Designing a support workflow for Grupop: macros, triggers, and an AI I didn’t plan for

I wanted to see how Zendesk could handle a multifaceted support query that is already addressed within Grupop Help. Many of the FAQ were pretty cut and dried, but one that enabled a more in-depth investigation of Zendesk’s capabilities related to cancellations.

Cancellation questions are one of the most common - and most sensitive - things a teacher might email support about. Get the response wrong, or too slow, and you risk losing a subscriber unnecessarily; get it too automated and impersonal, and you undermine the trust Grupop is built on.

I wanted to design a proper support workflow around a cancellation request in an industry-standard tool, as a genuine test of matching a tool’s real capabilities to a real product problem, rather than just following a tutorial.

When I initially signed up, I was prompted for information about Grupop so I created 3 articles that contained information already available on grupop.ie. I was also prompted to personalise an AI agent to test and demonstrate using information from those articles, which I named Pop! Pop was able to answer questions relating to the context of the articles in a popup chat.

Once those aspects of the registration process were complete, I set about designing a cancellation request workflow. This required creating the following:

1. A Help Centre article on cancellation

An article dealing specifically with cancellation, which I would use as a reference when responding to cancellation queries. This covered the difference between cancelling a trial, a monthly plan, and an annual plan. In a real deployment, this would be the self-service layer most people asking this question wouldn’t need to go beyond.

2. A “Cancellation Intent” ticket field

A simple dropdown (Info Request / Cancel Now) that lets an agent classify what a customer is actually asking for, once they’ve read the ticket - rather than guessing from a subject line.

3. Two macros and one trigger

• A trigger scans new tickets for vocabulary related to cancellations, automatically tags them and raises priority - on the reasoning that any mention of cancellation, even a vague one, deserves a fast look, whether or not the customer ends up actually leaving.
• Two macros - “Cancellation - Info Request” and “Cancellation - Cancel Now” - give the agent a one-click, on-brand reply for each real scenario, linking back to the Help Centre article.

While testing this with a dummy ticket in the trial, I was pleased to realise that Zendesk would automatically present articles within the ticket creation UI depending on the content of the ticket subject field. Some of the original articles I had imported during the registration process, along with the article I had written specifically for this task, were those recommended.

Upon submitting the ticket, Zendesk correctly flagged the ticket as containing the word “cancel” and instantly assigned High priority status to the ticket. Enthused by this success, I raced to see how the macros were applied but realised that Pop (the AI agent I had been prompted to create during registration) had already read the ticket, pulled directly from the Help Centre articles I’d written, and sent a fully-formed reply - before either of my macros ever got a chance to fire!

My first instinct was that this broke my careful macro workflow. On quick reflection, however, I decided it did the opposite: it validated it. The AI’s reply was only as good as the article it pulled from - so a correct, well-structured AI response is really a signal that the underlying content was written clearly enough for a machine to use it correctly, not just a human. And although this response lacked specifics in the second half of the email, they would have been specifics that would have been provided had I uploaded all of Grupop’s policy documents to the Zendesk Knowledge Base, which would certainly have been the case in a production environment. The information that had been provided to the AI agent via the articles was utilised impressively well to answer a quite specific question.

Rather than treat the AI layer as making the macros redundant, I’ve kept both in this design, deliberately:

• The AI Agent would handle the majority of first-contact cancellation queries automatically, using the Help Centre articles as its source of truth.
• The macros remain as a fallback layer - for tickets where AI confidence is low, where a customer replies again after the automated response (a strong signal they need a human), or as a safety net if AI Agents were ever switched off.
• The trigger’s tagging/priority logic applies regardless of which path resolves the ticket, so nothing “cancellation-shaped” goes unnoticed even when the AI handles it silently.

Ironically this mirrors a decision I made deliberately only hours previously, rather than by default: I initially explored a more ambitious version of this - a fully automated multi-branch workflow asking customers to self-select their plan type and intent before ever reaching a human. I stepped back from that once I recognised it was pushing Zendesk toward being a decision-tree/chatbot builder, which isn’t what the tool is actually for. Scoping back down to “self-service article → light triage → human-reviewed macro reply, with AI as an emergent bonus layer” is a better match for what Zendesk does well.

This exercise wasn’t about needing Zendesk today - Grupop’s support volume doesn’t require it yet. It was about proving, on a real piece of Grupop’s own content and logic, that I understand how to design a support system that scales past founder-led email: where self-service, light automation, and human judgement each do the job they’re actually suited to, rather than over-engineering any one layer.

• Extend the same Help Centre → trigger → macro pattern to the other two or three most common FAQ topics, rather than just cancellations.
• If Grupop’s support volume ever genuinely grew, revisit whether a proper conversational AI flow (rather than macros) becomes worth the added complexity - the groundwork here would carry over directly, since the Help Centre content is the shared foundation either way.
• Experiment with AI agents suggesting responses for the human agent rather than automatically responding directly to the customer. Beginning with the most basic FAQ topics would be the best place to start.
• Should Grupop move forward with AI responding to customer tickets, I would add a short satisfaction check after AI-resolved tickets, to catch cases where the AI’s answer looked right but didn’t actually solve the customer’s problem.
• Consider renaming the AI agent from Pop to something a little more professional - if a customer is enquiring about cancellations, receiving an automated response signed “Pop” may be perceived as insincere!