Every pet store that grows online eventually reaches a point where message volume outpaces what a small team can answer within a reasonable time. The common fix is to automate support, but automating everything without criteria creates a different problem: a frustrated customer trying to resolve something specific with a system trained only for generic questions. This piece breaks down where automation genuinely helps and where it costs more than it appears to save.
Hugo Galvao de Franca Filho, founder and director of Enjoy Pets, does not treat this decision as a choice between automation and human support but as a question of where each one handles a specific type of conversation better. Understanding that split avoids both the extreme of automating everything and the extreme of keeping a large team answering repetitive questions no human needs to handle personally.
The type of question automation handles well
Order tracking, delivery timelines, stock availability, and exchange policy are questions that repeat hundreds of times with the same structural answer, changing only the specific order data. This is where automation delivers real value: an immediate response, available outside business hours, with no dependence on a person being available at that exact moment.
For Hugo Galvao, automating this block of questions frees up the human team for what actually requires judgment, instead of keeping them busy repeating information already available on any order dashboard. The metric that matters here is not how many conversations the chatbot resolved; it is how many repetitive conversations stopped consuming the time of someone who could be handling something else.
Where automation creates more problems than it solves
A question about how a pet reacted to a food change, a symptom that may or may not signal a health issue, or a complaint about a product that arrived damaged all require reading context that an automated flow does not reproduce. Trying to fit that kind of conversation into a pre-built decision tree tends to make the experience worse, not better, because the customer feels processed rather than heard.
Hugo Galvao de Franca Filho is direct about this limit: any question involving an animal’s health or a customer’s dissatisfaction goes to human support without exception, even if that means responding more slowly than a bot could. Response speed matters less, in these cases, than the sense that someone actually understood the problem before suggesting a solution.
How to decide what to automate without testing customer by customer?
The safest way to draw this line is to look at the store’s own conversation history and sort questions by volume and by how much the answer varies. High-volume questions with an almost identical answer every time are natural candidates for automation; low-volume questions with an answer that shifts depending on context almost always need a person reading the message before replying.
That mapping, in the practice Hugo Galvao applies at Enjoy Pets, gets revisited periodically because the type of question coming in changes as the catalog grows and new products get added. A question that is rare today can become recurring within a few months, and the automation criteria need to track that shift instead of staying fixed from the first implementation.
The risk of measuring success only by conversation volume closed
Support dashboards tend to highlight how many conversations the bot closed without escalating to a human, a number that looks positive in isolation but hides how many of those conversations ended with the customer giving up rather than getting the problem actually solved. Without cross-referencing that figure with repeat purchase rate or post-support ratings, the metric misleads more than it informs.
Hugo Galvao considers this distinction the point most overlooked by stores that automate without tracking the outcome closely. A practical example of how this balance between automation and human support gets applied day to day can be seen at www.enjoypets.com.br, where the criteria for escalating to a person follow exactly this logic of question type, not just message volume.
