It wasn't one site break.
It was four things compounding — and three of them are ours to fix.
Between May 10 and June 10, daily new customers fell from 623 to 367 (−41%) and blended CAC rose from $88 to $129 (+47%). We walked the entire funnel with Daily Stand, Northbeam, Meta, Intelligems, Recharge, Gorgias, Clarity, and outside market research. The site did not break: returning-visitor conversion is flat, checkout and payments work, prices and offers never changed, and organic demand is actually up versus April. What broke is the cold-traffic acquisition machine — in a specific, dateable sequence.
The single most important fact we found: spend fell only 14%, customers fell 41%. If efficiency had simply held, the spend cut would have lowered CAC to ~$76. The entire CAC increase is an efficiency collapse — concentrated 100% in new visitors (CVR −27%) while returning-visitor CVR moved −1%. Whatever happened, it happened to cold paid traffic, not to the website as customers experience it.
What happened, in one chart
Daily new customers & blended CAC — March 1 to June 10
The four compounding causes
| Cause | What it did | Share of damage | Confidence | |
|---|---|---|---|---|
| CAUSE | 1. Meta engine break | New ASC campaign launched May 10 (learning reset) + budgets scaled +30–40% into decaying ad sets + the main ASC engine killed May 29 with no working replacement. Facebook first-time transactions: −108/day — half the total loss — plus a ~44/day halo loss in unattributed/brand demand. | ~60% | HIGH |
| CAUSE | 2. Spend rotated into non-converting channels | Podcast spend 2–4×’d (peaking $13.7k/day) producing ~0 tracked first-time purchases in-window (+$14/customer on blended CAC); Google's June raise went to conquesting at $170 CAC with zero added volume. | ~15–20% of CAC rise | HIGH |
| CONTRIBUTOR | 3. A/B test swarm on the money page | Test launches ramped 1→3→8→9 per two weeks. From May 15 on, 25–50% of Rise 2 traffic was on unproven redirect variants; the Side Cart test put 67% of ALL traffic on losing carts for 8 days. A continuous conversion tax, and a possible Meta signal corruptor. | ~10–15% | MED-HIGH |
| AMPLIFIER | 4. Macro demand headwind | Iran-war energy shock (gas ~$4.50/gal) — Dallas Fed research says consumption effects land 2–3 months after the March shock, i.e., exactly May. Real, but explains maybe 5–15 points of softening — not a 41% cliff. June 2025 did 564/day on the same weeks. | ~10–15% | MED |
One honest nuance about the baseline: the 623/day "golden run" (Apr 28–May 9) was partly bought with the spring sale and a perfectly tuned Meta engine. The underlying business in March–April ran ~500–550/day at $95–100. So the fix has two stages: repair the break (back to ~520–560/day, ~$95) and then earn the goal (600 @ $90) with creative volume, CVR work, and offer strategy — not just by turning dials back.
Where the customers went
Lost first-time transactions per day (May 1–9 → May 26–Jun 10)
The CAC bridge: $88 → $129
Built by the growth-diagnostic working session (analytics, growth, CX, market research lanes), June 11, 2026. Full evidence in the following pages — start with Timeline.
Two break dates. Both are dateable to the day.
Conversion broke May 10–11 (with traffic flat). Traffic broke May 28–29 (the Facebook cut). Everything else follows from those two events. Here is the reconstructed sequence with the evidence for each step.
Sitewide conversion rate by period (Intelligems, orders ÷ sessions)
Daily spend mix — the channel rotation, weekly (Mar 2 – Jun 8)
Meta efficiency by phase
Walking the funnel, stage by stage
Each stage gets a verdict. The pattern that emerges: every broken stage is specific to new visitors arriving from paid; every stage that touches existing customers is intact.
