# Google Ads Account Review: 2020-2026
A data-driven review of the Hairpin Leg Co Google Ads account across six years, two agencies, one conversion tracking breakdown, and a global pandemic.
**Data source**: `google_ads_raw_daily` — 107k rows of daily account/campaign/ad-group metrics, 2020-01-01 to 2026-02-12. All spend figures in GBP.
---
## Executive Summary
- **Total spend 2020-2026**: ~£829k across ~78 campaigns
- **Peak year**: 2020 (£191k spend, £1.25M conversion value, 6.57x ROAS) — COVID-era home improvement boom
- **Steady decline 2021-2022**: Spend halved from £191k to £85k as the market normalised; ROAS held at 7x
- **Conversion tracking broke in stages**: Gradual degradation started Feb 2023 (conversions halved), then massively over-counted May-Jul 2023 (10-40x inflation). The months of bad signal poisoned all PMax campaign models beyond recovery
- **The disaster had a long tail**: 3-4 months of near-zero productive spend (Feb-May), plus permanent destruction of AP's PMax campaign learning data (£35k of cumulative investment in algorithmic models) — all abandoned in the transition
- **Agency handover 18 Aug 2023**: Clean same-day cutover from Adpulse (AP) to Snowball Creations (SC)
- **SC rebuilt the account around Shopping**: 58% of SC's total spend is Shopping (vs. 3% under AP). Search dropped from AP's primary channel to a supporting role
- **SC's search campaigns fail because CPC is 4x higher — but ~40% of the gap is fixable**: The #1 search term "hairpin legs" cost AP £0.51/click and SC £1.80/click. Quality scores (9) and CTR (23.6%) are fine — the gap is not a relevance problem. Market inflation accounts for ~2-2.5x (Shopping CPC doubled over the same period). The remaining ~1.5x was caused by Maximise Conversions ratcheting: both SC search campaigns experienced a specific trigger event (Feb 2024: a Shopping budget surge that shifted conversion attribution; Feb 2025: Shopify checkout broke tracking), after which Maximise Conversions permanently escalated bids. A Manual CPC approach could realistically halve the bid-strategy component
- **Brand Search is leaking**: 7.2% of Brand Search spend goes to non-brand terms at £4.31 CPC and 0.85x ROAS due to broad match types — an immediate optimisation opportunity
- **Current ROAS (SC era) is ~3.3x**, roughly half of AP's 7x — but AP's numbers include the COVID boom years when every DTC brand was printing money
- **Current run rate**: ~£140k/yr spend, ~£485k/yr conversion value. Six active campaigns, all Shopping/PMax except Brand Search
---
## Account Timeline
```
2020 2021 2022 2023 2024 2025 2026
| | | | | | |
|==COVID BOOM==| | | | | |
| £191k spend | | | | | |
| 6.6x ROAS | | | | | |
| |========| | | | |
| |POST-COVID DECLINE | | | |
| | Spend halving | | | |
| | ROAS holds at 7x | | | |
| | |===========| | | |
| | | |TRACKING BROKE| | |
| | | |May-Jul '23 | | |
| | | | | | |
| | | | AP→SC | | |
| | | | 18 Aug '23 | | |
| | | | |==========|===========|
| | | | |SC STEADY STATE |
| | | | | Shopping-led |
| | | | | ~3.3x ROAS |
```
---
## Annual Performance
| Year | Spend | Clicks | Conversions | Conv. Value | ROAS | CPC | CTR | CVR |
|------|------:|-------:|------------:|------------:|-----:|----:|----:|----:|
| 2020 | £191,034 | 441,298 | 13,061 | £1,254,360 | 6.57x | £0.43 | 1.39% | 2.96% |
| 2021 | £173,215 | 377,928 | 11,928 | £1,224,804 | 7.07x | £0.46 | 1.31% | 3.16% |
| 2022 | £84,680 | 166,497 | 5,717 | £607,233 | 7.17x | £0.51 | 1.34% | 3.43% |
| 2023 | £97,702 | 117,969 | 21,225* | £2,301,186* | 23.55x* | £0.83 | 1.37% | 17.99%* |
| 2024 | £125,846 | 92,749 | 3,873 | £412,250 | 3.28x | £1.36 | 0.95% | 4.18% |
