The top landing pages behind the highest-spend campaigns — where they leak, why, and the two changes most likely to move conversion rate.
Ten live campaigns spend into seven distinct landing pages. Most of the money lands on very few of them: a single LeadsHook form, reached via experiment /e/v/299, absorbs $36,798 — 43% of measured spend — from four different campaigns selling four different products.
The 90-day and 30-day pictures agree, so this is structural rather than a blip: CvR 14.6% over 90 days vs 14.8% over 30, CPA $89 vs $91. Nothing has recently broken — these are standing conditions.
The best-built page in the account is being fed the worst traffic, and the worst-built page is carrying the most money.
The short-term-rental page (start.gerrards.co.nz) does almost everything a modern quote page should — specific headline, named objection, trust badges, progress bar, compact options. It converts at 9.6%, because only 36% of its spend lands on short-term-rental intent.
The /e/v/299 form does almost nothing right — no headline at all, no value proposition, no progress indicator, five giant option cards spanning 1.8 screens. It converts at 20.2%, carried entirely by high commercial intent.
That asymmetry is the opportunity. Gerrard's already owns the pattern that works. It just isn't pointed at the money.
Source: diag_ads.csv, weeks 2026-05-11 → 2026-08-03. General Business excluded here because it splits across two pages by ad group — see §2.
Every dollar, traced to the page it lands on. This is the view that makes the problem obvious.
| Campaign | Landing page | Cost | Clicks | Conv | CvR | Ad promise → first question |
|---|---|---|---|---|---|---|
| Commercial Property | /e/v/299 | $16,107 | 679 | 111.0 | 16.3% | "Commercial Building" → partial Building/Property |
| General Business | …x5ulc + …BRd9t5 | $16,615 | 842 | 127.3 | 15.1% | Two pages, split by ad group |
| CMV | /e/v/299 | $10,134 | 781 | 217.6 | 27.9% | EXACT "Commercial vehicle" → Commercial Vehicles |
| Rural | page.gerrards.co.nz/ | $7,905 | 777 | 126.0 | 16.2% | "Lifestyle Block" → matches, but no form on page |
| Airbnb | start.gerrards…/f/short-term-rental | $7,385 | 490 | 47.0 | 9.6% | EXACT "Short Term Rental" → Short-term rental |
| Truck | /e/v/299 | $6,910 | 505 | 75.5 | 15.0% | "Commercial Truck" → partial Commercial Vehicles |
| Contract Works | …Q3bH1J | $5,202 | 536 | 67.0 | 12.5% | "Contract Works" → matches |
| Digger | /e/v/299 | $3,647 | 249 | 44.0 | 17.7% | NO MATCH "Digger/Excavator" → no such option |
| Trades V3 | /e/v/349 | $1,577 | 176 | 34.0 | 19.3% | "Trades" → generic option list |
| Total measured | 7 distinct pages | $75,482 | 5,035 | 849.4 | 16.9% |
/e/v/299 and /e/v/349 are LeadsHook split tests. Both resolve into the same decision tree that General Business sends traffic to directly (…x5ulc0Xq…). Six campaigns — Commercial Property, CMV, Truck, Digger, Trades and General Business — are effectively sharing one generic insurance application form.
The /e/v/299 redirect also costs about 0.6 s of LCP (2.14 s via the experiment vs 1.46 s hitting the tree directly), because it adds a full server round-trip before anything renders.
Affects $36,798 / 2,214 clicks Highest expected value
Someone searches "digger insurance nz". They read an ad that says "Digger Insurance — Protect Your Digger Investment". They click. The page then asks:
WHAT DO YOU WANT TO PROTECT?
Business Liability · Commercial Vehicles · Tools & Equipment · Building/Property · Stock/Inventory
There is no option for a digger. No excavator, no machinery, no plant. The visitor has to decide for themselves whether a 20-tonne excavator counts as a "Commercial Vehicle" or as "Tools & Equipment" — and they have to do it while re-answering a question their click already answered.
