AI Automation
AI Integration
AI Email Response Generator is an AI-powered customer support workflow developed for Hollow Technologies to improve how incoming support emails are reviewed, prioritized, and answered. Built with n8n, Gmail, Gemini, and Google Sheets, the workflow automatically processes incoming support messages, classifies their content, identifies their priority, and records relevant information in a spreadsheet. A separate Gemini interaction generates contextual response drafts directly within the original Gmail conversations, allowing support staff to review and send replies themselves. The implementation separates email classification from response generation, uses structured AI outputs, and incorporates request pacing to accommodate API rate limits. Hollow Technologies handled approximately 60–80 support emails daily. The client reported that average response times decreased from approximately 18 hours to minutes after introducing the workflow. This was a paid client engagement completed in April 2026, including workflow delivery, documentation, and handover.
Project details
Context
Hollow Technologies needed a more efficient way to manage incoming customer support emails. The support inbox received approximately 60–80 messages daily, requiring staff to read individual emails, determine their importance, understand the customer's request, and prepare an appropriate response. Handling these activities manually created delays, with the client reporting an average response time of approximately 18 hours before automation. The objective was to reduce the repetitive work involved in reviewing and drafting replies while retaining human control over customer communication. I developed an n8n workflow connecting Gmail, Gemini, and Google Sheets. When an email arrives, the workflow uses Gemini to classify the message and determine relevant information, including its priority and a summary of its contents. The classification results are recorded in Google Sheets, providing a structured view of incoming support requests. A separate Gemini interaction generates a contextual reply draft within the original Gmail conversation. Support staff can review the generated response, make changes where necessary, and decide whether to send it. The workflow was delivered with supporting documentation and handed over to Hollow Technologies in April 2026.
The workflow was designed around tools already familiar to the client. Gmail remained the primary interface for receiving and responding to customers, while Google Sheets provided a structured record of classified messages. Gemini handled the language-processing tasks, including understanding incoming requests and generating contextual response drafts. The automation deliberately separated classification from response generation. This allowed each AI interaction to focus on a distinct responsibility: the first analyzed the message and its priority, while the second prepared a relevant reply. Importantly, the workflow created drafts rather than automatically sending customer-facing messages. Support staff retained responsibility for reviewing and approving outgoing replies.
Contribution
I developed Hollow Technologies' AI-powered support email automation in April 2026, integrating Gmail, Gemini, Google Sheets, and n8n. My responsibilities included workflow architecture, email classification, priority identification, structured information logging, contextual response generation, rate-limit handling, and final documentation and handover.
My contribution focused on automating the repetitive stages of customer support email handling without removing human oversight. The workflow needed to understand incoming requests, organize useful information, and prepare responses that support staff could review directly in Gmail. I approached the implementation by separating the process into distinct stages for message classification, information logging, and response drafting. This provided a clearer division between AI-generated analysis and the customer-facing communication process.
Designed and implemented an n8n workflow connecting incoming Gmail messages with AI classification, spreadsheet logging, and automated response drafting. Coordinated the processing stages so incoming messages could be analyzed and prepared for support staff without requiring manual intervention at every step.
Integrated Gemini to analyze incoming support emails and classify their contents. The classification stage identifies relevant message information and produces structured outputs that the workflow can process.
Implemented AI-assisted priority classification to help distinguish incoming support requests based on their contents. Included the resulting priority information in the structured records generated for the support team.
Integrated Google Sheets to record message summaries, priority information, and related classification results. This provided a separate, organized view of incoming support activity alongside the existing Gmail inbox.
Built a separate Gemini-powered drafting stage that generates replies based on the contents of incoming support messages. The workflow creates contextual response drafts within the original Gmail threads, keeping replies associated with the corresponding customer conversations.
Implemented the response workflow around Gmail drafts rather than automatic message sending. This allowed support staff to review, edit, and approve generated replies before communicating with customers.
Separated classification and drafting into distinct Gemini interactions. Implemented structured-output parsing and request pacing to accommodate API rate limits and support consistent processing of model-generated information.
Delivered the completed workflow and supporting documentation to Hollow Technologies. The handover provided the client with the automation and information needed to operate the workflow within its support process.
The implementation involved two separate AI operations rather than a single prompt attempting to handle every responsibility. The classification stage focused on understanding and organizing incoming messages, while the drafting stage focused on generating an appropriate response. This separation helped maintain clear responsibilities within the automation and allowed classification results to be recorded independently of the generated reply. The completed workflow reduced manual preparation work while preserving human responsibility for outgoing customer communications.
Implementation
The AI Email Response Generator was implemented using n8n to coordinate Gmail, Gemini, and Google Sheets. The automation processes incoming support emails through two distinct AI stages: classification and response generation. The first Gemini interaction analyzes the email and produces structured classification information. This information is processed and recorded in Google Sheets. The second Gemini interaction generates a contextual response, which is saved as a draft in the original Gmail conversation. The implementation also uses structured-output parsing and paced requests to account for API rate limits.