| 2025 | £142,739 | 120,676 | 4,897 | £485,532 | 3.40x | £1.18 | 1.16% | 4.06% |
| 2026 (to 12 Feb) | £14,235 | 15,412 | 503 | £46,773 | 3.29x | £0.92 | 1.26% | 3.26% |
*\* 2023 conversion/value figures are corrupted by the May-July tracking breakdown. Spend, clicks and impressions are reliable.*
---
## Quarterly Performance
| Quarter | Spend | Clicks | Impr. | Conv. | Conv. Value | ROAS | CPC | CTR | CVR |
|---------|------:|-------:|------:|------:|------------:|-----:|----:|----:|----:|
| 2020-Q1 | £27,995 | 59,001 | 4.6M | 1,649 | £146,292 | 5.23x | £0.47 | 1.28% | 2.80% |
| 2020-Q2 | £41,622 | 111,627 | 8.2M | 3,227 | £302,973 | 7.28x | £0.37 | 1.36% | 2.89% |
| 2020-Q3 | £53,583 | 121,478 | 8.5M | 3,463 | £353,971 | 6.61x | £0.44 | 1.44% | 2.85% |
| 2020-Q4 | £67,834 | 149,192 | 10.6M | 4,722 | £451,124 | 6.65x | £0.45 | 1.41% | 3.16% |
| 2021-Q1 | £63,894 | 159,653 | 11.0M | 4,725 | £464,486 | 7.27x | £0.40 | 1.45% | 2.96% |
| 2021-Q2 | £39,860 | 90,112 | 6.8M | 2,917 | £303,282 | 7.61x | £0.44 | 1.32% | 3.24% |
| 2021-Q3 | £35,422 | 67,989 | 6.4M | 2,234 | £241,695 | 6.82x | £0.52 | 1.07% | 3.29% |
| 2021-Q4 | £34,039 | 60,174 | 4.6M | 2,053 | £215,342 | 6.33x | £0.57 | 1.31% | 3.41% |
| 2022-Q1 | £25,515 | 53,977 | 4.1M | 1,725 | £186,492 | 7.31x | £0.47 | 1.31% | 3.20% |
| 2022-Q2 | £22,137 | 44,501 | 3.4M | 1,407 | £147,899 | 6.68x | £0.50 | 1.33% | 3.16% |
| 2022-Q3 | £18,679 | 35,458 | 2.6M | 1,268 | £140,569 | 7.53x | £0.53 | 1.38% | 3.58% |
| 2022-Q4 | £18,350 | 32,561 | 2.4M | 1,317 | £132,271 | 7.21x | £0.56 | 1.38% | 4.04% |
| 2023-Q1 | £16,259 | 30,333 | 2.1M | 766 | £78,739 | 4.84x | £0.54 | 1.45% | 2.52% |
| **2023-Q2** | **£9,745** | **17,678** | **1.7M** | **7,224*** | **£789,344*** | **81x*** | **£0.55** | **1.06%** | **40.87%*** |
| **2023-Q3** | **£22,440** | **38,745** | **2.4M** | **11,653*** | **£1,286,619*** | **57x*** | **£0.58** | **1.61%** | **30.08%*** |
| 2023-Q4 | £49,257 | 31,213 | 2.5M | 1,583 | £146,484 | 2.97x | £1.58 | 1.27% | 5.07% |
| 2024-Q1 | £40,995 | 32,514 | 3.2M | 1,123 | £107,839 | 2.63x | £1.26 | 1.03% | 3.45% |
| 2024-Q2 | £28,251 | 23,344 | 2.9M | 950 | £107,844 | 3.82x | £1.21 | 0.81% | 4.07% |
| 2024-Q3 | £23,100 | 16,548 | 2.0M | 817 | £87,278 | 3.78x | £1.40 | 0.83% | 4.94% |
| 2024-Q4 | £33,500 | 20,343 | 1.8M | 984 | £109,289 | 3.26x | £1.65 | 1.16% | 4.83% |
| 2025-Q1 | £41,570 | 28,473 | 2.6M | 1,156 | £117,174 | 2.82x | £1.46 | 1.09% | 4.06% |
| 2025-Q2 | £45,884 | 30,280 | 2.5M | 1,278 | £135,864 | 2.96x | £1.52 | 1.23% | 4.22% |
| 2025-Q3 | £28,552 | 25,455 | 2.2M | 1,268 | £124,159 | 4.35x | £1.12 | 1.17% | 4.98% |
| 2025-Q4 | £26,733 | 36,468 | 3.1M | 1,196 | £108,336 | 4.05x | £0.73 | 1.16% | 3.28% |
| 2026-Q1 (partial) | £14,235 | 15,412 | 1.2M | 499 | £46,697 | 3.28x | £0.92 | 1.26% | 3.24% |
*\* Bolded rows contain corrupted conversion data. Spend/clicks/impressions remain reliable.*
---
## Phase Analysis
### Phase 1: COVID Boom (Q1 2020 - Q2 2021)
The pandemic home improvement surge drove the account's highest-ever performance.
| Metric | Value |
|--------|------:|
| Total spend | £294,790 |
| Total conversion value | £1,764,176 |
| Blended ROAS | 5.98x |
| Avg. monthly spend | £19,653 |
| Avg. CPC | £0.42 |
| Avg. CVR | 2.97% |
| Peak month | Nov 2020: £27,430 spend, 7.30x ROAS |
**What happened**: Stay-at-home orders drove a surge in home office and furniture demand. Cheap search traffic (£0.37 CPC at the low point in May 2020) combined with massive impression volumes meant the account could scale aggressively while maintaining 6-8x ROAS.
