Embedded Files
Overview
Mastering Diabetes was receiving a tremendous volume of comments across social media channels, with their service team spending several hours daily responding to each comment. Many of these were low-risk engagement comments—gratitude, thank-yous, requests for guides and blueprints—that didn't require specialist judgment but were consuming valuable human resources. The team needed a way to automate responses to these routine comments while preserving human oversight for health-related inquiries that required expertise and care.
Solution
I developed an AI-powered comment classification and response system using a multi-layered approach:
Comment Collection: Make.com automated the collection of incoming comments from social media and stored them in Google Sheets for tracking
Flagging & Triggering: A custom Google Apps Script flagged CTAs (calls-to-action) as they were collected, triggering automated workflows in n8n when new comments arrived
Intelligent Classification: The system used chain prompting to classify comments into two categories: simple gratitude/CTA requests (automated response) vs. health-related inquiries (human escalation)
FAQ Search & Retrieval: For relevant queries, the system searched a vectorised FAQ database in Pinecone to find the most appropriate information
Brand Tone Refinement: A final AI layer ensured all responses matched Mastering Diabetes's brand voice and quality standards before delivery
Challenges
The primary challenge emerged during testing, as the AI was generating responses that were too enthusiastic and didn't align with the brand's careful, health-conscious tone. This was critical because health education requires trust and consistency.
I addressed this by:
Reducing the temperature parameter to make responses more deterministic and measured
Refining prompts to enforce strict adherence to brand guidelines
Adding a quality gate before any response left the system
Results
The system successfully eliminated the need for manual CTA processing, freeing the service team from routine comment responses so they could focus entirely on health-related inquiries. In the first 10 days alone, the system processed 2,000 CTAs and collecting email addresses for the marketing list. It improved both operational efficiency and business impact. The team regained valuable time while maintaining the careful, trusted tone essential for health education.
Page updated
Google Sites
Report abuse