Anatomy of a funnel contraction

Four sources, reconciled quarter by quarter across 2021–2025: Shopify order data in BigQuery, channel-level ad spend from planning records and Triple Whale, storefront session analytics, and daily message-level Klaviyo data from February 2023 onward. Reconciling sources like these is exactly the work I build code-first warehouses for.

The indexing scheme: annual revenue is indexed to the peak year = 100, CAC is indexed to the efficient 2021 baseline = 100, and spend is shown as a share of that year's revenue. ROAS multiples, conversion rates, percentage changes, and per-send economics are unchanged, because the lesson lives in the ratios, not the absolutes.

2021 was the efficient baseline: 8.0x ROAS with ad spend at 12.6% of revenue. In 2022 the brand more than doubled spend (+114%) and bought its all-time revenue peak, up 41%. The price was the cost structure. CAC rose 74% above baseline, cost per session rose 73%, and ROAS fell from 8.0x to 5.3x. 2023 ran the same machine more efficiently and held revenue within 1% of peak, but CAC never returned to baseline.

Every time quarterly spend was pushed toward the top of its range, the same thing happened: CAC blew past twice the baseline and ROAS fell under 5x. Q3 2022 hit the wall first, at a CAC index of 217 and 4.4x ROAS. Q2 2023 came within 10% of it. Q1 2024 hit 217 again. Three touches in seven quarters is not bad luck. It is a structural ceiling on how many customers the market would sell at a profitable price, and the margin structure could not absorb acquisition above it.

After the third touch, the pullback was not optional. Margin requirements forced spend down 56% from 2023 to 2025, and every downstream stage contracted in sequence. The key exoneration sits in the middle: the email signup rate improved from 2.3% to 4.0% while list growth fell. Fewer people visited, so fewer people signed up. The funnel converted better at every stage it controlled. It was starved, not broken.

The quarterly campaign data makes the dose-response visible. The lowest-volume quarter in the dataset earned the highest revenue per thousand sends. A quarter with five times that volume earned the worst. The mechanism is deliverability: pushing volume into a list degrades inbox placement, which degrades opens, which degrades revenue per send. There is an optimal send band, and this account overshot it for years. My industry panel report found the same overshoot pattern across 226 brands, from the outside.

Inside the flow-send collapse was a correction, not a loss. One flow had been misconfigured to trigger for every site visitor already on the list, and it accounted for more than half of all flow volume in 2024 at $0.04 of revenue per send. The flows that actually convert earned $1.06 per send (abandoned cart) and $4.83 per send (checkout abandon). Every unopened junk send was dragging sender reputation down underneath them. Fixing it removed a huge share of the volume and protected the flows that made money, which is what a send-volume chart will never tell you on its own.

By 2025 every efficiency metric was at its best level since the baseline year: CAC down 33% from the peak-spend year, ROAS up 34% to 7.1x, cost per session down 27%, session-to-customer conversion up 27%. The cost was volume. Revenue indexed at 54 against a peak of 100. That is what delivering profitability through a CAC ceiling looks like: the machine gets better while it gets smaller.

The first quarter after the study window suggests the cost environment shifted. Sessions up 18% year over year, new customers up 40%, orders up 68%, on the cheapest Google CPC in the dataset. The open question for the operator is the one this report exists to frame: how far can spend scale back up before it finds the wall again?

This is one business in one strongly seasonal consumer vertical, so the rhythm and the specific ceiling are its own. Measurement is a blend of platform-reported sources rather than a single attribution model. One quarter of spend data is missing (Q4 2024, a low-spend quarter). A welcome-flow outage in late 2025 adds operational noise to the flow figures, and it is flagged rather than smoothed. And the scale figures are indexed, so the report argues entirely in ratios. That is a constraint I accept happily, because the ratios are the finding.

What it can say: acquisition costs cap out structurally, not anecdotally. Owned channels are downstream of paid traffic, and starving the top of the funnel shows up 18 months later dressed as an email problem. And efficiency and volume can move in opposite directions for years, which is why a single topline number tells you almost nothing about whether a marketing operation is working.