The starting point

SnoPro delivers ski and snowboard hire to where people are staying, rather than making them queue in a shop. It runs two businesses in one, in opposite hemispheres: Queenstown and Wanaka from June to October, and Niseko in Japan from December to March. When one season ends the other is already selling.

We came in on the New Zealand side, on a mature account that had been running for years. Then in late 2025 they decided to launch Japan: no brand recognition, no campaign history, and a booking funnel that had never been tested outside a domestic market.

Japan: a funnel with no history behind it

The instinct with a new market is to spend and see what happens. We did the opposite and started with measurement. Before meaningful budget went live, we wrote the event specification for the Japanese booking funnel from scratch: what counts as intent, what counts as a started booking, what carries value and in which currency, and which of those are worth optimising toward. Search campaigns were built around how Japanese skiers actually search, rather than translated across from the New Zealand account.

The first season ran from a standing start and filled. The shape of it is worth more than the size: bookings arrived well ahead of the snow, with the bulk of the season's volume landing in the pre-season months for trips taken after New Year. Plan the budget against when people book, not when they ski, and the account stops fighting its own calendar.

Those bookings are every booking the site took, from every source. We are not claiming the ads produced all of them, and the last section explains why we could not claim it even if we wanted to.

New Zealand: holding ground while search changes underneath it

The New Zealand side is the older account and a different kind of work. Audience roughly doubled between 2023 and 2025.

2026 is the year that got interesting. Sessions year to date are down 16% on the same window last year. Underneath that number:

Channel	Change, 2026 vs 2025
Organic Search	−30%
Direct	+26%
AI Assistant	new channel
Paid Search	down, by design

Organic search is down 30%, which is the AI search shift arriving on schedule. Direct is up 26% across the same window, and that is the other half of the same story: when an assistant answers the question and names the business, the visit lands as direct rather than as a click on a result. GA4 now reports AI Assistant as a channel in its own right, and the pages it sends people to are the location and comparison content, not the home page.

That is the whole reason total sessions fell 16% while organic fell 30%. The content built to rank is being read and cited by the thing that is replacing ranking, so the demand arrives through a different door instead of not arriving. Sites without that content depth are taking the organic hit without the direct offset.

Paid search is down separately and on purpose. Targeting was rebuilt around tighter regional ad groups, and bidding moved onto booking value rather than click volume. Both trade clicks for bookings by design.

What these numbers can and cannot tell you

Both booking systems send their completion events server-side. That is the right call for reliability, and it has one consequence worth being upfront about: those events arrive with no session context, so they cannot be traced back to the channel that produced them. On the Japan property, effectively none of the completed bookings can be attributed to a source at all.

So there is no honest return-on-ad-spend figure for either market, and we do not publish one. What the ad platform reports as conversion value in these accounts is dominated by checkout starts, which is a perfectly good optimisation signal and a bad revenue number. Treating the first as the second is the most common way paid search reporting flatters itself, and it is worth checking whoever runs your account is not doing it.

What we do instead: optimise against the funnel-stage events that do carry session context, hold those to a floor, and read the real result at the business level against total spend. It is less tidy than a single multiplier. It has the advantage of being true.

What made it work

Launching in a new market is a measurement problem before it is a budget problem. Every dollar in the opening weeks buys information as much as clicks, and information you cannot read is just an expense.

The New Zealand year makes the same point from the other direction. An account read only on last year's metrics looks like it is going backwards. Read on the right ones, it is a brand being found through a channel that did not exist in the previous reporting period, off the back of content written years before anyone needed it to work this way.