diff --git a/lib/optimizely.rb b/lib/optimizely.rb
index 1894ed18..c64ea794 100644
--- a/lib/optimizely.rb
+++ b/lib/optimizely.rb
@@ -137,9 +137,9 @@ def initialize(
       # Initialize CMAB components if cmab service is nil
       if cmab_service.nil?
         @cmab_client = DefaultCmabClient.new(
-          nil,
-          CmabRetryConfig.new,
-          @logger
+          http_client: nil,
+          retry_config: CmabRetryConfig.new,
+          logger: @logger
         )
         @cmab_cache = LRUCache.new(Optimizely::DefaultCmabCacheOptions::DEFAULT_CMAB_CACHE_SIZE, Optimizely::DefaultCmabCacheOptions::DEFAULT_CMAB_CACHE_TIMEOUT)
         @cmab_service = DefaultCmabService.new(
diff --git a/lib/optimizely/cmab/cmab_client.rb b/lib/optimizely/cmab/cmab_client.rb
index 8b6b7fa1..229eb11a 100644
--- a/lib/optimizely/cmab/cmab_client.rb
+++ b/lib/optimizely/cmab/cmab_client.rb
@@ -43,15 +43,22 @@ class DefaultCmabClient
     # Client for interacting with the CMAB service.
     # Provides methods to fetch decisions with optional retry logic.
 
-    def initialize(http_client = nil, retry_config = nil, logger = nil)
+    def initialize(http_client: nil, retry_config: nil, logger: nil, prediction_endpoint: nil)
       # Initialize the CMAB client.
       # Args:
       #   http_client: HTTP client for making requests.
       #   retry_config: Configuration for retry settings.
       #   logger: Logger for logging errors and info.
+      #   prediction_endpoint: Custom prediction endpoint URL template.
+      #                        Use #{rule_id} as placeholder for rule_id.
       @http_client = http_client || DefaultHttpClient.new
       @retry_config = retry_config || CmabRetryConfig.new
       @logger = logger || NoOpLogger.new
+      @prediction_endpoint = if prediction_endpoint.to_s.strip.empty?
+                               'https://prediction.cmab.optimizely.com/predict/%s'
+                             else
+                               prediction_endpoint
+                             end
     end
 
     def fetch_decision(rule_id, user_id, attributes, cmab_uuid, timeout: MAX_WAIT_TIME)
@@ -64,7 +71,7 @@ def fetch_decision(rule_id, user_id, attributes, cmab_uuid, timeout: MAX_WAIT_TI
       #   timeout: Maximum wait time for the request to respond in seconds. (default is 10 seconds).
       # Returns:
       #   The variation ID.
-      url = "https://prediction.cmab.optimizely.com/predict/#{rule_id}"
+      url = format(@prediction_endpoint, rule_id)
       cmab_attributes = attributes.map { |key, value| {'id' => key.to_s, 'value' => value, 'type' => 'custom_attribute'} }
 
       request_body = {
diff --git a/lib/optimizely/helpers/sdk_settings.rb b/lib/optimizely/helpers/sdk_settings.rb
index 3ca2dc72..6ccc82f8 100644
--- a/lib/optimizely/helpers/sdk_settings.rb
+++ b/lib/optimizely/helpers/sdk_settings.rb
@@ -22,7 +22,7 @@ module Optimizely
   module Helpers
     class OptimizelySdkSettings
       attr_accessor :odp_disabled, :segments_cache_size, :segments_cache_timeout_in_secs, :odp_segments_cache, :odp_segment_manager,
-                    :odp_event_manager, :fetch_segments_timeout, :odp_event_timeout, :odp_flush_interval
+                    :odp_event_manager, :fetch_segments_timeout, :odp_event_timeout, :odp_flush_interval, :cmab_prediction_endpoint
 
       # Contains configuration used for Optimizely Project initialization.
       #
@@ -35,6 +35,7 @@ class OptimizelySdkSettings
       # @param odp_segment_request_timeout - Time to wait in seconds for fetch_qualified_segments (optional. default = 10).
       # @param odp_event_request_timeout - Time to wait in seconds for send_odp_events (optional. default = 10).
       # @param odp_event_flush_interval - Time to wait in seconds for odp events to accumulate before sending (optional. default = 1).
+      # @param cmab_prediction_endpoint - Custom CMAB prediction endpoint URL template (optional). Use %s as placeholder for rule_id. Defaults to production endpoint if not provided.
       def initialize(
         disable_odp: false,
         segments_cache_size: Constants::ODP_SEGMENTS_CACHE_CONFIG[:DEFAULT_CAPACITY],
@@ -44,7 +45,8 @@ def initialize(
         odp_event_manager: nil,
         odp_segment_request_timeout: nil,
         odp_event_request_timeout: nil,
-        odp_event_flush_interval: nil
+        odp_event_flush_interval: nil,
+        cmab_prediction_endpoint: nil
       )
         @odp_disabled = disable_odp
         @segments_cache_size = segments_cache_size
@@ -55,6 +57,7 @@ def initialize(
         @fetch_segments_timeout = odp_segment_request_timeout
         @odp_event_timeout = odp_event_request_timeout
         @odp_flush_interval = odp_event_flush_interval
+        @cmab_prediction_endpoint = cmab_prediction_endpoint
       end
     end
   end
diff --git a/lib/optimizely/optimizely_factory.rb b/lib/optimizely/optimizely_factory.rb
index f2d5ad72..dbf1410b 100644
--- a/lib/optimizely/optimizely_factory.rb
+++ b/lib/optimizely/optimizely_factory.rb
@@ -119,6 +119,12 @@ def self.cmab_custom_cache(custom_cache)
       @cmab_custom_cache = custom_cache
     end
 
