the reconciliation
press enter
They are both true, from the same data, at the same time — because blended CVR and CAC are not measuring the same customers. This page reconciles the three numbers the team is quoting, with the math shown.
Cost per session is flat vs the $90 era ($1.61 → $1.62). The entire CAC move is the red term: new customers per session fell 1.78% → 1.39% (-22%). Blended CVR — all purchases ÷ sessions — contains a second population that CAC doesn’t care about: returning customers. They keep buying regardless of this month’s ads, so they prop the blended number up exactly when new-customer acquisition sags.
Black: blended CVR (every purchase ÷ sessions) — noisy, range-bound, no trend. Rose: new customers per session — the number CAC is built on. Bars: returning-customer purchases per day, the cushion that hides the rose line inside the black one.
Exactly the segmentation you’d want: filter to new visitors only and the decline is unmistakable — new-visitor CVR peaked at 2.37% (late Apr–early May) and is 1.62% now (-32%), while returning-visitor CVR is trendless noise. Blended mixes the two, which is why it looks “stable”. (GA4 “new visitor” ≈ but ≠ new customer — the Shopify NC-per-session line above is the cleaner business measure; both fall the same way.) And to connect this to the funnel: this rate decline IS the LP→PDP break — fewer new visitors reaching product pages is the same problem viewed from the people side. The stage-level detail is on the radar and the full breakdown.
Blended CVR hit its quarter HIGH (2.57%) in the week of May 25 — while new customers were a mediocre 495/day. The gap was 259/day of returning-customer purchases (Strawberry Matcha LTO buyers, subscribers adding on). A blended-CVR chart literally cannot see the difference between a great acquisition week and a good promo to the existing base.
Left: new customers per day, peak → now, split into the traffic effect and the rate effect. Right: CAC, peak → now — the cost-per-session step is barely visible; it’s a conversion-rate problem.
Method: traffic effect holds the peak NC-rate constant and applies the session decline (33,401 → 27,198/day); rate effect applies the NC-rate decline (1.78% → 1.39%) to the remainder. CAC bridge: at today’s cost per session with the PEAK conversion rate, CAC would be ~$91 — i.e. nearly all of the $90→$117 move is the rate term. Where the rate damage sits in the funnel (traffic not reaching product pages, rise-2 diversion) is mapped on the drop-off radar and the diverging steps.
Compared to March, today’s new-customer rate is basically unchanged (1.4% → 1.39%) and March CAC already averaged ~$101 — today’s pain vs March is mostly traffic volume (35,466 → 27,198 sessions/day) at a higher cost per session ($1.42 → $1.62). Compared to the Apr 27 – May 10 peak — the only two $90 weeks of the quarter — it’s the conversion rate. Both framings are correct; people anchored to different windows will talk past each other. The peak fortnight is also when the test program was quietest, which is the argument that 1.78% is attainable, not a fluke.
This is the dial that sets CAC. It recovers if the rise-2 routing fix works (URL tests paused Jun 11). Track it weekly, not blended CVR — blended cannot see this problem.
The media-side dial: auction CPMs, creative, Google Shopping reinstatement, channel mix. Flat vs the peak era, +14% vs March.