Integrated Gmail with n8n to process incoming customer support messages.
The workflow uses the incoming email content as the starting point for classification and response generation.
Keeping Gmail as the central communication platform meant support staff could continue working within their existing inbox rather than switching to a new application.
The workflow also preserves the connection between the incoming message and its generated response draft.
Implemented a Gemini interaction dedicated to analyzing incoming support emails.
The model interprets the message content and generates structured classification information, including a summary and priority assessment.
Rather than combining classification and response writing in one operation, the classification stage operates independently.
This allows the workflow to process the email's meaning and relevant support information before generating a customer-facing response.
Implemented processing for the structured information generated during email classification.
The workflow parses the model's output so classification results can be mapped to the appropriate downstream fields.
This provides a consistent way to pass message summaries and priority information from the AI stage into Google Sheets.
Using structured outputs also avoids relying on manually interpreted, free-form AI responses for subsequent automation steps.
Connected the classification stage to Google Sheets to organize incoming support information.
The workflow records message summaries, priority information, and related classification details in a structured spreadsheet.
This gives support staff an additional way to review incoming requests and their assigned priorities without manually compiling the information.
Google Sheets serves as the logging and organization layer, while Gmail remains the primary interface for customer correspondence.
Implemented a second Gemini interaction dedicated to generating a response to the incoming customer message.
The drafting stage uses the message context to prepare a relevant reply.
Keeping response generation separate from classification allows the workflow to treat these as different tasks, with distinct outputs and processing responsibilities.
The generated text is used to create a Gmail draft rather than being sent directly to the customer.
Integrated the generated response into the original Gmail conversation.
Instead of creating an unrelated message or automatically sending a reply, the workflow prepares a draft associated with the customer's existing email thread.
This keeps the generated response accessible in the same conversation that support staff would normally use to handle the request.
Staff can inspect the original message, review the generated draft, make corrections, and send the response when appropriate.
The workflow makes separate Gemini requests for classification and response drafting.
Because these operations interact with an external model API, request pacing is important when handling multiple incoming messages.
Implemented paced requests to accommodate model API rate limits during workflow processing.
This helped structure the automation around the constraints of the integrated service rather than assuming every request could be made immediately.
Completed the automation for Hollow Technologies in April 2026.
The delivered workflow connected incoming Gmail messages, structured Gemini classification, Google Sheets logging, and contextual response drafts.
I also prepared documentation and handed over the workflow to the client.
The handover concluded the paid engagement.
Outcomes
The AI Email Response Generator helped Hollow Technologies reduce the time spent preparing responses to incoming support messages. The client's support inbox handled approximately 60–80 messages daily. Following the introduction of the automation, the client reported response times falling from an average of approximately 18 hours to minutes. The workflow also introduced AI-assisted classification, priority identification, structured support logging, and contextual response drafts within existing Gmail conversations. Support staff retained control over final responses, reviewing and sending drafts themselves rather than allowing the automation to communicate with customers without approval. The results reflect the client's reported experience. No independently measured accuracy, customer satisfaction, or cost-saving figures are claimed.
Outcome
Automated incoming email classification and contextual reply drafting, helping Hollow Technologies reduce the time required to respond to customer support requests.
Measured impact
Previously approximately 18 hours on average.
The client reported the improvement for a support inbox handling approximately 60–80 emails daily. These figures represent client feedback, not independently conducted performance measurements.
Deliverable
Implemented Gemini-powered classification of incoming support messages, producing structured information about message content and priority.
The classification stage operates separately from reply generation, allowing support information to be processed and organized independently.
Deliverable
Connected Google Sheets to the workflow to record support message summaries, priorities, and classification results.
The integration provides a structured record of incoming support information alongside the existing Gmail inbox.
Deliverable
Developed a Gemini-powered drafting process that creates contextual response drafts within the original Gmail conversations.
Generated responses remain associated with the incoming customer messages, allowing support staff to inspect and edit them before sending.
Deliverable
Implemented a human-review workflow in which AI prepares responses but support staff retain responsibility for approving and sending customer-facing messages.
The automation creates Gmail drafts rather than automatically sending AI-generated replies.
Deliverable
Designed independent Gemini interactions for email classification and response drafting, with structured-output processing between workflow stages.
The separation gives each AI operation a distinct responsibility and allows classification results to be logged independently of response generation.
Deliverable
Implemented request pacing to accommodate Gemini API rate limits while processing classification and drafting requests.
The implementation accounts for external API constraints without claiming measured throughput or guaranteed execution times.
Outcome
Delivered the AI support email automation and supporting documentation to Hollow Technologies as a completed paid client engagement.
The project was completed and handed over in April 2026. No ongoing maintenance engagement is claimed.