**Campaign structure**: Dominated by AP text search campaigns — `AP | Text | Hairpin Table Legs` and `AP | Text | Wicked Hairpins` were the core workhorses, supplemented by brand, competitor, and product-category text ads. Shopping was minimal (£1.2k total during this period).
**The peak and inflection point**: Q4 2020 hit £67,834 spend — the single biggest quarter. But Q1 2021, while still strong (£63,894 spend, 7.27x ROAS), was the last quarter of boom-level activity. By Q2 2021, spend had dropped 38% QoQ as the post-lockdown normalisation began.
### Phase 2: Post-COVID Normalisation (Q3 2021 - Q1 2023)
A long, steady decline in volume while ROAS held remarkably stable.
| Metric | Value |
|--------|------:|
| Total spend | £154,721 |
| Total conversion value | £1,123,301 |
| Blended ROAS | 7.26x |
| Avg. monthly spend | £8,596 |
| Avg. CPC | £0.52 |
| Avg. CVR | 3.38% |
**What happened**: The post-pandemic demand correction played out over 18 months. Spend dropped from £35k/quarter to £16k/quarter, but this was primarily budget-driven — CPCs crept up only modestly (£0.52 to £0.56) and ROAS actually held steady or improved (7.5x in Q3 2022).
**The signal worth noting**: CVR climbed from 3.06% to 4.04% through this period. The account was getting more efficient even as it shrank — likely AP tightening keyword targeting as budgets decreased, and lower-intent browsers dropping out of the market post-COVID.
**AP introduced PMax in mid-2022**: `AP | Performance Max | Hairpin Table Legs`, `AP | Performance Max | Industrial Table Legs`, etc. launched in July 2022. By Q4 2022, PMax was running alongside text search and a small amount of Standard Shopping.
**Q1 2023 was the canary**: ROAS dropped to 4.84x — the lowest in 3 years. Conversions fell disproportionately to the spend reduction (766 conversions on £16,259 spend vs. 1,317 on £18,350 the prior quarter). Something was going wrong with conversion tracking.
### Phase 3: Conversion Tracking Breakdown & AP Wind-Down (Q2 2023 - Aug 2023)
**Conversion data from May-July 2023 is completely unreliable and should be discarded for any analysis.**
The numbers tell the story of the corruption:
| Month | Spend | Clicks | Conversions (reported) | CVR (reported) | ROAS (reported) |
|-------|------:|-------:|-----------------------:|---------------:|----------------:|
| Apr 2023 | £1,506 | 4,068 | 98 | 2.42% | 6.82x |
| **May 2023** | **£2,120** | **3,935** | **1,194** | **30.33%** | **61.97x** |
| **Jun 2023** | **£6,119** | **9,675** | **5,932** | **61.31%** | **105.84x** |
| **Jul 2023** | **£6,564** | **19,671** | **10,970** | **55.77%** | **184.87x** |
| Aug 2023 | £7,416 | 10,118 | 340 | 3.36% | 4.70x |
The corruption hit every AP campaign — PMax, Text Search, Shopping, DSA — suggesting the issue was at the conversion action/tag level, not campaign-specific. Some individual campaigns reported absurd numbers: `AP | Performance Max | Desks` showed 1,693 conversions in July on 2,548 clicks (66% CVR).
**The non-AP campaign that appeared**: `Hairpin Table Legs - Google Business Manager` ran from 5-18 May 2023 with £106 spend. This is the only non-AP/non-SC campaign in the account history — likely a manual experiment or GMC auto-campaign.
**AP's last spend**: 18 August 2023 across `AP | Text | Hairpin Table Legs`, `AP | Text | Brand`, and `AP | Text | SKAG | Top Converting`. The same day, SC's first three campaigns launched.
### Phase 4: SC Onboarding & Restructure (Aug 2023 - Q1 2024)
SC launched with a fundamentally different strategy from AP.