| Funnel stage | Verdict | Evidence (May 1–9 → May 26–Jun 10) |
|---|---|---|
| Auction costs (CPM/CPC) | NOT THE TRIGGER | Meta CPM fell $49→$28–35 during the break. Costs are up ~20% YoY industry-wide (structural), but the May break is not an auction-price event. |
| Ad click-through | SOFTENING | CTR 2.1%→1.5–1.6%. People click less as creative rotated away from proven ads; secondary, not primary. |
| Traffic volume | BROKE MAY 28–29 | Sessions 22.6k→19.0k/day (−16%); new visitors −15.9%. Direct result of the Facebook cut. |
| New-visitor conversion | BROKE MAY 10–11 | New-visitor CVR 2.55%→1.87% (−27%). This is ~64% of the customer loss. |
| Returning-visitor conversion | INTACT | 3.74%→3.69% (−1.3%). The strongest single piece of evidence against "the site broke." |
| Landing page → PDP | BROKE | Second-page-is-PDP fell 20.8%→15.4% of visitors; visitors ping-pong to content pages instead. Canonical Rise 2 CVR −25% (2.42%→1.82%) and its traffic −48%. |
| PDP → Add to cart | LEAKING | Visitor→ATC 4.30%→3.65% (−15%) — ~70% of the CVR drop happens at or before ATC. |
| Cart → Checkout | INTACT | 94.4% → 94.4%. Identical. |
| Checkout → Purchase | WATCH | 62.7%→58.6% (−4.1pp; the prior doc's "59→52" was exaggerated). But Jun 4–9 abandonment ran 32–36%/day vs 24–27% baseline — recent, unexplained, needs session replays. |
| Offer / price / shipping | UNCHANGED | AOV flat ($48.6→$49.4), median order identical ($34), discount rate flat (13.2→13.3%), zero true price tests, Recharge price points unchanged. |
New vs returning visitor CVR
Checkout funnel steps
The canonical Rise 2 page — correcting the prior analysis
The test swarm
Channel first-time CAC by phase (Northbeam, clicks-only)
"It has to be a site issue we can fix" — we tested that. Here's everything we cleared.
This was the leading internal theory, so we attacked it from five independent directions: funnel analytics, checkout data, payment systems, customer tickets, and subscription infrastructure. Ten specific failure modes checked; all ten cleared. A real site break would show up in returning visitors, in support tickets, and in payment logs. It shows up in none of them.
| Hypothesis | How we checked | Result |
|---|---|---|
| Site outage / broken templates | Returning-visitor CVR (Intelligems), Shopwatch theme log | CLEARED Returning CVR flat (3.74→3.69%). No published theme changes — heavy churn was all unpublished dev branches. |
| Checkout/payment failure | Recharge failed-charge log, funnel steps, Gorgias tickets | CLEARED Cart→checkout identical (94.4%). Failed charges are 100% ordinary card declines; no gateway errors, no novel error types. Zero ticket cluster about checkout. |
| Price increase / offer removed | Recharge price points, Intelligems discount & AOV data, ticket keyword search | CLEARED $40/$100 price points unchanged Apr–Jun. Discount rate flat. "Price went up" complaints: zero; price-related cancel reasons declined as a share. |
| Subscription widget broke | Recharge new-sub starts by window | CLEARED Sub starts fell slightly less than total NCs at every step (−16% then −34/−39% vs −23%/−41%). Both purchase paths declined proportionally → upstream cause. |
| Hero SKU out of stock / PDP break | Recharge product mix of new subs | CLEARED 30-serving Original holds 63–67% share throughout; Coffee SKU share actually growing. |
| A single bad A/B test shipped | Full Intelligems experiment audit, variation-level CVRs | CLEARED (as a single cause) Nothing started or ended May 9–13; nothing shipped to 100% before Jun 8. The swarm is a contributor; no one test explains the cliff. |
| Customer-visible breakage (any kind) | Gorgias top issues, CSAT, keyword search | CLEARED CSAT 4.85/5. No spike in site/checkout/discount complaints. (Caveat: Gorgias tooling only exposes ~2 weeks of history — detection power limited, but Recharge independently corroborates.) |
| Demand collapse for the brand | Northbeam organic + unattributed transactions | CLEARED Organic first-time transactions UP 2× vs April (21→43/day). People who find us still buy. (Brand search is softening −18% — that's the Meta halo, see Causes.) |
| Seasonality ("it's just summer") | Daily Stand YoY: same calendar days 2025 | CLEARED AS PRIMARY Jun 1–10, 2025: 564/day at $98 on the same ~$55k spend. May→June 2025 showed no cliff. We are −35% YoY at similar spend. |
| A May Meta algorithm break | External research: advertiser communities, agency reports, status pages | CLEARED No documented May 2026 platform break; when Meta breaks broadly, it's loud (cf. March 2026). A 1–3 day Ads Manager outage May 5–7 can't explain 5 weeks. |
Year-over-year: same days, same spend, very different output
What this means strategically: there is no single "fix the bug, get 600 back" lever. That's actually good news in disguise — the causes we found are operational and reversible (media structure, budget allocation, test discipline), not a mystery regression or a market that stopped wanting the product.
