1Executive summary
Two unrelated incidents hit paid media performance in the first week of July. In CA/UK/NZ, a conversion-tracking switch undercounted real purchases, inflating CPA — this has been reverted. In the US, a planned Meta creative refresh landed the same day a top-performing campaign wound down, triggering a delivery disruption; a separate, genuine site-side conversion-rate decline in the US is still being investigated independently of paid media. All three are broken down below, with a proposed follow-up audience strategy test for US Meta on the second tab.
Blended CPA (Google + Meta combined) for each market, before vs after the tracking switch.
Canada — blended CPA
$53 → $106 (+98%)
Google alone: $44 → $151 (+244%). Meta wasn't affected by the tracking switch, so blending softens the visible hit.
UK — blended CPA
$35 → $73 (+106%)
Google alone: $33 → $95 (+191%). Same pattern — Meta held steady and diluted the blended number.
NZ — blended CPA
$34 → $35 (+4%)
Google alone: $50 → $69 (+40%), but NZ volume is thin — 3–5 conversions total, so treat with caution either way.
US — blended CPA (Google + Meta)
$91 → $168 (+85%)
Separate root cause — not the CA/UK/NZ tracking issue. See sections 3–4.
2Timeline: what changed, and when
Two independent event chains running in parallel — CA/UK/NZ tracking (green) and US Meta delivery (navy).
Jul 1
Planned "Summer NH" creative refresh launched — US Meta
Dozens of new UGC video and image ads deployed into the Broad 25–65 M/F Advantage+ ad sets (Northern & Southern). Coordinated, planned launch.
Meta
Jul 1
Flash Sale campaigns paused as the sale rush concluded — US Meta
The three Flash Sale campaigns (Advantage+, Remarketing + Retention, Dynamic Remarketing) had driven a strong sale rush through late June and were intentionally paused once that promotion wound down — on the same day the new Summer creative went live.
Meta
Jul 2–3
Compounding effect begins: new-creative learning phase + freed-up budget
With Flash Sale off and unproven new creative live in the same ad sets, Meta's delivery broadened sharply — clicks nearly tripled (434 → 1,259/day) at similar spend, while purchases fell (9.7 → 3.8/day). Same window: the Purchase pixel briefly logged zero conversions on Jul 3.
Meta
Impact begins
Jul 2–3
GA4 set as primary conversion source — CA/UK/NZ
The original "Purchase" pixel (reliable count, flat $1 value) was set to secondary. The new GA4 action under-counted real purchases by 66–72% relative to the prior week, starving Maximize Conversions bidding of signal in all three accounts.
Google Ads
Impact begins
Jul 7–8
CA/UK/NZ reverted to pixel-primary; GA4 kept running in parallel for reporting
Restores a reliable count signal for bidding. Value-based bidding remains a future goal once the GA4 import gap is validated against real orders.
Recovery in progress
Ongoing
US Meta ad sets still absorbing the Jul 1 disruption
No reversal action taken yet. Delivery should stabilize as the new creative exits its learning phase, assuming no further competing changes.
Meta
Note: dates above reflect the ad platform's own account timezone. Because we report in AU time, some of these events may land on a different calendar day when viewed in Ads Manager or Business Manager directly — worth double-checking against the local view before quoting exact dates externally.
3Google vs Meta — US market, side by side
Same week, same market, two different root causes. Meta's decline traces to a specific, dated delivery disruption; Google's traces to a genuine conversion-rate drop that shows up site-wide — not a competitor or budget issue.
Google Ads — US
Conversion rate issue
Purchase pixelFiring normally throughout — not the cause
Clicks (Jun24–30 → Jul1–6 avg/day)695 → 694 (flat)
Conversion rate~2.3% → ~1.3% (−43%)
CPA (avg/day)$84.74 → $145.88 (+72%)
- Not a tracking problem: the pixel logged conversions correctly on every day in this window — the drop is in real visitor behavior, not measurement.
