Executive summary
Two separate, overlapping problems — not one root cause.
① Chronic — a slow-building UX friction problem (Jan–Jun)
Checkout dead-click rate climbed steadily from 67.4% in March to 75.7% in June. Quick-backs rose in step, 20.4% → 33.2%. This has been degrading baseline checkout conversion for months — it did not suddenly appear on July 1.
② Acute — a sharp, unexplained event starting July 1
Google Ads and Meta CPA hit all-time highs the week of Jul 13–19. GA4 purchase conversion rate cut roughly in half (≈2% → ≈1%) starting July 2, across every browser and device — not just Safari/WebKit as previously diagnosed. The chronic dead-click rate did not spike in July (it plateaued), so it does not fully explain this acute drop on its own.
Best current lead for the July 1 event
The "Pay Now" button's dead-click rate broke its own 3-month trend: stable 33–36% (Mar–May) → dipped to 14.1% in June → spiked to 41.7% in July, its highest rate on record. This is the cleanest, most isolated signal found so far connecting a specific checkout element to the timing of the acute drop.
What's been ruled out
Findings that looked promising but didn't hold up under closer testing.
| Hypothesis | Verdict | Why |
|---|---|---|
| San Jose GA4 session spike = real demand | Ruled out | Known artifact of Meta data-center IPs / iCloud Private Relay geolocation, not genuine users |
| Payment processing / gateway broke | Ruled out | Mixpanel Pay Attempt and CheckoutStepCompleted move in parallel with no widening ratio; GA4 transactions/purchases/revenue stay internally consistent |
| JS errors are the main driver | Ruled out | Only 51 of 904 checkout users (5.6%) hit any JS error over the full period — too low volume to explain a 20–30% funnel-wide decline |
| Safari/WebKit-specific bug is the whole story | Partially ruled out | Conversion rate fell across every browser/device combo post-Jul 1, including desktop Chrome (−40%) and Edge (−50%) — not just Safari |
| Dead clicks caused the July 1 cliff | Partially ruled out | Aggregate dead-click rate was flat/plateaued in July (75.72%→75.31%); it's a chronic issue, not a new July trigger — except "Pay Now" specifically, which did spike |
Google Ads & Meta US — CPA trend
Weekly spend, conversions and CPA. Both channels reach their highest CPA on record in the last two weeks of the window.
Google Ads US · peak weekly CPA
$119.57All-time high
Week of Jul 13–19, vs $41–65 typical Jan–May range
Meta US · peak weekly CPA
$179.54All-time high
Week of Jun 29–Jul 5, still elevated at $118–152 since
Google Ads conversions nearly halved while spend held flat
Week of Jun 15–21: ~190 conversions on $12.3k spend. Week of Jul 13–19: ~84 conversions on $10.1k spend — spend stayed within a similar band, but reported conversions fell by more than half. This is the Maximize Conversions feedback loop in effect: a weaker conversion signal pushes the algorithm to bid less efficiently, compounding the original problem.
Where the spend is going (Jul 1–19)
Both platforms show the same pattern: efficient, high-intent segments are starved of spend relative to broad/prospecting segments.
Google Ads — campaign CPA (Jul 1–19)
| Campaign | Spend | Conv. | CPA |
|---|---|---|---|
| Branded | $4 | 15.1 | $0.29 |
| Non-Brand (Southern + Northern) | $1,721 | 83.6 | $20.59 |
| PMAX Southern Generic BAU | $6,809 | 33.2 | $204.79 |
| SEM Southern Generic Occasions | $6,484 | 30.3 | $213.75 |
| PMAX Northern Generic BAU | $6,462 | 42.9 | $150.53 |
| SEM Consolidated — Empire State Bldg | $857 | 1.2 | $703.35 |
The three "Generic" PMAX/SEM campaigns = ~70% of total US Google spend, all at 4–10x the CPA of non-brand/branded.
Meta — campaign CPA (Jul 1–19)
| Campaign | Spend | Purch. | CPA |
|---|---|---|---|
| Southern Dynamic Remarketing | $478 | 9 | $53 |
| Northern Dynamic Remarketing | $441 | 10 | $44 |
| Southern Advantage+ | $6,425 | 31 | $207 |
| Northern Advantage+ | $3,835 | 25 | $153 |
| Southern Traffic objective | $1,049 | 1 | ~$1,049 |
| Northern Traffic objective | $253 | 0 | — |
Remarketing runs 3–4x more efficient than Advantage+ broad, but gets a fraction of the budget. Traffic-objective campaigns are burning spend for near-zero purchases.
Distinct anomaly — not part of the current trend
May 28–30: Meta clicks spiked 5–12x normal (7,838 / 9,026 / 11,940 vs ~1,000–1,300 typical) while spend and purchases stayed flat. GA4 sessions spiked in lockstep (8,354 / 8,964 / 10,506) with bounce rate hitting 73–84%. Classic signature of low-quality/invalid traffic, not real demand. Self-resolved by June 1 — worth a dedicated quality check if it recurs.
GA4 — the July 1 conversion cliff
US purchase conversion rate (purchases ÷ sessions), daily. Transactions, purchases and revenue all move together throughout — this rules out a GA4 tracking/tagging gap. The drop reflects real completed purchases, not a measurement artifact.
Pre-Jul 1 (Jun 24–Jul 1)
1.7%–2.6%Jul 1 itself: 2.64%, best day
Post-Jul 1 (Jul 2–19)
0.5%–1.4%Sustained 18 days
Browser & device breakdown — why this rules out "Safari-only"
Conversion rate by segment, 11 days pre-Jul 1 vs 18 days post. Every browser fell, including desktop Chrome and Edge — platforms unaffected by the previously-diagnosed WebKit issue.