**Day 1 campaigns (18 Aug 2023)**:
1. `(OLD) General Search Top Performers - SC` — Search (paused Mar 2024)
2. `Brand Search - SC` — Brand search (still active)
3. `Shopping Top Performers - SC` — Shopping (still active)
**Q4 2023 ramp-up**: SC's first full quarter saw £49,257 spend — almost triple the preceding quarter. CPC jumped to £1.58 as the Shopping campaigns brought more expensive, higher-intent traffic.
| SC Quarter | Spend | Clicks | ROAS | CPC | CVR |
|------------|------:|-------:|-----:|----:|----:|
| 2023-Q3 (partial) | £12,123 | 14,041 | 4.32x | £0.86 | 3.53% |
| 2023-Q4 | £49,257 | 31,213 | 2.97x | £1.58 | 5.07% |
| 2024-Q1 | £40,995 | 32,514 | 2.63x | £1.26 | 3.45% |
**The low-ROAS onboarding dip**: Q4 2023 through Q1 2024 was the account's worst ROAS period (2.63-2.97x). This is typical of agency transitions — new campaigns need data to optimise, and SC was simultaneously learning the account and building a Shopping-centric structure that AP never had.
**January 2024**: `(OLD) General Search Testing - SC` launched — an experimental search campaign that ran until March 2024 with £5,768 spend and poor 0.54x ROAS. It was correctly killed.
### Phase 5: Steady State Under SC (Q2 2024 - Present)
The account found its footing through mid-2024 and has been running in a recognisable pattern since.
| Period | Quarterly Spend (avg) | ROAS (avg) | CPC (avg) | CVR (avg) |
|--------|----------------------:|------------|-----------|-----------|
| Q2-Q4 2024 | £28,284 | 3.62x | £1.42 | 4.61% |
| Q1-Q4 2025 | £35,685 | 3.40x | £1.32 | 4.33% |
| Q1 2026 (partial) | £14,235* | 3.28x | £0.92 | 3.24% |
*\* Q1 2026 is only 43 days of data.*
**Campaign evolution under SC**:
| Campaign | Launched | Status | Lifetime Spend | Lifetime ROAS |
|----------|----------|--------|---------------:|--------------:|
| Shopping Top Performers - SC | Aug 2023 | Active | £112,036 | 3.00x |
| Brand Search - SC | Aug 2023 | Active | £55,234 | 6.48x |
| (OLD) General Search Top Performers - SC | Aug 2023 | Paused | £39,345 | 1.74x |
| General Search (SC) | Oct 2024 | Paused | £39,290 | 2.01x |
| Shopping Testing - SC | Feb 2024 | Paused | £30,341 | 1.36x |
| Shopping Catch All - Over £20 - SC | Jan 2025 | Active | £29,453 | 4.54x |
| Brand Shopping - SC | Dec 2024 | Active | £26,000 | 4.00x |
| (OLD) General Search Testing - SC | Jan 2024 | Paused | £5,768 | 0.54x |
| PMax: Magnetic Knife Rack Only Shopping (SC) | Nov 2025 | Paused | £3,111 | 2.71x |
| PMax: Knobs Only Shopping (SC) | Oct 2025 | Paused | £2,461 | 2.87x |
| PMax: Table Tops (SC) | Jan 2026 | Active | £723 | 4.39x |
| Shopping Catch All - Under £20 - SC | Jan 2026 | Active | £334 | 3.16x |
| Knobs Only Shopping (SC) | Oct 2025 | Paused | £104 | 0.58x |
**Notable patterns**:
- SC has tried generic search twice (`(OLD) General Search Top Performers`, `General Search (SC)`) — both paused with ROAS under 2x. The account currently has **no non-brand search campaigns running**
- Shopping Testing ran for 17 months before being paused at 1.36x ROAS — a long testing window
- Recent PMax experiments (knobs, knife racks, table tops) show 2.7-4.4x ROAS in limited data. `PMax: Table Tops` is the newest active campaign (Jan 2026)
- The Shopping Catch All campaigns (split by £20 price threshold in Jan 2026) suggest SC is segmenting Shopping by product value
---
## Agency Comparison
### Headline Numbers
| Metric | AP (excl. May-Jul 2023) | SC (Aug 2023 - Feb 2026) |
|--------|------------------------:|-------------------------:|
| Total spend | £245,452 | £344,200 |
| Total clicks | 550,248 | 274,091 |
| Conversions | 17,529 | 11,352 |
| Conversion value | £1,725,990 | £1,143,445 |
| ROAS | 7.03x | 3.32x |
| CPC | £0.45 | £1.26 |
| CTR | 4.16% | 1.09% |
| CVR | 3.19% | 4.14% |
### Why the Comparison Is Misleading
A direct AP vs SC comparison overstates AP's advantage for several reasons:
1. **AP's data includes the COVID boom (2020-2021)** — a once-in-a-generation demand event. Every DTC advertiser saw inflated performance during this period. Stripping out the boom years, AP's post-COVID performance (Feb 2021 - Apr 2023) shows £138,630 spend at 7.24x ROAS — still better than SC, but the gap narrows.