+    # Convenience method for setting custom CMAB prediction endpoint.
+    # @param prediction_endpoint String - Custom URL template for CMAB prediction API. Use %s as placeholder for rule_id.
+    def self.cmab_prediction_endpoint(prediction_endpoint)
+      @cmab_prediction_endpoint = prediction_endpoint
+    end
+
     # Returns a new optimizely instance.
     #
     # @params sdk_key - Required String uniquely identifying the fallback datafile corresponding to project.
@@ -202,7 +208,11 @@ def self.custom_instance( # rubocop:disable Metrics/ParameterLists
       )
 
       # Initialize CMAB components
-      cmab_client = DefaultCmabClient.new(logger: logger)
+      cmab_prediction_endpoint = nil
+      cmab_prediction_endpoint = settings.cmab_prediction_endpoint if settings&.cmab_prediction_endpoint
+      cmab_prediction_endpoint ||= @cmab_prediction_endpoint
+
+      cmab_client = DefaultCmabClient.new(logger: logger, prediction_endpoint: cmab_prediction_endpoint)
       cmab_cache = @cmab_custom_cache || LRUCache.new(
         @cmab_cache_size || Optimizely::DefaultCmabCacheOptions::DEFAULT_CMAB_CACHE_SIZE,
         @cmab_cache_ttl || Optimizely::DefaultCmabCacheOptions::DEFAULT_CMAB_CACHE_TIMEOUT
diff --git a/spec/cmab/cmab_client_spec.rb b/spec/cmab/cmab_client_spec.rb
index daa9ccd4..ae348fab 100644
--- a/spec/cmab/cmab_client_spec.rb
+++ b/spec/cmab/cmab_client_spec.rb
@@ -55,7 +55,7 @@
   end
 
   context 'when client is configured without retries' do
-    let(:client) { described_class.new(nil, Optimizely::CmabRetryConfig.new(max_retries: 0), spy_logger) }
+    let(:client) { described_class.new(http_client: nil, retry_config: Optimizely::CmabRetryConfig.new(max_retries: 0), logger: spy_logger) }
 
     it 'should return the variation id on success' do
       WebMock.stub_request(:post, expected_url)
@@ -132,7 +132,7 @@
   end
 
   context 'when client is configured with retries' do
-    let(:client_with_retry) { described_class.new(nil, retry_config, spy_logger) }
+    let(:client_with_retry) { described_class.new(http_client: nil, retry_config: retry_config, logger: spy_logger) }
 
     it 'should return the variation id on first try' do
       WebMock.stub_request(:post, expected_url)
@@ -195,4 +195,54 @@
       expect(spy_logger).to have_received(:log).with(Logger::ERROR, a_string_including('Max retries exceeded for CMAB request'))
     end
   end
+
+  context 'when custom prediction endpoint is configured' do
+    let(:custom_endpoint) { 'https://custom.endpoint.com/predict/%s' }
+    let(:custom_url) { 'https://custom.endpoint.com/predict/test_rule' }
+    let(:client_with_custom_endpoint) { described_class.new(http_client: nil, retry_config: Optimizely::CmabRetryConfig.new(max_retries: 0), logger: spy_logger, prediction_endpoint: custom_endpoint) }
+
+    it 'should use the custom prediction endpoint' do
+      WebMock.stub_request(:post, custom_url)
+             .with(body: expected_body_for_webmock, headers: expected_headers)
+             .to_return(status: 200, body: {'predictions' => [{'variation_id' => 'custom123'}]}.to_json, headers: {'Content-Type' => 'application/json'})
+
+      result = client_with_custom_endpoint.fetch_decision(rule_id, user_id, attributes, cmab_uuid)
+
+      expect(result).to eq('custom123')
+      expect(WebMock).to have_requested(:post, custom_url)
+                     .with(body: expected_body_for_webmock, headers: expected_headers).once
+    end
+  end
+
+  context 'when no prediction endpoint is provided' do
+    let(:client_with_default) { described_class.new(http_client: nil, retry_config: Optimizely::CmabRetryConfig.new(max_retries: 0), logger: spy_logger, prediction_endpoint: nil) }
+
+    it 'should use the default prediction endpoint' do
+      WebMock.stub_request(:post, expected_url)
+             .with(body: expected_body_for_webmock, headers: expected_headers)
+             .to_return(status: 200, body: {'predictions' => [{'variation_id' => 'default123'}]}.to_json, headers: {'Content-Type' => 'application/json'})
+
+      result = client_with_default.fetch_decision(rule_id, user_id, attributes, cmab_uuid)
+
+      expect(result).to eq('default123')
+      expect(WebMock).to have_requested(:post, expected_url)
+                     .with(body: expected_body_for_webmock, headers: expected_headers).once
+    end
+  end
+
+  context 'when empty string prediction endpoint is provided' do
+    let(:client_with_empty_endpoint) { described_class.new(http_client: nil, retry_config: Optimizely::CmabRetryConfig.new(max_retries: 0), logger: spy_logger, prediction_endpoint: '') }
+
+    it 'should fall back to the default prediction endpoint' do
+      WebMock.stub_request(:post, expected_url)
+             .with(body: expected_body_for_webmock, headers: expected_headers)
+             .to_return(status: 200, body: {'predictions' => [{'variation_id' => 'fallback123'}]}.to_json, headers: {'Content-Type' => 'application/json'})
+
+      result = client_with_empty_endpoint.fetch_decision(rule_id, user_id, attributes, cmab_uuid)
+
+      expect(result).to eq('fallback123')
+      expect(WebMock).to have_requested(:post, expected_url)
+                     .with(body: expected_body_for_webmock, headers: expected_headers).once
+    end
+  end
 end