diff --git a/app/jobs/mutex_application_job.rb b/app/jobs/mutex_application_job.rb
index 58c7cbf36..4143eab2a 100644
--- a/app/jobs/mutex_application_job.rb
+++ b/app/jobs/mutex_application_job.rb
@@ -14,6 +14,19 @@
 class MutexApplicationJob < ApplicationJob
   class LockAcquisitionError < StandardError; end
 
+  def self.retry_on_lock_conflict(wait:, attempts:, on_exhaustion: :raise)
+    retry_on LockAcquisitionError, wait: wait, attempts: attempts do |job, error|
+      raise error if on_exhaustion == :raise
+
+      job.public_send(on_exhaustion, *job.arguments)
+    end
+  end
+
+  # Redis::LockManager#unlock is not owner-checked. If a job runs past the TTL,
+  # Redis can expire the key, a newer job can acquire it, and the older job can
+  # then delete the newer job's lock on unlock. Current mutex users treat locks as
+  # short race dampeners, so this is acceptable for now. Future iterations should
+  # move Redis::LockManager to token-checked unlocks.
   def with_lock(lock_key, timeout = Redis::LockManager::LOCK_TIMEOUT)
     lock_manager = Redis::LockManager.new
 
diff --git a/app/jobs/webhooks/instagram_events_job.rb b/app/jobs/webhooks/instagram_events_job.rb
index 6383daff8..c27363127 100644
--- a/app/jobs/webhooks/instagram_events_job.rb
+++ b/app/jobs/webhooks/instagram_events_job.rb
@@ -1,6 +1,14 @@
 class Webhooks::InstagramEventsJob < MutexApplicationJob
   queue_as :default
-  retry_on LockAcquisitionError, wait: 1.second, attempts: 8
+  # This lock is only a short race dampener for first-message conversation creation.
+  # ContactInbox creation is already protected by a unique index, but conversation
+  # lookup is `find active conversation || create`, so concurrent first messages from
+  # the same IG contact can create duplicate conversations.
+  #
+  # ActiveJob retries are not FIFO, so a longer retry window does not preserve message
+  # order. Use deterministic backoff so the final attempt happens after the 3s lock TTL,
+  # then process without the lock instead of dropping the webhook.
+  retry_on_lock_conflict wait: ->(executions) { executions.seconds }, attempts: 3, on_exhaustion: :process_without_lock
 
   # @return [Array] We will support further events like reaction or seen in future
   SUPPORTED_EVENTS = [:message, :read].freeze
@@ -9,11 +17,19 @@ class Webhooks::InstagramEventsJob < MutexApplicationJob
     @entries = entries
 
     key = format(::Redis::Alfred::IG_MESSAGE_MUTEX, sender_id: contact_instagram_id, ig_account_id: ig_account_id)
-    with_lock(key) do
+    # Keep the lock TTL just long enough for the first job to fetch profile data and
+    # create the contact/conversation. A longer TTL would add user-visible latency for
+    # hot contacts without giving us ordering guarantees.
+    with_lock(key, 3.seconds) do
       process_entries(entries)
     end
   end
 
+  def process_without_lock(entries)
+    Rails.logger.warn("[#{self.class.name}] Processing without lock after lock retry exhaustion")
+    process_entries(entries)
+  end
+
   # https://developers.facebook.com/docs/messenger-platform/instagram/features/webhook
   def process_entries(entries)
     entries.each do |entry|
diff --git a/spec/jobs/mutex_application_job_spec.rb b/spec/jobs/mutex_application_job_spec.rb
index 91a56407d..4c8fa3394 100644
--- a/spec/jobs/mutex_application_job_spec.rb
+++ b/spec/jobs/mutex_application_job_spec.rb
@@ -55,4 +55,57 @@ RSpec.describe MutexApplicationJob do
       end.to raise_error(StandardError)
     end
   end
+
+  describe '.retry_on_lock_conflict' do
+    let(:job_class) do
+      Class.new(described_class) do
+        retry_on_lock_conflict wait: 1.second, attempts: 1, on_exhaustion: :process_without_lock
+
+        attr_reader :fallback_args
+
+        def perform(lock_key, _payload)
+          with_lock(lock_key) { raise 'lock should not be acquired' }
+        end
+
+        def process_without_lock(lock_key, payload)
+          @fallback_args = [lock_key, payload]
+        end
+      end
+    end
+
+    let(:payload) { { 'message' => 'hello' } }
+
+    before do
+      stub_const('LockConflictTestJob', job_class)
+    end
+
+    it 'runs the configured handler with the original job arguments when lock retries are exhausted' do
+      allow(lock_manager).to receive(:lock).with(lock_key, Redis::LockManager::LOCK_TIMEOUT).and_return(false)
+
+      job = job_class.new(lock_key, payload)
+
+      expect { job.perform_now }.not_to raise_error
+      expect(job.fallback_args).to eq([lock_key, payload])
+    end
+
+    context 'without an exhaustion handler' do
+      let(:job_class) do
+        Class.new(described_class) do
+          retry_on_lock_conflict wait: 1.second, attempts: 1
+
+          def perform(lock_key)
+            with_lock(lock_key) { raise 'lock should not be acquired' }
+          end
+        end
+      end
+
+      it 'raises the lock acquisition error when retries are exhausted' do
+        allow(lock_manager).to receive(:lock).with(lock_key, Redis::LockManager::LOCK_TIMEOUT).and_return(false)
+
+        expect do
+          job_class.perform_now(lock_key)
+        end.to raise_error(StandardError) { |error| expect(error.class.name).to eq('MutexApplicationJob::LockAcquisitionError') }
+      end
+    end
+  end
 end