2. **AP was predominantly Search; SC is predominantly Shopping**. Different channel, different economics:
- AP: Started at 98% Search (2020), shifted to ~42% Search / 42% PMax by 2023
- SC: 33-59% Search, 41-67% Shopping, with PMax growing to 15% in 2026
Shopping typically carries higher CPCs and lower headline ROAS but captures more high-intent, bottom-funnel traffic.
1. **The market changed**. Google Ads CPC inflation across e-commerce has been well documented — the same traffic costs more in 2025 than it did in 2021 regardless of agency.
2. **SC's CVR is higher** (4.14% vs 3.19%). The clicks SC is buying convert at a better rate, which means the per-conversion cost gap is smaller than the CPC gap suggests.
### Fairer Comparison: Cost Per Conversion
| Agency | CPC | CVR | Cost per Conversion |
|--------|----:|----:|--------------------:|
| AP (excl. corrupt data) | £0.45 | 3.19% | £14.00 |
| SC | £1.26 | 4.14% | £30.32 |
SC's cost per conversion is roughly double AP's. Some of this is market inflation, some is the Shopping-heavy strategy, and some may be genuine performance difference. Without controlling for product mix, seasonality, and the macro CPC trend, it's impossible to cleanly attribute.
### Campaign Mix Over Time
| Period | Search | Shopping | PMax | Display/Other |
|--------|-------:|--------:|-----:|--------------:|
| 2020 (AP) | 98% | 1% | — | ~1% |
| 2021 (AP) | 95% | 3% | — | 1% |
| 2022 (AP) | 59% | 4% | 34% | 2% |
| 2023 (AP, to Aug) | 42% | 4% | 42% | 1% |
| 2023 (SC, from Aug) | 59% | 41% | — | — |
| 2024 (SC) | 33% | 67% | — | — |
| 2025 (SC) | 40% | 57% | 3% | — |
| 2026 YTD (SC) | 29% | 56% | 15% | — |
AP ran the account as a predominantly text search operation for its entire tenure — even after introducing PMax in mid-2022, search remained the largest channel. SC flipped this immediately, making Shopping the primary channel within its first full quarter.
---
## Current State Assessment (Last 90 Days)
Six active campaigns as of February 2026:
| Campaign | Spend (90d) | Clicks | Conv. | ROAS | CPC |
|----------|------------:|-------:|------:|-----:|----:|
| Shopping Catch All - Over £20 - SC | £8,565 | 10,920 | 403 | 4.27x | £0.78 |
| Brand Search - SC | £6,488 | 3,290 | 237 | 3.51x | £1.97 |
| Shopping Top Performers - SC | £5,510 | 3,860 | 191 | 3.35x | £1.43 |
| Brand Shopping - SC | £3,215 | 3,692 | 154 | 4.07x | £0.87 |
| PMax: Table Tops (SC) | £723 | 858 | 22 | 4.39x | £0.84 |
| Shopping Catch All - Under £20 - SC | £334 | 534 | 27 | 3.16x | £0.63 |
**Combined 90-day performance**: £24,835 spend, 23,154 clicks, 1,034 conversions, £95,125 conversion value, **3.83x ROAS**, £1.07 avg CPC, 4.47% CVR.
**Observations on the current setup**:
- **No non-brand search running**: Both generic search campaigns have been paused. The account is entirely brand search + Shopping/PMax. This means the account captures demand but does very little to generate it through search.
- **Shopping Catch All is the biggest spender**: The £20 split is recent (Jan 2026). The over-£20 segment at 4.27x ROAS is outperforming Shopping Top Performers at 3.35x. Worth watching whether the "catch all" is genuinely better, or whether it's benefiting from product mix.
- **Brand Search CPC at £1.97**: Expensive for branded terms. Could indicate competitor bidding on the Hairpin Leg Co brand, or broad match creep pulling in non-brand queries.
- **PMax: Table Tops is new and promising**: 4.39x ROAS on £723 spend. Early data, but table tops are a new product category worth growing. Previous PMax experiments (knobs, knife racks) were paused at 2.7-2.9x ROAS — below the threshold to justify continued spend.
---
## Deep Dive: The Tracking Disaster and Its Aftermath
### The Timeline of Collapse
The conversion tracking breakdown wasn't a sudden event — it was a slow-motion crash that started months before the obvious corruption in May 2023.
**October 2022 - January 2023 (the baseline)**:
The account was running at a stable ~£6k/month, ~12k clicks, ~4% CVR, ~7x ROAS. This was the "normal" post-COVID state.