Four campaigns hit this identical page, which lets us compare alignment while holding the page constant. The one campaign whose ad promise maps exactly onto an option converts far better than the three that don't:
Campaign on /e/v/299 | Ad → option alignment | Clicks | CvR | CvR on-target terms only |
|---|---|---|---|---|
| CMV | Exact match | 781 | 27.9% | 32.8% |
| Commercial Property | Partial | 679 | 16.3% | 27.1% |
| Digger | No option exists | 249 | 17.7% | 15.8% |
| Truck | Partial | 505 | 15.0% | 12.8% |
CMV converts 11.8 points higher than the other three combined (27.9% vs 16.1%, z = 6.59, p < 0.001). Restricting to on-target search terms only — which controls for the fact that CMV also buys cleaner traffic — the gap holds: the two exact/near-match campaigns run 27–33%, the two mismatched ones run 13–16%.
This is a natural experiment, not a controlled test. Verticals differ in intrinsic conversion propensity, and CMV also buys better traffic (81% of its spend lands on commercial-vehicle intent, versus 41% for Truck and 36% for Digger). Some of the gap is category, not alignment.
But two parts of this need no statistics at all: the page has no headline, and a digger has no option to click. Both are defects on inspection, and both are worth fixing whether or not the alignment hypothesis is fully correct.
Pass the vertical in the URL and have LeadsHook skip or pre-select the category question. Truck ads land with "Commercial Vehicles → Truck" already chosen; Digger ads land on a machinery branch that currently has no entry point at all.
This removes a full step from 100% of the traffic on this page and eliminates the self-classification problem entirely.
The page currently has no <h1>. A visitor's first screen is a review bar, a 250 px logo block, then a bare question. There is no statement of what this is, what they get, how long it takes, or why Gerrard's.
Add a vertical-specific block above the question, matching the pattern the short-term-rental page already uses:
Five options at 118 px each plus gaps means the last option sits at 1,104 px — 1.8 screens down on an iPhone. Visitors choose from what they can see. The short-term-rental page fits three options in 1.1 screens using a 2-up grid; port that grid.
It touches the largest single pool of spend ($36.8k), it is a content and routing change rather than a rebuild, the winning pattern already exists in the account and can be copied, and two of its three components fix defects that are certain rather than hypothesised.
Affects $7,905 / 777 clicks Certain defect
The Rural / lifestyle-block campaign lands on page.gerrards.co.nz/. Two images account for almost the entire payload:
| Asset | Size | Intrinsic | Shown (CSS px) | Needed at 3× DPR | Lazy? |
|---|---|---|---|---|---|
hero-modern-home.jpg | 8,025 KB | 2763 × 1547 | 358 × 200 | 1074 × 600 | No |
rural-family.jpg | 7,140 KB | 2752 × 1536 | 358 × 400 | 1074 × 1200 | No |
| Two images | 15.1 MB | Page total 15.49 MB · 31.7 s to full load on throttled 4G | |||
Two things are wrong here, and the smaller one is the dimensions. Even on a 3× retina phone these are only about 2.5× wider than needed — roughly 3–7× surplus pixels. The real fault is compression: 8 MB for a 2763 × 1547 JPEG means the file is essentially uncompressed. A well-encoded WebP at that same full resolution would be around 400 KB; resized to what the phone actually displays, closer to 120 KB.
Resize and re-encode and this page goes from 15.49 MB to under 400 KB — a 97% reduction — with no visible quality change.
First paint is fast (LCP 0.67 s) because the hero is text, so this does not show up in a naive speed check. The damage is downstream: the browser spends half a minute pulling 15 MB while the visitor is reading and scrolling, on a page where the proof content — "Why Lifestyle Block Owners Choose Us", the insurer panel, the FAQ — sits below those images. On NZ mobile data this is also a real cost to the visitor.
Rural CvR has fallen from 16.2% across 13 weeks to 10.6% in the last four. I have not established the cause and this page weight may not be it — but it is the largest unforced defect on that page.
Unlike every other top page, this one has no form on it. Every CTA links out to quote.gerrards.co.nz/s/iapoc…. So Rural traffic pays for two page loads before it can answer a single question — ad → content page → quote form.
Worse, the CTA placement is thin: one button above the fold at 634 px, then nothing until 4,430 px — 6.7 screens of scrolling. There is a sticky CTA element in the markup, but it renders with class hidden-cta and zero height on mobile.
Export at 2× displayed width as WebP with a JPEG fallback, add loading="lazy" and explicit width/height. 15.1 MB → under 400 KB. No test needed — this is a defect fix, not a hypothesis.