Where the prior report was wrong
The Cowork diagnostic got the broad shape right (upstream problem, not a Rise 2 rebuild) but four load-bearing claims didn't survive re-verification:
| Prior claim | What the data actually says |
|---|---|
| "Blended CAC $92→$95" | $88→$129. The prior pull used Northbeam lifetime attribution and the wrong denominator. Company standard (tracker spend ÷ Shopify NCs) shows a far bigger problem. |
| "Canonical /pages/rise-2 is healthy (CVR flat 1.84→1.87%)" | Wrong. Canonical CVR fell 2.42%→1.82% (−25%), traffic −48%, orders −61%. The "flat 1.84%" matched only the post-cliff level. |
| "Variant-page mix is the main CVR drag" | Real but ~10× too small: ≈3 orders/day. The canonical page's own decline is the story. |
| "Checkout→buy dipped 59%→52%" | 62.7%→58.6% (−4.1pp). Directionally right, magnitude exaggerated. |
| "Main driver is auction inflation (eCPC up)" | Meta CPM/CPC fell during the break. The driver is account restructuring + post-click conversion collapse + spend mix, not auction prices. |
Root causes, ranked
Ordered by estimated share of the damage, each with the mechanism, the evidence, and our confidence. Together they reconstruct the full −256 customers/day and +$41 CAC.
Meta account restructuring broke the engine — then the fix made it worse HIGH CONFIDENCE · ~60% OF VOLUME LOSS
The sequence: a new ASC scaling campaign launched May 10 at 1:54am (fresh learning phase on the most important engine) while spend spiked +28% the same day. Over the next two weeks budgets were raised +30–40% into ad sets whose CPA was visibly degrading ($113→$148). On May 29 the whole engine was killed (−$10k/day). Every replacement has underperformed: new ASC $139 CPA, "Purgatory" $252, MixedAudiences-Testing $164 — versus ~$106 in the golden run. Facebook first-time transactions fell −108/day (half the total loss), and the halo that Meta prospecting feeds — unattributed and brand-search conversions — fell another ~44/day. Brand-search transactions are down 18% despite 2.2× brand spend: the top of the funnel is starving the bottom.
Why it reads as "something broke on the site": Meta responded to the disruption by buying cheaper, lower-intent inventory (CPM fell $49→$28–35) — so the same landing page received colder traffic and converted 27% worse. The page didn't change; the people did.
Budget rotated from converting to non-converting channels HIGH CONFIDENCE · MECHANICAL
Late May: podcast spend jumped from ~$3.3k to as high as $13.7k/day (June average $7.0k) — with 0–2 tracked first-time transactions in the entire window. At June volume that's +$14 per customer on blended CAC, all by itself. Google's June raise (+$3.9k/day vs early May) went to conquesting (CAC $94→$170, zero incremental transactions) and to saturating brand search ($13→$36 CAC on a shrinking demand pool). Podcast may pay back on a lag as a brand bet — but it's sitting inside the CAC budget line, distorting the metric the team is managed on.
The A/B test swarm taxed the money page — and may be corrupting Meta's signal MED-HIGH CONFIDENCE · CONTRIBUTOR
Launch cadence went 1 → 3 → 8 → 9 per two-week window. From May 15 onward, 25–50% of Rise 2 traffic sat on unproven redirect variants at all times; 71 of 79 active Meta ads point at this one URL, so every test touches nearly all paid volume simultaneously. The Side Cart Redesign (May 18–26) put ~67% of all site traffic on variants that converted worse. Several tests were killed within 1–2 days — too fast to learn anything, long enough to lose orders. Estimated direct drag ~0.1–0.2pp of sitewide CVR. The open question (highest-value thing to validate next): whether redirecting ad traffic mid-session feeds Meta inconsistent landing-page signals and degrades its optimization — the post-click collapse signature is consistent with this, but unproven.