**February 2023 — the first warning sign**:
CVR dropped from 3.29% (Jan) to 2.40% (Feb). Conversions fell from 426 to 234 on only modestly lower click volume (12,966 → 9,762). ROAS dropped from 7.23x to 4.31x. This hit **all campaign types simultaneously**:
| Campaign Type | Q4 2022 Conv. | Jan 2023 Conv. | Feb 2023 Conv. | Mar 2023 Conv. |
|---------------|----------:|----------:|----------:|----------:|
| PMax | 736 | 238 | 132 | 31 |
| Search | 571 | 185 | 101 | 75 |
| Shopping | 10 | 3 | 1 | 0 |
PMax was hit hardest — from 238 conversions in January to 31 in March, an 87% collapse. Search also declined but held up better (185 → 75). The fact that it hit PMax disproportionately is significant: PMax campaigns use Smart Bidding that directly depends on conversion signal quality. If the conversion tag started misfiring (either under-counting or flapping), PMax's algorithm would respond by pulling back spend because it couldn't find "converting" traffic anymore.
**March 2023 — spend collapses**:
Monthly spend dropped from £5,147 (Feb) to £4,616 (Mar), but impressions cratered from 723k to 298k. The PMax campaigns effectively stopped spending: the five PMax campaigns went from £2,196 in February to £1,818 in March, with Wall Hooks and Industrial Table Legs dropping to near-zero. This is what PMax does when it loses confidence in the conversion signal — it gradually goes dormant.
**April 2023 — the near-shutdown**:
Total spend collapsed to £1,506 for the entire month. Only 4,068 clicks. But then it gets weird: the daily data shows spend at £100+/day through April 21, then a complete gap — **zero spend from April 22-24**, and near-zero (£8.91) on April 25. The account effectively went dark.
**May 1-17, 2023 — zombie state**:
The account limped along at £7-20/day with no conversions recorded. The original campaigns were barely spending. Then on **May 18**, two things happened simultaneously:
1. AP launched six "NEW" campaigns (`AP | NEW | Hairpin Table Legs`, `AP | NEW | Rest Of Products`, etc.) — a clear rebuild attempt
2. The `AP | Test | Text 18/05/2023` campaign launched — explicitly a test
**May 19 onwards — the explosion**:
The day after the NEW campaigns launched, fake conversions began. May 19 alone reported 930 conversions on 206 clicks (451% CVR, an obvious impossibility). From here, the corruption accelerated through June and peaked in July at 10,970 fake conversions for the month.
### Were the Campaigns Recoverable?
Almost certainly not. Here's why:
**PMax campaigns were poisoned by algorithmic learning**. Performance Max campaigns build audience and bidding models based on conversion history. During May-July 2023, these campaigns were fed thousands of false conversion signals. The algorithm "learned" that virtually every impression led to a conversion, which would have:
- Destroyed the bidding model (bidding too aggressively on junk traffic)
- Corrupted the audience signals (every audience segment appears to convert)
- Invalidated the creative performance data (every asset appears to work)
Even after the tracking was fixed, a PMax campaign with 3 months of poisoned training data can't simply recover. Google's Smart Bidding models have a lookback window that would continue to reference the corrupted period for months afterward.
**The AP "NEW" campaigns were born into the corruption**. The six campaigns launched on May 18 accumulated their entire conversion histories during the corrupted period. `AP | NEW | Rest Of Products` showed 2,185 conversions on just £1,352 spend — numbers that were obviously fake. These campaigns never had a clean baseline to return to.
**Evidence from the post-corruption period**: In August 2023 (after the corruption stopped), the remaining AP campaigns show marginal performance:
| Campaign | Jan 2023 (pre) | Aug 1-17 2023 (post) | Notes |
|----------|---:|---:|---|
| AP \| PMax \| Industrial Table Legs | 52 conv, £1,312 | 13 conv, £423 | 75% drop in efficiency |
| AP \| PMax \| Hairpin Table Legs | 86 conv, £1,073 | 36 conv, £516 | Still running but damaged |
| AP \| PMax \| Wall Hooks | 83 conv, £925 | 14 conv, £251 | Barely alive |
| AP \| Text \| Hairpin Table Legs | 58 conv, £1,101 | 8 conv, £517 | CPC doubled (£0.81 → £1.68) |
| AP \| Text \| Brand | 85 conv, £537 | 55 conv, £681 | Relatively resilient |
| AP \| Text \| SKAG | 26 conv, £268 | 39 conv, £821 | Higher spend, lower efficiency |
Brand search was the only campaign that looked roughly normal in August — because brand campaigns are less dependent on algorithmic bidding signals. Everything else was either barely spending or spending inefficiently. The PMax campaigns were running on poisoned models. The text search campaigns had doubled their CPCs (presumably because the bid strategy was still recalibrating).