- Not competitor pressure: search impression share held flat (~10–14%) before and after; rank-lost impression share was already high and unchanged — no new competitive event on Jul 1–3.
- Not a budget cap: budget-lost impression share did not spike — campaigns weren't newly constrained.
- Real conversion rate drop: clicks stayed flat but the rate at which visitors converted fell sharply from Jul 2 — and it lines up with a site-wide engagement decline (see Section 4) across all channels, not just Google.
In short: Google Ads is sending the same quality and volume of traffic as before — fewer of those visitors are converting once they land. That points to the site experience, not the ad account.
Daily clicks: Google (stable) vs Meta (surge from Jul 1)
US market only. Meta's click volume roughly triples starting the day the new creative and campaign pause landed; Google shows no equivalent shift.
Daily CPA: Google vs Meta
Both worsen, but Meta's rise is sharper and tracks exactly to the July 1 account changes; Google's is a gradual drift consistent with a conversion-rate issue rather than a sudden shock.
4US site engagement — building the case for a real, site-wide effect
Filtered to amazingco.me/us specifically. If this were purely a paid-media problem, engagement quality on the site itself shouldn't move. It did.
Bounce rate
41.2% → 45.6%
+4.4pp (Jun24–30 → Jul2–7 avg)
Avg. session duration
175s → 120s
−31%
Pages per session
4.11 → 3.46
−16%
Daily bounce rate and sessions, amazingco.me/us
Sessions rise from Jul 2 (driven by the Meta traffic surge), but bounce rate rises with them and session quality falls — more visitors, but a worse visit on average.
Daily US sessions by channel — Google vs Meta
Cross-network and Paid Search (Google) track close together throughout. Paid Social (Meta) breaks away sharply from Jul 2.
Daily US bounce rate by channel — Google vs Meta
Paid Social (Meta) runs chronically high (~60-77%) the entire period, before and after — it doesn't get worse, there's just far more of it from Jul 2. Google's channels stay in a healthier band throughout, with Cross-network actually trending down.
Correcting the site-wide read: once broken out by channel, the story is more precise than "the whole site got worse." No individual channel's own bounce rate meaningfully declined — Cross-network actually improved, and Paid Search and Paid Social both held steady. The site-wide bounce rate rise (41.2% → 45.6%) is a mix-shift effect: Paid Social has always run a much higher bounce rate (~65%) than Google's channels (~34–44%), and that segment's volume more than tripled — pulling the blended average up even though no visitor group individually behaved worse. This means the drop in Google's purchase conversion rate is happening deeper in the funnel (past the landing page, e.g. cart or checkout), not at first engagement — worth pointing any technical audit there rather than at the top of the funnel.
5CA / UK / NZ — before vs after the tracking switch
Both Google-only and blended (Google + Meta) figures are shown, so it's clear how much of the headline number is a Google-specific tracking artifact versus the true blended-channel picture.
| Market | Google CPA before | Google CPA after | Google change | Blended CPA before | Blended CPA after | Blended change |
|---|
| Canada | $43.82 | $150.54 | +244% | $53.34 | $105.76 | +98% |
| United Kingdom | $32.69 | $95.17 | +191% | $35.40 | $73.08 | +106% |
| New Zealand | $49.68 | $69.49 | +40% | $33.51 | $34.97 | +4% |
What "+244%" means: total spend ÷ total conversions in the 7 days before the switch (Jun 25–Jul 1), compared to the same calculation for the 6 days after (Jul 2–7). Google-only CPA rose sharply because the conversion count collapsed while spend stayed flat. The blended figure adds Meta's spend and conversions from the same markets and days — Meta wasn't touched by this tracking change, so it dilutes the headline impact but is the more accurate reflection of total channel performance.
NZ volume is thin (3–5 Google conversions across the whole "after" window) — directionally consistent but not statistically strong on its own.
