 + phase2)
-            if i % 2 == 0:  # Phase 1 responses
-                mock_responses.append(PromptResponse(
-                    text=json.dumps(["customers"]),
-                    error=None
-                ))
-            else:  # Phase 2 responses
-                mock_responses.append(PromptResponse(
-                    text=json.dumps({
-                        "query": f"query {{ customers {{ id name }} }}",
-                        "variables": {},
-                        "confidence": 0.9
-                    }),
-                    error=None
-                ))
-        
-        # Mock the flow context to return prompt service responses  
-        prompt_service = AsyncMock()
-        prompt_service.request = AsyncMock(
-            side_effect=mock_responses
-        )
-        flow.side_effect = lambda service_name: prompt_service if service_name == "prompt-request" else flow_response if service_name == "response" else AsyncMock()
+        # Mock responses for all requests - create individual prompt services for each flow
+        prompt_services = []
+        for i in range(5):  # 5 concurrent requests
+            phase1_response = PromptResponse(
+                text=json.dumps(["customers"]),
+                error=None
+            )
+            phase2_response = PromptResponse(
+                text=json.dumps({
+                    "query": f"query {{ customers {{ id name }} }}",
+                    "variables": {},
+                    "confidence": 0.9
+                }),
+                error=None
+            )
+            
+            # Create a prompt service for this request
+            prompt_service = AsyncMock()
+            prompt_service.request = AsyncMock(
+                side_effect=[phase1_response, phase2_response]
+            )
+            prompt_services.append(prompt_service)
+            
+            # Set up the flow for this request
+            flow_response = flows[i].return_value
+            flows[i].side_effect = lambda service_name, ps=prompt_service, fr=flow_response: (
+                ps if service_name == "prompt-request" else 
+                fr if service_name == "response" else 
+                AsyncMock()
+            )
         
         # Act - Process all messages concurrently
         import asyncio
@@ -503,7 +512,8 @@ class TestNLPQueryServiceIntegration:
         await asyncio.gather(*tasks)
         
         # Assert - All requests should be processed
-        assert prompt_service.request.call_count == 10
+        total_calls = sum(ps.request.call_count for ps in prompt_services)
+        assert total_calls == 10  # 2 calls per request (phase1 + phase2)
         for flow in flows:
             flow.return_value.send.assert_called_once()
 
diff --git a/tests/integration/test_structured_query_integration.py b/tests/integration/test_structured_query_integration.py
index b06aebc4..3ad3fc36 100644
--- a/tests/integration/test_structured_query_integration.py
+++ b/tests/integration/test_structured_query_integration.py
@@ -562,34 +562,50 @@ class TestStructuredQueryServiceIntegration:
             messages.append(msg)
             flows.append(flow)
         
-        # Mock responses for all requests (6 total: 3 NLP + 3 Objects)
-        mock_responses = []
-        for i in range(6):
-            if i % 2 == 0:  # NLP responses
-                mock_responses.append(QuestionToStructuredQueryResponse(
-                    error=None,
-                    graphql_query=f'query {{ test_{i//2} {{ id }} }}',
-                    variables={},
-                    detected_schemas=[f"test_{i//2}"],
-                    confidence=0.9
-                ))
-            else:  # Objects responses  
-                mock_responses.append(ObjectsQueryResponse(
-                    error=None,
-                    data=f'{{"test_{i//2}": [{{"id": "{i//2}"}}]}}',
-                    errors=None,
-                    extensions={}
-                ))
+        # Set up individual flow routing for each concurrent request
+        service_call_count = 0
         
-        call_count = 0
-        def mock_client_side_effect(name):
-            nonlocal call_count
-            client = AsyncMock()
-            client.request.return_value = mock_responses[call_count]
-            call_count += 1
-            return client
-        
-        integration_processor.client.side_effect = mock_client_side_effect
+        for i in range(3):  # 3 concurrent requests
+            # Create NLP and Objects responses for this request
+            nlp_response = QuestionToStructuredQueryResponse(
+                error=None,
+                graphql_query=f'query {{ test_{i} {{ id }} }}',
+                variables={},
+                detected_schemas=[f"test_{i}"],
+                confidence=0.9
+            )
+            
+            objects_response = ObjectsQueryResponse(
+                error=None,
+                data=f'{{"test_{i}": [{{"id": "{i}"}}]}}',
+                errors=None,
+                extensions={}
+            )
+            
+            # Create mock services for this request
+            mock_nlp_client = AsyncMock()
+            mock_nlp_client.request.return_value = nlp_response
+            
+            mock_objects_client = AsyncMock()
+            mock_objects_client.request.return_value = objects_response
+            
+            # Set up flow routing for this specific request
+            flow_response = flows[i].return_value
+            def create_flow_router(nlp_client, objects_client, response_producer):
+                def flow_router(service_name):
+                    nonlocal service_call_count
+                    service_call_count += 1
+                    if service_name == "nlp-query-request":
+                        return nlp_client
+                    elif service_name == "objects-query-request":
+                        return objects_client
+                    elif service_name == "response":
+                        return response_producer
+                    else:
+                        return AsyncMock()
+                return flow_router
+            
+            flows[i].side_effect = create_flow_router(mock_nlp_client, mock_objects_client, flow_response)
         
         # Act - Process all messages concurrently
         import asyncio