### What Actually Happened — A Reconstruction
Piecing together the timeline:
1. **Late Jan/early Feb 2023**: Conversion tracking starts degrading. Either the tag stops firing reliably, a conversion action gets misconfigured, or a site change breaks the tracking snippet. PMax immediately responds by reducing spend.
2. **Feb-Mar 2023**: A cascading failure. PMax can't find converting traffic (because it can't measure conversions), so it stops spending. Account revenue drops. This looks like poor campaign performance, but it was actually a measurement failure.
3. **Late Mar-Apr 2023**: AP (or the client) dramatically cuts budgets in response to the apparent performance decline. Spend drops to £1,506/month. A 10-day complete pause happens Apr 22 - May 1.
4. **May 18 2023**: AP launches "NEW" campaigns — a rebuild attempt. They also create a test campaign. Whatever was wrong with conversion tracking has either been "fixed" in a way that overcorrects (now tracking phantom conversions) or a completely different measurement problem begins.
5. **May 19 - Jul 25 2023**: Massively inflated conversions across all campaigns. The PMax campaigns, hungry for conversion signals after months of starvation, gorge on the false data and scale up aggressively.
6. **Jul 26 2023**: The corruption abruptly reduces (from hundreds/day to tens/day). Something was fixed or changed.
7. **Aug 18 2023**: SC takes over. AP campaigns are paused. SC starts fresh with new campaigns and (presumably) properly configured conversion tracking.
### The Real Cost
The tracking disaster cost more than just the spend during May-July 2023 (~£18k of spend on corrupted data). The real costs were:
- **3-4 months of near-zero productive spend** (Feb-May 2023): The account was effectively dormant while the tracking issue went undiagnosed, costing perhaps £15-20k in missed revenue at historical run rates
- **Permanently damaged PMax campaigns**: 5 PMax campaigns with £35k in historical spend had their algorithmic models poisoned beyond recovery
- **Loss of years of campaign learning data**: AP's text search campaigns had 3 years of quality score history, keyword performance data, and audience signals that were abandoned in the transition
- **The SC cold-start penalty**: SC had to build everything from scratch with zero historical conversion data to leverage
---
## Deep Dive: Why SC's Search Campaigns Fail
SC has spent £84,403 on three generic search campaigns over 2.5 years, all paused with ROAS under 2x. AP ran profitable search campaigns (7-8x ROAS) for the preceding 3 years. The same search terms, the same products, the same landing pages.
**Full analysis**: [Search Campaign CPC Analysis](search-campaign-cpc-analysis.md)
### Key Findings
The 4x CPC gap between AP search (£0.51/click) and SC search (£1.82/click) on identical terms like "hairpin legs" is **not** caused by quality score, account history, ad copy, or landing pages. SC's quality scores are 9, CTR is comparable or better, CVR is consistently higher, and CPCs were near-identical at the transition point (AP £0.77 → SC £0.95).
The gap opened *after* the transition and decomposes into two factors:
| Factor | Magnitude | Fixable? |
|--------|-----------|----------|
| Quality Score / account history | **None** — QS is 9, CTR comparable, CPCs matched at transition | N/A |
| Ad copy / landing page quality | **None** — CVR is better under SC | N/A |
| Market-wide CPC inflation | **~2-2.5x** (evidenced by Shopping CPC trend) | No |
| Maximise Conversions amplification | **~1.5x on top** (two documented ratcheting events) | **Yes** |
Both SC generic search campaigns used Maximise Conversions and both experienced the same pattern: stable CPCs for 3-4 months, a specific trigger event, CPC doubling, and failure to revert:
- **Feb 2024**: A Shopping budget surge (£200/day → £734/day for 9 days) shifted conversion attribution away from search via last-click. Maximise Conversions interpreted the drop as "bids too low" and permanently ratcheted CPCs from £0.90 to £2.50+. Campaign paused at £2.82/click.
- **Feb 2025**: Shopify's checkout broke conversion tracking, feeding bad data to Maximise Conversions. Same outcome — CPC doubled and never reverted. Campaign paused at 2.01x blended ROAS.
### Recommendation
A Manual CPC approach at £1.20-1.50 caps could halve the bid-strategy component. At £1.30/click with SC's demonstrated 5.2% CVR and ~£60 AOV, ROAS would be ~2.4x — marginal but potentially justified if search captures incremental demand that Shopping doesn't. Any test should use exact match on the top 5-10 terms with hard CPC caps; no Maximise Conversions until 30+ conversions/month.