A real macro headwind arrived in May — the amplifier, not the cause MED CONFIDENCE · 5–15 PTS
The March energy shock (gas heading toward $5/gal, CPI 3.3%) hits consumption with a 2–3 month lag per Dallas Fed research — landing exactly in May. Consumer confidence dipped in May; discretionary wellness is exactly the category that gets cut. Competitors (RYZE in Target with 4,100+ active Meta ads, AG1 in Target since April) keep auction pressure up. All real; none of it explains −41% in four weeks when last June did 564/day — and high grocery coffee prices (+19% YoY) are actually a tailwind for our "cheaper than your coffee habit" positioning.
Structural fragility underneath it all: creative concentration MED CONFIDENCE · RISK MULTIPLIER
One founder-style creative carries ~25–30% of all Meta spend (it's fine — CPA $104→$115, CTR rising). The #2 creative — a spring-sale sticker video — fatigued from $74 to $139 CPA while being scaled 7×, and is still running with stale sale messaging in June. Nothing in the new-creative pipeline matches the founder ad (CTRs 0.9–1.5% vs 2.3%). In the Andromeda era, Meta rewards creative volume; we are one ad away from a much worse month. This is also why the brand/positioning work already underway matters — it's the feedstock for the creative volume we need.
Watch list (real, not the trigger)
| Item | Status | Note |
|---|---|---|
| Checkout abandonment Jun 4–9 (32–36%/day vs 24–27% baseline) | OPEN | Recent, unexplained by offer data. Needs session replays + payment log review this week. |
| Meta health/wellness policy tightening (Q1 2026) | CHECK | Industry rejection rates +34% for supplements. Verify our account has no new restrictions/disapprovals from April–May. |
| Subscription onboarding churn | CHRONIC | 35% of May 1–9 new subs already cancelled ("didn't realize I signed up"). Pre-existing, not a May change — but it overstates durable NCs and burns LTV + support load. |
| Site performance floor (CLS ~0.7, mobile script errors ~6% of sessions) | CHRONIC | Couldn't verify trend (Clarity history unavailable via API) — chronic tax, raise-the-floor fix, not the May trigger. |
Roadmap: repair the break, then earn the goal
Stage 1 (weeks 1–2) reverses the self-inflicted damage — target ~520–560/day at ≤$100. Stage 2 (weeks 3–6) rebuilds toward 600 at $90 with creative volume and CVR recovery. Stage 3 fixes the structural debt so this can't happen silently again.
Stage 1 — This week: stop the bleeding
Stage 2 — Weeks 2–6: rebuild toward 600 @ $90
Stage 3 — Structural (30–90 days)
The recovery math
| Lever | NC/day | Blended CAC | Basis |
|---|---|---|---|
| Today (Jun 1–10) | 367 | $129 | Daily Stand |
| + Meta engine rebuilt to ~$32k/day at golden-run efficiency | +80–100 | −$15–20 | FB FT CAC ~$145–155 achieved Apr–May at this spend |
| + Halo recovery (unattributed + brand search follows Meta) | +25–40 | −$5–8 | Halo fell ~44/day when Meta was cut; partially reverses |
| + Test freeze → new-visitor CVR 1.87%→2.2%+ | +35–55 | −$8–12 | Returning CVR proves the site can convert; Apr baseline was 2.28% sitewide |
| + Podcast & conquesting reallocation/recategorization | +10–25 | −$10–16 | Mechanical |
| Stage 1+2 landing zone (4–6 weeks) | ~520–580 | ~$92–100 | |
| To 600 @ $90 | Creative volume + offer window + CVR program | The last 10% is earned, not restored — see Stage 2 | |
Honesty note: 600 @ $90 was only sustained for ~12 days this year, during a promo with the Meta engine at peak. Treat Stage 1+2 as the commitment, and the goal as the target the creative + CVR + offer program has to earn.
Every question we asked, how we answered it
Four parallel workstreams (growth/media, site analytics, CX/subscriptions, market research) ran against live systems on June 11, 2026. This page is the audit trail: sources, queries, answers, dead ends, and what we still don't know.