Funnel stage erosion
Stage-to-stage conversion ratios, pre- vs post-Jul 1 (Mixpanel: AddedToCart, CheckoutStepCompleted, Pay Attempt).
| Transition | Pre-Jul 1 | Post-Jul 1 | Change |
|---|---|---|---|
| AddedToCart → CheckoutStepCompleted | 72.8% | 67.2% | −5.6 pts |
| CheckoutStepCompleted → Pay Attempt | 60.1% | 58.0% | −2.1 pts |
Reading this correctly
The bigger relative erosion is at cart→checkout-entry, roughly double the decline seen at checkout→payment. This is a broad-based decline across the whole back half of the funnel, with a mild concentration at the earlier step — not one stage that catastrophically broke while others held steady.
Clarity — checkout dead clicks, Mar–Jul
US traffic, checkout page. Dead clicks and quick-backs climb steadily through June, then plateau in July — right when the acute conversion drop hits.
| Month | Dead clicks | Quick backs | Sessions |
|---|---|---|---|
| March | 67.41% | 20.36% | 948 |
| April | 70.52% | 24.31% | 1,201 |
| May | 75.05% | 30.25% | 1,395 |
| June | 75.72% | 33.16% | 1,565 |
| July MTD | 75.31% | 30.56% | 733 |
Read this as two phenomena, not one
The overall rate climbing Mar→Jun is a slow, chronic degradation — real and worth fixing, but it can't explain a sharp new drop on July 1 since it plateaus (doesn't spike) right when the cliff hits. The element-level breakdown below is where the July-specific signal actually shows up.
Element-level breakdown — "Pay Now" is the standout
% of checkout sessions with a dead click on each named element, by month.
Pay Now
41.7%Jul spike
Stable 33–36% Mar–May, dipped to 14.1% in June, then broke its own record in July — the cleanest lead found for the Jul 1 event.
Redeem
35.7%Rebounded
Also dipped in June (15.7%) and rebounded in July, but stayed within its Mar–Apr historical range (39–45%) — not a new record.
CSS font-fallback artifact
349Growing
"/* Optimist Extra Light *" — a raw CSS comment rendering as clickable text, implying a failed font load. Present since March (54→110→349 dead clicks Mar–May), a genuine standalone bug worth a direct fix.
The June dip explained
Both Redeem and Pay Now dipped sharply in June (15.7% / 14.1%) while the site-wide dead-click rate hit its highest point (75.72%) that same month. Non-Redeem/Pay Now elements picked up the volume: the blank/unlabeled area (27.5% of sessions), the payment-method selector (4.3%), masked input fields (4.9%), and the coupon/gift field (9.4%). This suggests June's spike was a separate, still-unidentified issue distinct from the June/July Pay Now-specific pattern.
Confirmed: JS errors are not the volume driver
Checkout users w/ any JS error (Jun 24–Jul 19)
51Small
Out of 904 total unique checkout users = 5.6%
Dead-click rate within JS-error sessions
30.91%
Lower than the general checkout population (75.02%) — errors and dead clicks are not the same population
hCaptcha anomaly (single live session)
2xWatch
Same challenge ID requested twice in a row — plausible silent friction point, not yet confirmed at volume
UK / CA / NZ — Smart Bidding recovery post-tracking migration
Weekly Google Ads performance, pre- vs post- the Jul 2 conversion tracking migration (reverted Jul 8). As of Jul 19 — 3 days before the Jul 22 checkpoint — none show a confirmed recovery.
UK
↓ Still declining
Conversions fallen 3 straight weeks post-migration (130→110→107→84). CPA climbed each week to £51.25, the worst on record.
CA
↕ Volatile
Most unstable market. Brief bounce Jul 6–12 reversed last week — CPA back to $69, matching the worst week since migration.
NZ
↑ Tentative positive
Best week of the four in the most recent period, but conversion counts are single-digit — too low-volume to call a confirmed recovery.
Building the case: TOFU vs. BOFU
Where the evidence currently sits, and what would close the remaining gaps.
Evidence pointing to BOFU
- Sessions held steady/grew through July — this isn't a traffic-quality (TOFU) story where volume and quality shift together messily
- Conversion rate cliff hit uniformly across every browser/device — consistent with a checkout-side break, not a targeting problem
- "Pay Now" dead-click rate broke its 3-month trend exactly at the point of the cliff
- CheckoutStepCompleted → Pay Attempt ratio held steady — rules out payment gateway failure specifically, but doesn't rule out earlier BOFU friction
Gaps still open
- No confirmed root cause for the July 1 event — "Pay Now" dead clicks are the strongest lead, not yet a proven mechanism
- June's dead-click spike (75.72%, driven by different elements) remains unexplained and undated against any known change
- No deploy/change log available — root cause needs code-level access (GTM export, view-source, or dev collaboration) that hasn't yet been obtained
- New Relic (best available error/AJAX-failure source) inaccessible — account provisioning still pending
Recommended next steps
- Watch 5–10 individual Clarity session recordings on "Pay Now" dead clicks specifically, post-Jul 1, to confirm what's actually happening on click (disabled state? overlay? silent JS no-op?)
- Request New Relic org access, or ask whoever administers it to pull checkout-page JS errors + failed AJAX/fetch calls to the payment gateway domain, June 24–Jul 19
- Ask whoever owns hCaptcha to check challenge/verify success rates around Jul 1 — the double-challenge-ID pattern seen in one live session is worth checking at volume
- Get a clean answer on what changed in June to spike non-Pay-Now/Redeem dead clicks (blank/unlabeled area, payment selector, masked fields) — this may be a second, separate incident
- Treat the chronic dead-click climb (Mar→Jun) as a standing UX fix item, separate from the July 1 investigation — both are real, but conflating them will weaken the case