Returning = the customer had an earlier order inside the current filter. Wolf & Badger customers are matched on name only (no email), so their repeat rate is a floor, not a point estimate.
Each line is one acquisition-year cohort. Height at month 0 is the first-order value; the slope after that is repeat revenue. Steeper is better; a flat line means the cohort never came back.
Blank cells: the cohort is too young for that horizon. Repeat rate = share of the cohort with a second order within N months of the first. LTV = kept revenue per acquired customer (returns removed).
Of customers acquired at least N days ago, the share that placed a second order within N days. Read it as: how much of the eventual repeat happens early.
September is the peak month across years; June is the trough. Campaign calendars below are anchored to this.
"Reachable" = customers with an email on file. The segment CSV delivered alongside this dashboard contains the actual lists.
Source: 16 monthly Farfetch partner statements, April 2025 to July 2026. Values are the purchase price Farfetch pays Moos, excluding VAT, converted from RON at 0.20. There is no customer identity in these files, so lifetime value and repeat rate cannot be measured here; this is a unit-economics and assortment view of the channel.
OS = one size (gloves and accessories). Sized garments return at roughly the same rate regardless of size here, unlike the own site where 40–42 stand out: on Farfetch the problem is the buyer, not the fit chart.
Own-store revenue is what the customer paid (incl. VAT, before payment fees); Farfetch payout is what Moos receives ex VAT after Farfetch's margin. They are not the same measure, which is exactly why the per-unit line matters: it is the cash Moos books per garment that stays sold.
Own store = Website orders shipped to the USA in the order book. Wolf & Badger rows labelled USA are excluded from the store view because their country field is unreliable (it contains London, Dublin and Paris). Farfetch US comes from the partner statements. Context from the 17 September sync is folded into the opportunities below.
Two-arm test, 80% power, 95% confidence, equal split. The instrumentation deployed in early September counted 119 US visitors in seven days with zero conversions, so the defaults are deliberately generous. The conclusion does not change much with the inputs: US-only conversion tests take quarters, not weeks. Test site-wide and read the US as a segment, or use pre/post with other markets as the control.
Two tracks. Deploy = ship to the store now, read the result pre/post against the other markets; these are fixes or changes where a control group adds nothing. Test = run as a site-wide A/B and read the US segment, or as a staged rollout; these are changes with a real downside if wrong. Each card names the evidence, the metric and what "working" looks like.