Guest ops — Case Study
backThe question
A hundred-odd people, two messaging channels, a venue that had to be paid for up front, and a kitchen somebody had to fill. Run in a group chat, this is a week of answering the same four questions. The brief I set myself: know, at any moment and without asking anyone, who is coming, who has paid, and what is still missing from the table.
What I built
A guest pipeline in a spreadsheet — one row per person, one column per state they can be in — replacing the thread everyone scrolls. It is a small CRM wearing a party hat, and the design problems are the ones any outreach pipeline has: two channels with different delivery evidence, a funnel that only moves in one direction, and a money column that has to be reconciled against something real.
- One row per guest, with the channel the invitation went out on
- Delivery state kept separate from reply state — sent, delivered, read
- A reply column and a decision column, because "answered" and "coming" are not the same event
- A contribution column for the shared cost, and a column for what each person is bringing
- A free-text note per row for the reason — which is the only part a number cannot hold
Technical decisions
Delivered is not read, and read is not an answer — three separate states, because collapsing them is how you end up chasing someone who never got the message and ignoring someone who did. The channel with no read receipts gets its own state, rather than being guessed at.
One row per person, never per message — a guest invited on two channels is still one guest. The duplicate is a note on the row, not a second row, so the headcount can be trusted.
Decline with a reason — a no that records "away that weekend" or "working" is worth keeping; a blank cell is not. The reasons are what told me the real constraint was the date, not the plan.
Money is tracked, never assumed — the contribution column holds what actually arrived, so the budget is reconciled against transfers instead of intentions.
What I learned
The useful output was not the headcount, it was the funnel: how many invitations converted at each step, which channel answered, and which day of the week a reply arrives on. The same pipeline, relabelled, is the one I use for job applications — and the one any small team uses for leads.
The other lesson was about people: a tracker with real names, numbers and reasons in it is a privacy problem the moment it leaves your hands. This write-up describes the system and none of the guests, and the file itself never went anywhere it did not need to go.