Embed step 1 of the quote flow directly in the hero — same move that makes the short-term-rental page work. The visitor answers before they navigate, which both removes the hop and creates commitment. Fall back to fixing the sticky CTA (hidden-cta) if embedding isn't quick.
13.3 screens of page with two CTAs is under-served. Add one after the "Why Lifestyle Block Owners Choose Us" block and one after the FAQ, and un-hide the sticky mobile bar.
Measured on a real render — iPhone 13 viewport, 4 Mbps / 70 ms latency, 4× CPU throttle, which approximates a mid-range Android on NZ mobile data.
| Page | LCP | CLS | Full load | Requests | Weight | Hosts | Verdict |
|---|---|---|---|---|---|---|---|
/e/v/299 (via experiment) | 2.14 s | 0.000 | 6.4 s | 85 | 1.49 MB | 34 | Redirect costs 0.6 s |
…x5ulc (direct) | 1.46 s | 0.000 | 5.3 s | 82 | 1.51 MB | 34 | 1.1 MB of script |
…BRd9t5 | 1.55 s | 0.000 | 5.4 s | 81 | 1.33 MB | 34 | 1.1 MB of script |
…Q3bH1J Contract Works | 1.58 s | 0.001 | 5.6 s | 78 | 1.39 MB | 34 | 1.1 MB of script |
…PW2Fl7 (299 variant B) | 1.63 s | 0.022 | 5.4 s | 82 | 1.52 MB | 34 | 1.1 MB of script |
page.gerrards.co.nz/ Rural | 0.67 s | 0.008 | 31.7 s | 35 | 15.49 MB | 15 | 15 MB of images |
start.gerrards… Airbnb | 0.91 s | 0.232 | 0.5 s | 32 | 0.46 MB | 6 | CLS 0.232 = poor |
Google Ads, GA4 (two property IDs), GTM, Facebook, TikTok, Bing UET, Microsoft Clarity, LinkedIn, DoubleClick, jsDelivr. 1.1 MB of JavaScript — roughly three-quarters of total page weight, and more than the images.
The page declares GA4 G-VMS9KN9BEK in its own config while loading G-V25NBEFNZ6 — worth reconciling against the "only GTM-KNHW726H" rule.
The short-term-rental page scores CLS 0.232 — Google's "poor" band starts at 0.25 and "needs improvement" at 0.1. It loads 217 KB of fonts, more than any other resource type, and the text reflows as they swap in.
Fix: font-display:optional or preload the two faces actually used above the fold, and reserve height on the badge row.
/e/v/299 is 11,093 px — 16.7 screens. Below the form sits a review wall, an "Trusted Insurance Advisors for Rural New Zealand" block (on the page serving Commercial Property, CMV, Truck and Digger), a how-it-works section and a footer. That rural heading is a template leak and reads as wrong to every visitor on that page.
/e/v/299 → …x5ulc / …PW2Fl7 $36,798 · CvR 20.2%page.gerrards.co.nz/ Rural $7,905 · CvR 16.2% → 10.6%hidden-ctastart.gerrards…/f/short-term-rental-insurance/ $7,385 · CvR 9.6%Category eyebrow (AIRBNB · BOOKABACH · BOOKING.COM), specific H1, a value proposition that names the actual objection ("Most standard home policies exclude paying guests"), trust badges, an explicit time cost, "Step 1 of 8" progress, and a compact 2-up option grid that fits in 1.1 screens. It is 0.46 MB and loads in half a second.
Its problem is not the page — it's the traffic. Only 36% of spend lands on short-term-rental intent. The rest goes to "house insurance for seniors nz", "house insurance calculator", "initio insurance nz", "generate insurance", "home insurance comparison nz". On-target terms convert at 16.1%; off-target at 3.7%.
Fixing this page would be optimising the wrong thing. The lever here is negatives and match-type discipline, not CRO — with one exception: CLS 0.232 is worth fixing, and 8 steps is worth testing against a shorter path.
…Q3bH1J Contract Works $5,202 · CvR 12.5%Better than its siblings: the first question is specific ("What type of project are you looking to protect?"), it explains why it's asking ("Different builds carry very different risk profiles"), and it offers only two options, both above the fold. That explanatory subhead is a pattern worth porting to every other question in the account. It still has no H1 and no progress indicator.
