Inside the rollout
A household New Zealand food brand lives or dies by sentiment. Across Instagram, Facebook, TikTok comments, Google reviews and the contact form, thousands of mentions land every week, and they spike hard whenever a new block drops or a campaign goes live. The community team was proud of replying to everyone, but they were drowning. Genuine complaints, a cracked product, a wrong promo code, an allergy question, sat in the same undifferentiated flood as fan love and meme replies, and the ones that actually needed a fast, careful human answer were the easiest to miss.
Our approach
We built an agent that watches every channel and does three things on each mention: classifies it (praise, question, complaint, PR risk, spam), scores how urgent it is, and drafts a reply in the brand's exact voice with the right tone for the situation. Anything sensitive, allergies, safety, media, gets flagged and routed straight to a named person with the full context attached. Everything routine arrives in a queue with a reply already written, so the team approves or tweaks instead of starting from a blank box. The voice was tuned on years of the brand's own published replies, so it sounds like them, not like a bot.
Results at a glance
We used to measure ourselves on how late we stayed up clearing the queue. Now every mention is sorted and has a draft reply attached within minutes of landing, the urgent ones are on the right person's screen before they've even refreshed, and nothing about a safety or allergy comment can quietly slip through. The team went from firefighting to actually choosing what deserves a real, human answer. During our last campaign week it handled the spike without us adding a single temp.
