A prospecting system that works properly does nothing for most of the day. That reads like a defect and it is the whole design: what the machine buys you is not throughput, it is a clock. It can arrive in someone's inbox at nine in their morning, on a day when you are asleep, without a person having set an alarm at three. Below is the actual twenty-four hours of one we run — the sends, the pauses, the roughly forty minutes that cannot be given to software, and the points where the sequence has to stop and wait for a human being.
Auckland runs on UTC+12, and UTC+13 between late September and early April. Nouméa sits on UTC+11 and never moves. Port Vila is on UTC+11, Suva on UTC+12 — Fiji has not run daylight saving since the 2021-22 season — and Papeete is on UTC-10. Put those on one clock and the arithmetic turns blunt. A message that should land in a Papeete inbox at eight on Monday morning has to leave Auckland at six on Tuesday morning. Twenty-one hours separate the two towns, and they sit in the same ocean.
Between Kanaky (New Caledonia) and New Zealand the gap is the opposite problem: one hour in the southern winter, two in summer. Small gaps are the ones people get wrong, precisely because they feel ignorable. Send at nine your time in February and you have arrived at seven in theirs, in the tranche a person clears on a phone before getting to a desk — the one that gets deleted rather than read. Nothing about that is a copywriting failure. It is a scheduling failure, and scheduling is the one thing a machine does perfectly.
| Time | What happens | Human minutes |
|---|---|---|
| 05:40 | Verification pass on the day's forty addresses. Bounced, role-based (info@, contact@) and already-in-the-CRM entries are dropped before a single message is composed. | 0 |
| 06:00 | Papeete window — eight in the morning there, the previous calendar day. Six messages. | 0 |
| 08:30 | Overnight replies land in one place, sorted into three piles: answer today, answer this week, remove from the list. | 20 |
| 09:00 | Nouméa and Port Vila window — eight local. Fifteen messages, twelve to twenty minutes apart. | 0 |
| 10:00 | Auckland and Suva window. The rest of the day's allowance. | 0 |
| 11:30 | Second touches — only to people who never replied, never to people who did. | 0 |
| 14:00 | Anything that asks a question, mentions a price or asks who we are has stopped its own sequence and is waiting for a person. | 15 |
| 16:00 | Nothing sends. Tomorrow's list is enriched: company size, a named person, one specific reason to write to them. | 5 |
| 18:00 | Counters — bounces, complaints, replies. A bounce rate above two per cent holds tomorrow's send automatically, without asking. | 0 |
| 20:00 – 05:40 | Idle. Deliberately. | 0 |
Ninety minutes of machine activity spread across eighteen hours, and forty minutes of human work. The spread is the product: compressing the same forty messages into one afternoon would be quicker to run and worth measurably less, because two thirds would land outside the hours in which anyone answers anything.
Apple shipped Mail Privacy Protection with iOS 15 in September 2021. When it is on, Mail pre-loads remote content through a proxy, which means an open is recorded whether or not a human ever looked at the message. Enough recipients use Apple Mail that the metric can no longer carry a decision — and a system that branches on opens is not automating judgement, it is automating a coin toss and reporting the result as insight.
So the only triggers we let a sequence act on are the four that correspond to something a person actually did: a reply, a bounce, an unsubscribe, and time elapsed. That removes most of the clever branching sold with these tools, along with the class of mistake where a machine sends a warmer second message because a proxy in another hemisphere fetched an image.
Three tasks resist automation, and they are the three that decide the outcome. Choosing who goes on the list — which is judgement about a market, not a filter on a database. The first sentence of the message, which has to contain something only someone who looked would know. And the reply, which is a conversation.
The arithmetic underneath is unglamorous. Forty messages a day, five days, is two hundred a week. We plan on one reply in twenty; that is a planning figure we revise downward for a cold list, not a promise. Ten conversations a week, then, arriving in a mailbox that someone has to actually work. Which points at the real failure mode of a system like this, and it is never too few replies. It is an owner who quietly stopped reading them in week three because the machine kept producing them at a rate the calendar could not absorb.
Lists rot silently — people change jobs, companies close — and a system with no verification step keeps writing into the gaps while the dashboard stays green. The bounce rate moves first, which is why the 18:00 counter holds the next day's send instead of emailing someone about it.
The other failure is regional and appears in no vendor documentation. A schedule that runs itself will send during a cyclone warning, on a day of mourning, or into a town where nobody is at work for reasons every local person knows and no calendar API contains. In this part of the world that happens several times a year. The requirement is a stop control a person can reach from a phone in under a minute, and the willingness to use it early. A machine writing cheerfully through a week like that burns more goodwill than a quarter of outreach earns back.
Ours is doing something for about ninety minutes, spread across roughly eighteen hours, and idle the rest. The spread is deliberate: messages have to arrive during working hours in four different time zones, which on an Auckland clock means separate windows at six, nine and ten in the morning. Compressing the same volume into one block would be faster to run and worth considerably less, because most of it would land when nobody is at a desk.
No. Apple Mail Privacy Protection, shipped with iOS 15 in September 2021, pre-loads remote content through a proxy, so an open is logged whether or not a person read anything. The proportion of recipients affected is large enough that the metric cannot support a decision. Trigger only on events that correspond to a real action: a reply, a bounce, an unsubscribe, or time elapsed.
In both places, within conditions. New Zealand's Unsolicited Electronic Messages Act 2007 recognises express, inferred and deemed consent — the last covering a business address conspicuously published without a statement refusing such messages, where the content is relevant to that person's role. The Department of Internal Affairs enforces it, with penalties reaching NZ$200,000 for an individual and NZ$500,000 for an organisation. In New Caledonia, under the regime supervised by the CNIL, business-to-business prospecting does not require prior consent provided the message concerns the person's professional function and offers a working way to object in every message. In both jurisdictions, accurate sender identification and a functioning unsubscribe are not optional.
Fewer than the tools suggest. Forty a day across a warmed domain is a defensible working figure for a small firm, and the binding constraint is not the sending cap but the complaint budget: Google asks bulk senders to hold spam rates under 0.3 per cent, which at that volume is about one complaint a week. If the replies are not coming, volume is the wrong variable to change — the list is.
Kanaky Tech is an AI automation agency working across New Zealand and the Pacific. Start with a free AI opportunity audit: we map how your business actually runs, rank what is worth automating, and give you a clear scope before anything is built — no obligation.