diff --git a/.rubocop.yml b/.rubocop.yml
index d8c9e2b..0c4ee6b 100644
--- a/.rubocop.yml
+++ b/.rubocop.yml
@@ -36,3 +36,6 @@ Style/StringLiterals:
 
 Style/StringLiteralsInInterpolation:
   EnforcedStyle: single_quotes
+
+RSpec/ExampleLength:
+  Max: 20
diff --git a/lib/esse/import/bulk.rb b/lib/esse/import/bulk.rb
index 00a6b7b..67a020a 100644
--- a/lib/esse/import/bulk.rb
+++ b/lib/esse/import/bulk.rb
@@ -22,12 +22,12 @@ def initialize(type: nil, index: nil, delete: nil, create: nil)
       # Return an array of RequestBody instances
       #
       # In case of timeout error, will retry with an exponential backoff using the following formula:
-      #  wait_interval = (retry_count**4) + 15 + (rand(10) * (retry_count + 1)) seconds. It will retry up to max_retries times that is default 3.
+      #  wait_interval = (retry_count**4) + 15 + (rand(10) * (retry_count + 1)) seconds. It will retry up to max_retries times that is default 4.
       #
       # Too large bulk requests will be split into multiple requests with only one attempt.
       #
       # @yield [RequestBody] A request body instance
-      def each_request(max_retries: 3)
+      def each_request(max_retries: 4, last_retry_in_small_chunks: true)
         # @TODO create indexes when by checking all the index suffixes (if mapping is not empty)
         requests = [optimistic_request]
         retry_count = 0
@@ -43,6 +43,8 @@ def each_request(max_retries: 3)
         rescue Faraday::TimeoutError, Esse::Transport::RequestTimeoutError => e
           retry_count += 1
           raise Esse::Transport::RequestTimeoutError.new(e.message) if retry_count >= max_retries
+          # Timeout error may be caused by a too large request, so we split the requests in small chunks as a last attempt
+          requests = requests_in_small_chunks if last_retry_in_small_chunks && max_retries > 2 && retry_count == max_retries - 2
           wait_interval = (retry_count**4) + 15 + (rand(10) * (retry_count + 1))
           Esse.logger.warn "Timeout error, retrying in #{wait_interval} seconds"
           sleep(wait_interval)
@@ -73,6 +75,18 @@ def optimistic_request
         request
       end
 
+      def requests_in_small_chunks(chunk_size: 1)
+        arr = []
+        @delete.each_slice(chunk_size) { |slice| arr << Import::RequestBodyAsJson.new.tap { |r| r.delete = slice } }
+        @create.each_slice(chunk_size) { |slice| arr << Import::RequestBodyAsJson.new.tap { |r| r.create = slice } }
+        @index.each_slice(chunk_size) { |slice| arr << Import::RequestBodyAsJson.new.tap { |r| r.index = slice } }
+        Esse.logger.warn <<~MSG
+          Retrying the last request in small chunks of #{chunk_size} documents.
+          This is a last resort to avoid timeout errors, consider increasing the bulk size or reducing the batch size.
+        MSG
+        arr
+      end
+
       # @return [Array<RequestBody>]
       def balance_requests_size(err)
         if (bulk_size = err.message.scan(/exceeded.(\d+).bytes/).dig(0, 0).to_i) > 0
diff --git a/spec/esse/import/bulk_spec.rb b/spec/esse/import/bulk_spec.rb
index 6dbefbf..2a1d8be 100644
--- a/spec/esse/import/bulk_spec.rb
+++ b/spec/esse/import/bulk_spec.rb
@@ -7,7 +7,7 @@
     let(:index) { [Esse::HashDocument.new(id: 1, source: { foo: 'bar' })] }
     let(:create) { [Esse::HashDocument.new(id: 2, source: { foo: 'bar' })] }
     let(:delete) { [Esse::HashDocument.new(id: 3, source: { foo: 'bar' })] }
-    let(:bulk) { Esse::Import::Bulk.new(index: index, create: create, delete: delete) }
+    let(:bulk) { described_class.new(index: index, create: create, delete: delete) }
 
     it 'yields a request body instance' do
       expect { |b| bulk.each_request(&b) }.to yield_with_args(Esse::Import::RequestBodyAsJson)
@@ -18,7 +18,7 @@
       expect(Esse.logger).to receive(:warn).with(an_instance_of(String)).twice
       retries = 0
       expect {
-        bulk.each_request(max_retries: 3) { |request|
+        bulk.each_request(max_retries: 3, last_retry_in_small_chunks: false) { |request|
           retries += 1
           raise Faraday::TimeoutError
         }
@@ -31,7 +31,7 @@
       expect(Esse.logger).to receive(:warn).with(an_instance_of(String)).twice
       retries = 0
       expect {
-        bulk.each_request(max_retries: 3) { |request|
+        bulk.each_request(max_retries: 3, last_retry_in_small_chunks: false) { |request|
           retries += 1
           raise Esse::Transport::RequestTimeoutError
         }
@@ -49,6 +49,29 @@
       end
     end
 
+    context 'when on last retry and last_retry_in_small_chunks is true' do
+      let(:index) do
+        %w[foo bar baz].each_with_index.map { |name, idx| Esse::HashDocument.new(id: idx + 10, source: { name: name }) }
+      end
+      let(:create) do
+        %w[foo bar baz].each_with_index.map { |name, idx| Esse::HashDocument.new(id: idx + 20, source: { name: name }) }
+      end
+      let(:delete) do
+        %w[foo bar baz].each_with_index.map { |name, idx| Esse::HashDocument.new(id: idx + 30, source: { name: name }) }
+      end
+      let(:bulk) { described_class.new(index: index, create: create, delete: delete) }
+
+      it 'retries in small chunks' do
+        expect(bulk).to receive(:sleep).with(an_instance_of(Integer)).exactly(3).times
+        requests = []
+        bulk.each_request(last_retry_in_small_chunks: true) { |request|
+          requests << request
+          raise Faraday::TimeoutError if [1, 2, 3].include?(requests.size)
+        }
+        expect(requests.size).to eq(3 + index.size + create.size + delete.size)
+      end
+    end
+
     context 'with a request entity too large error' do
       let(:index) { [Esse::HashDocument.new(id: 1, source: { name: 'Aaa' * 30 })] }
       let(:create) { [Esse::HashDocument.new(id: 2, source: { name: 'Bbbb' * 100 })] }