Creator Attribution and Holdout Measurement in Japan

A creator code tells you who remembered to type it. In Japan that is a smaller and more biased slice of the demand than most brands assume: around two thirds of handsets run iOS, Safari caps script-set cookies at seven days and at 24 hours when a visit arrives decorated with parameters from another domain, and much of the demand a creator generates completes on Amazon.co.jp or Rakuten Ichiba rather than on your own store. We keep the codes and the links and we report them, labelled as directional. The number we defend at the quarterly review is a lift figure from staggered flights, with an interval on it. What gets instrumented: Order events posted server-side from your backend to the GA4 Measurement Protocol, the Meta Conversions API and the TikTok Events API, so a purchase survives a browser that dropped its cookie two weeks ago. One code per creator per flight — never one per campaign, and never a code also running in a paid-social ad the same week, because a code appearing in two places measures neither. Amazon Attribution tags where you sell on Amazon.co.jp, the one marketplace in Japan returning creator-level data. A single post-purchase question in Japanese with creators named individually; the option meaning "saw it on social" is removed, because it tells you nothing you can act on. Where the instrumentation stops: Rakuten Ichiba shop pages do not accept your tags. We say so rather than quoting a modelled number in its place. For Rakuten we work from daily store sales in RMS, deliberate stagger in creator timing, and the survey. For products moving through drugstores and variety chains, prefecture-level movement from a retail panel such as Intage SRI+ is the only honest read available, and it arrives weeks late. Getting to a number worth acting on: Creator reach in Japan is not geo-targetable, so the design is time-based rather than geographic. Flights are staggered across the roster and read against a baseline built from your own pre-period, with match-back windows set per category: three days for a game at launch, twenty-one for skincare, sixty or more for equipment that involves a quotation step. Three flights before we put an interval on anything. One flight is one data point, and a line through one point is decoration. What we will not report: Last-click creator return on ad spend as a headline. A creator credited for a session that arrived through a branded search. A redemption counted twice because it also touched a paid-social ad. Attribution summing to one hundred percent — the monthly reconciliation carries an explicit unattributable line, and in the first quarter it is usually the largest figure on the page. What is not included: No media mix model. Below roughly one hundred million yen of annual measured media there is not enough variance in your spend history for one to say anything an experiment cannot say sooner and for less. No customer data platform or warehouse build. No affiliate network management: if you want A8.net or Rakuten Advertising operated for you, that is a different practice and we will name one. Who this is not for: Teams needing one number every Monday morning. Finance functions requiring every yen of revenue assigned to a source. Brands whose creator spend is small enough that the measurement costs more than the media — under about fifteen million yen a year, run the campaigns, keep clean records, and come back when the budget is worth defending.