Questions & answers
| Question | Source / method | Answer |
|---|---|---|
| When exactly did it break? | Daily Stand daily series Mar 1–Jun 10 (+ 2025 tabs for YoY) | CVR: May 10–11. Traffic: May 28–29. Two events, not one. |
| Traffic-driven or conversion-driven? | Intelligems sitewide sessions/orders by period | ~64% conversion / ~36% traffic; conversion broke first with flat traffic. |
| Did the site break? | Returning-visitor CVR, Shopwatch, Recharge, Gorgias, payment logs | No — 10 failure modes checked, all cleared (see Ruled Out). |
| Did Meta CPA break before or after the spend cut? | Meta daily insights, campaign creation log | Before — by 19 days. Cut was reactive. Re-broke after (Jun 4) because replacements underperform. |
| What structural Meta changes happened? | Campaign list created_time forensics | New ASC May 10 1:54am; 2 ABO campaigns May 13; engine killed May 29; 3 weak replacements since. |
| Did a bad A/B test cause it? | Full Intelligems experiment audit, variation CVRs | No single test; the swarm (1→3→8→9 launches/2wks) is a contributor. Side Cart test worst single item. |
| Did price/offer/shipping change? | Intelligems pricing flags, AOV/discount data, Recharge prices | No. Zero true price tests; all order economics flat. |
| Did customers notice anything? | Gorgias issues/CSAT/keyword search; Recharge sub starts | No complaint spike; CSAT 4.85; both purchase paths declined proportionally. |
| Is it the economy/competitors/season? | Web research (Dallas Fed, Conference Board, agency/community reports) + YoY internal | Real headwind worth 5–15pts, arriving ~May. Not −41%. June ’25 did 564/day same weeks. |
| Was there a May Meta algorithm break? | Advertiser communities, status trackers, agency analyses | No documented break (March 2026 was the disruption; we sailed through it fine until May 10). |
| Where did the customers go, by channel? | Northbeam clicks-only FT transactions by platform | FB −108/day, halo −44, YouTube −15, Google −10, organic +11. |
| What did the podcast spike buy? | Northbeam + Daily Stand cross-check | 0–2 tracked FT transactions; +$14/NC on blended CAC at June volume. |
Dead ends & data caveats — read before quoting numbers
| Limitation | Impact |
|---|---|
| Northbeam MCP export locks to lifetime attribution (7-day-click unavailable via API) | Channel CAC levels are inflated vs the dashboard convention; cross-window trends are valid. Blended headline numbers use Daily Stand, which is unaffected. |
| Meta landing-page performance API returned empty | Couldn't get CPA-by-URL from Meta's side; LP mix inferred from creative link audit (71/79 ads → /pages/rise-2). |
| Clarity API exports only the last 1–3 days | JS-error and CLS trends unverifiable; chronic-vs-acute judgment rests on the prior report's manual pulls. |
| Gorgias API only exposes ~2 weeks of tickets | No before/after ticket-volume comparison; complaint findings corroborated via Recharge instead. |
| Recharge counts subscription line items, caps some queries at 2,500–3,000 records | Trends valid; absolute levels don't map 1:1 to NC counts; cancellation volume comparison censored. |
| Meta "purchases" include returning customers | Meta CPA ≠ new-customer CAC; used for timing/trends only. |
| May 1–9 baseline was promo-elevated | All "vs May 1–9" deltas overstate decline vs a normal week by roughly 10–15%; April baselines shown where it matters. |
Open questions, in priority order
| # | Question | How to close it |
|---|---|---|
| 1 | Does Intelligems URL-redirect testing corrupt the Meta pixel/CAPI landing-page signal? | Instrument pixel URL on redirected sessions; A/B one direct-LP vs one redirect-LP ad set. (Roadmap #7) |
| 2 | What's behind Jun 4–9 checkout abandonment (32–36% vs 24–27%)? | Session replays + payment logs. (Roadmap #5) |
| 3 | Did Meta's Q1 health/wellness policy tightening restrict our account? | Audit rejections + sensitive-category flags. (Roadmap #9) |
| 4 | Channel CAC at 7-day-click (company standard) | Manual Northbeam dashboard pull; reconcile against this report's trends. |
| 5 | Rise 2-filtered Clarity history (CLS/error trend on the money page) | Clarity dashboard UI, custom date ranges. |