General Business splits across …x5ulc (20.0%) and …BRd9t5 (8.3%). That looks like a decisive page test, but it isn't:
The two pages are assigned by ad group, not randomly. "Business Insurance Core" → x5ulc at 21.4%; "Broker Services" → BRd9t5 at 5.6%. The one ad group that ran both — COMMERCIAL COVERAGE — actually reverses the result (8.9% on x5ulc vs 22.2% on BRd9t5, on only 57 and 18 clicks). So the gap is traffic, not page.
The real finding underneath is an ad-level mismatch: the Broker Services ad group ($1,740, 237 clicks, 5.6% CvR) runs the headline "Helpful Insurance Guidance | Personal Insurance Broker" and lands people on a business insurance application form. That is a promise/page mismatch worth fixing at the ad group, and it is the weakest-converting pocket of spend in the account.
The brief sets a target of "at least 4%" per page. That reads two ways, and the difference is enormous:
e.g. 16.2% → 16.8%. Well within reach of the changes above, on most of these pages.
e.g. 16.2% → 20.2%, a 25% relative gain. Achievable on the mismatched pages; a stretch on CMV, which already converts at 27.9%.
I'd rather state this plainly than imply a guarantee: the changes above are well-founded, but no one can promise a specific lift in advance. What I can do is show what it would take to prove one.
To detect a +4 point lift at 95% confidence and 80% power, you need roughly 900–1,700 clicks per arm. Against current traffic:
| Page | Baseline | Clicks/wk | n per arm | Weeks — 50/50 test | Weeks — ship to 100% |
|---|---|---|---|---|---|
/e/v/299 pooled | 20.2% | 170 | 1,694 | 20 | 10 |
| Commercial Property alone | 16.3% | 52 | 1,466 | 56 | 28 |
| Rural | 16.2% | 60 | 1,460 | 49 | 24 |
| Contract Works | 12.5% | 41 | 1,216 | 59 | 29 |
| Airbnb | 9.6% | 38 | 1,005 | 53 | 27 |
| Truck alone | 15.0% | 39 | 1,384 | 71 | 36 |
| Digger alone | 17.7% | 19 | 1,551 | 162 | 81 |
Classic 50/50 A/B testing does not work at this account's volume. Every page needs 5 months to 3 years to resolve a +4-point test. Anyone reporting a "winner" on these pages after two weeks is reading noise.
So the method has to change:
/e/v/299 campaigns share a page, so test the page across the pooled 170 clicks/week — 10 weeks, not 162.Rules that hold for every Gerrard's landing page, derived from what's actually working in this account.
/e/v/ experiment hop costs 0.6 s.font-display:optional.| # | Action | Page | Effort | Confidence | Test? |
|---|---|---|---|---|---|
| 1 | Compress + lazy-load the two Rural images (15.1 MB → <400 KB) | Rural | Trivial | Certain defect | Ship |
| 2 | Add a machinery/excavator option — or a Digger-specific entry | /e/v/299 | Low | Certain defect | Ship |
| 3 | Fix "Trusted Advisors for Rural NZ" on commercial pages | /e/v/299 | Trivial | Certain defect | Ship |
| 4 | Add H1 + value prop + trust row + progress bar | All quote pages | Low | High | Ship, watch step-1 |
| 5 | Pre-answer Q1 per vertical; compact the option grid | /e/v/299 | Medium | High | Pool 10 wks |
| 6 | Repoint the Broker Services ad group, or rewrite its ads | Gen. Business | Low | Medium | Before/after |
| 7 | Fix CLS 0.232 (font loading) | Airbnb | Low | Medium | Ship |
| 8 | Embed step 1 in the Rural hero; remove the hop | Rural | Medium | Medium | Before/after |
| 9 | Negatives sweep — Airbnb is 64% off-target spend | Airbnb | Low | High | Not CRO |
…BRd9t5 on that basis.
Sources — clients/gerrards-insurance/data/{summary,pages,diag_ads,search80,campSettings}.csv, exported 12 Aug 2026 05:00 NZ via gads-proxy, covering 24 Jan – 11 Aug 2026 (ad-level detail: 11 May – 3 Aug 2026).
Page metrics captured 12 Aug 2026 by headless Chromium at iPhone 13 viewport, 4 Mbps / 70 ms / 4× CPU.
Significance tests are two-proportion z-tests; power calculations assume 80% power, 95% two-sided confidence.