6Recommendations
Done
CA/UK/NZ reverted to pixel-primary conversion tracking. Allow 3–5 days (through ~Jul 12) for Smart Bidding to recalibrate before judging performance.
Priority — short term
Audit the US checkout and page-experience path directly (load time, payment step, mobile rendering) for Jul 2 onward — bounce rate, session duration, and pages/session all moved the same direction at the same time, which usually points to something visitors are hitting on-site rather than a traffic quality issue alone.
Process
Stagger future creative refreshes from campaign pauses — launching a large new creative batch the same day a top performer is paused compounds the learning-phase dip.
Monitor
Watch US Meta CPA daily; expect stabilization as the Jul 1 creative exits its learning phase, absent further changes.
1Prospecting audiences — before the sale vs after the creative refresh
Jun 16–22 (prior to the Flash Sale launch) vs Jul 1–7 (after the sale wound down and new Summer creative went live) — both 7-day windows, aligned in length for a fair comparison.
| Ad set type | Campaign | Period | Spend | Clicks | CPC | Purchases | CPA | CPA change | ROAS |
|---|
| Broad | 25-65 | MF (no interest signal) |
Advantage+ only
(Northern + Southern combined) |
Jun 16 – Jun 22 | $3,187 | 2,545 | $1.25 | 20 | $159 | — | 1.11x |
| Jul 1 – Jul 7 | $4,737 | 4,781 | $0.99 | 18 | $263 | +65% | 0.73x |
| Cart / viewed-but-not-purchased remarketing |
Dynamic Remarketing
(not Advantage+) |
Jun 16 – Jun 22 | $653 | 401 | $1.63 | 14 | $47 | — | 2.78x |
| Jul 1 – Jul 7 | $394 | 335 | $1.18 | 4 | $99 | +111% | 1.77x |
Broad | 18-55 | F&M and the Retention/Remarketing 180d pools ran the Flash Sale campaigns, which wound down intentionally on Jul 1 — spend drops to near-zero in the "after" window for those, so they're excluded from this comparison rather than shown as a misleading decline.
2Key findings
- The broad prospecting ad set is where the damage shows up clearest. Clicks nearly doubled (2,545 → 4,781, +88%) while CPC fell from $1.25 to $0.99 — Meta bought a lot more, cheaper traffic. But purchases stayed flat-to-down (20 → 18), so CPA rose 65% ($159 → $263) and the ad set flipped from marginally profitable (1.11x ROAS) to underwater (0.73x).
- This cheap-click surge is the same traffic inflating Meta's Paid Social sessions in GA4. That segment's session volume roughly tripled over the same window (see Section 4, Overview tab). More volume, lower quality, same root cause: new unproven creative plus freed-up budget pushed the algorithm into broader, less-qualified delivery.
- Even the remarketing pool wasn't fully insulated. Cart/viewed-but-not-purchased CPA more than doubled ($47 → $99) and ROAS fell from a strong 2.78x to a weaker 1.77x — smaller magnitude than broad prospecting, but the same direction, suggesting some of the disruption spilled beyond the ad sets that got new creative directly.
- No evidence this is an audience-targeting failure specifically. The broad ad set's own targeting configuration didn't change (see prior investigation) — the CPA spike tracks to volume and click quality, not a new targeting setting. That said, the results still make the case for testing a less-fragile setup (Section 3).
3Recommendations
- Don't touch the existing broad ad sets yet. They're still absorbing the Jul 1 disruption — editing targeting now would reset their learning phase again and confound any read on whether the change helped.
- Once delivery stabilizes, run a real test. Build a new ad set alongside the existing broad one using Meta's "Advantage+ detailed targeting" (a few interest/behavior suggestions layered on top of broad — not a full return to narrow targeting), and split budget for 2-3 weeks. This settles the internal debate with evidence instead of opinion.
- Shift some prospecting budget toward remarketing/retention in the meantime. Those pools ran 2x+ better ROAS in June and are a lower-risk place for spend while prospecting is unstable.