The old process ran on paper, repeated briefings and volunteer stamina. The principal, school staff and keyholder volunteers spent evening after evening explaining the same rules, answering the same questions and issuing keys. Building the roster meant contacting 30 or 40 people individually and negotiating availability until the gaps disappeared. At season end, school administration still had to chase keys back.
Last year, under pressure, I started redesigning that process with ChatGPT using free Google tools. The platform choice was deliberate from the beginning: Golden Bay High School uses Google Education, and my long-term goal has always been to hand the system back to the school rather than create a private platform that depends on me.
The first digital version removed a lot of repetition and proved the model. This year we stopped overhauling and started engineering the edges.
The hard part is the humans
A clean database assumes clean behaviour. A community system gets edited registrations, partners using different email addresses, payments made under another name, family and individual memberships, school-paid exceptions, short-term keys, volunteer preferences, training needs and useful notes arriving outside the formal workflow.
We treated that mess as the design problem rather than an inconvenience.
Registration data flows through Google Forms and Sheets. Payment and agreement status are reconciled in one authoritative admin workbook. Personalised mail merges draw on that data plus Gmail history and operational notes, including information first captured through WhatsApp. ChatGPT helps hold the system model together: matching records, identifying anomalies, shaping roster decisions and carrying the right context into the right communication.
We also built a tablet-friendly AppSheet collection desk so a volunteer could find a keyholder, check readiness and record payment or collection without working inside the full administration sheet. The prototype worked. Deploying it for a short collection period would have required moving beyond the free tier, so we deliberately stopped there. A technically successful build did not automatically justify deployment.
We weren’t trying to automate volunteering. We were trying to stop administration consuming the volunteers.
Refinement means finding the failures
This year I tried to strengthen the family process with a digital second-adult agreement. It exposed a new failure mode: mismatched emails, edited responses and people taking the wrong pathway through the form.
That was useful evidence. We fixed the live cases, simplified the logic and have already designed a cleaner approach for the next iteration. The system is now mature enough that the work is no longer “digitise the process”. It is observe, test, correct and reduce friction without losing the human judgement the process actually needs.
Designed for an owner who is not me
The handover is still a goal, not a completed milestone. But it has shaped the architecture from day one: free and familiar tools, one authoritative dataset, explicit exception handling, documented workflows and interfaces that do not require the next volunteer to understand every formula underneath them.
Keyholders are already noticing that the process feels more organised. That matters. But the harder test comes later.
The real measure of the system will not be whether I can run it efficiently. It will be whether someone else can.