### Brand Search: Match Type Problem
Brand Search is leaking non-brand spend: 7.2% of 2025 spend (£1,837) went to non-brand terms at £4.31 CPC and 0.85x ROAS. Tightening to phrase/exact match on core brand terms is an immediate optimisation opportunity.
---
## Key Questions & Observations
### Questions for Investigation
1. **What caused the conversion tracking breakdown?** The reconstruction above shows it started as a gradual degradation in Feb 2023 (under-counting), then flipped to massive over-counting in May 2023. Was this ever formally root-caused? Was it a tag misconfiguration, a GTM change, or a Shopify checkout update that broke the pixel?
2. **Is the Shopping-first strategy truly optimal, or is it optimal-by-default?** SC hasn't been able to make search work, so Shopping wins by elimination. But Shopping is also cannibalising the same search terms at different CPCs. An incrementality test (pausing Shopping for a specific product category while running search) would answer whether the two channels are truly additive.
3. **Would Manual CPC on generic search actually work?** The data suggests Maximise Conversions is responsible for ~1.5x of the CPC gap. Brand Search on Manual CPC maintained £0.40-0.85 CPCs for 22 months. But Brand Search has high intent and high CTR — generic search may not behave the same way on Manual CPC. A controlled test (exact match, top 5 terms, Manual CPC with £1.30 cap, 3-month duration) would answer this definitively.
4. **Brand Search match types need tightening** (now confirmed): 7.2% of 2025 Brand Search spend went to non-brand terms at £4.31 CPC and 0.85x ROAS. This is an immediate optimisation opportunity that doesn't require any strategic changes — just tighter keyword matching.
### Patterns Worth Noting
- **Seasonality**: The account consistently shows stronger Q4 (Black Friday/Christmas) and Q1 (new year home projects) performance. Q3 is typically the weakest quarter.
- **CPC trend**: £0.37 (May 2020) → £0.56 (Q4 2022) → £1.58 (Q4 2023) → £1.07 (current 90-day). The post-transition CPC spike has moderated but remains 2-3x the AP era.
- **CVR trend**: Steadily improving from ~2.9% (2020) → 3.4% (2022) → 4.1% (2024-present). This is a positive signal regardless of agency — the landing page/product-market fit is getting better.
- **Impression volume**: The account served 31.9M impressions in 2020 vs 10.4M in 2025. The audience has shrunk or the targeting has tightened considerably. Under SC, impression share data would help determine if this is by choice or by budget constraint.
### The Uncomfortable Question
The account spent £191k in 2020 at 6.6x ROAS and £143k in 2025 at 3.4x ROAS. Conversion value dropped from £1.25M to £486k. Is this:
(a) A market that has contracted post-COVID and can't support the old spend levels?
(b) A less effective agency strategy that's leaving money on the table?
(c) A structural shift (Shopping vs Search) that looks worse on ROAS but is actually more sustainable?
(d) CPC inflation making the same traffic simply more expensive?
The data points to **(a) + (d)** as the primary factors, with **(b)** playing a real but specific role: SC's strategy isn't broadly less effective, but the choice of Maximise Conversions on thin-volume search campaigns demonstrably amplified CPC inflation by ~1.5x beyond the market rate. SC's quality scores are 9, CTR is comparable, CVR is better — the ROAS decline is not a general competence issue.
The same search terms cost 2-2.5x more than they did in 2020-2022 (genuine market inflation), and the post-COVID home improvement market has normalised. SC's shift to Shopping-first is arguably the correct response. However, the specific choice of Maximise Conversions as the bid strategy for generic search — rather than Manual CPC — pushed those campaigns past the point of viability. A Manual CPC approach at £1.20-1.50 caps could potentially make search viable at ~2.4x ROAS, which is marginal but may be justified if it captures incremental demand that Shopping doesn't.
---
## Appendix: Data Quality Notes
- **Source**: `google_ads_raw_daily` table, filtered to `user_id = 3` (Hairpin Leg Co) and `level = 'account'` for aggregates
- **Date range**: 2020-01-01 to 2026-02-12 (107k rows across account/campaign/ad-group levels)
- **Corrupted period**: May 1 - July 31, 2023. Conversion and conversion value columns are unreliable. Spend, clicks, and impressions remain accurate
- **Agency classification**: AP campaigns identified by `AP ` or `AP|` prefix; SC campaigns identified by `- SC` or `(SC)` suffix. One orphan campaign (`Hairpin Table Legs - Google Business Manager`, £106 total spend) fits neither
- **Currency**: All values in GBP. `cost_micros` and `conversion_value_micros` divided by 1,000,000 for human-readable figures
- **Conversion model**: Figures use the `conversions` column (primary conversions only), not `all_conversions`