knobs-and-handles

Knobs & Handles — Google Ads Historical Audit

Long-term research document tracking the full history of knobs & handles in Google Ads from launch (2020) through the present. Covers campaign structure, conversion tracking issues, the item ID change, and why performance never recovered after the initial period.

Related: audit/2026-02-11/knobs-and-handles-analysis.md covers the Dec 2025–Feb 2026 period in detail, including product economics and order behaviour analysis.

Total Investment

Across all dedicated knob/handle campaigns (2021–2025): £10,650 spent, 21,701 clicks, 1,386 attributed conversions, £58,613 attributed revenue, 5.50x ROAS. The headline ROAS is misleading — it’s inflated by PMax attribution artefacts in mid-2023 and late 2025. Stripping those out, real performance was closer to 2–3x blended, with one genuinely strong quarter (Q3 2022).

Campaign Timeline

ID	Campaign	Type	Period	Spend	Clicks	Conv	Revenue	ROAS
11	AP | Shopping | Catch All	Shopping	Jan–Jun 2020	Mixed with all products	—	—	—	—
30	AP | Standard | Trial | Knobs And Handles	Shopping	Nov 2021–Jan 2022	£1,122	1,270	32	£1,238	1.10x
33	AP | Knobs & Handles | Performance Max Trial	PMax	Jan–Apr 2022	£1,149	1,876	41	£1,957	1.70x
47	AP | Performance Max | Knobs & Handles | NEW ID’S	PMax	Jul 2022–Feb 2023	£903	1,452	58	£5,089	5.64x
49	AP | Performance Max | All Knobs and Handles	PMax	Jul 2022–Aug 2023	£4,913	7,315	1,173	£43,208	8.80x*
74	Knobs Only Shopping (SC)	Shopping	Oct 2025	£104	85	2	£61	0.58x
75	PMax: Knobs Only Shopping (SC)	PMax	Oct–Dec 2025	£2,461	9,703	80	£7,060	2.87x**

* Campaign 49 ROAS massively inflated by Jun–Jul 2023 PMax attribution artefacts (see Phase 8). ** Campaign 75 Nov–Dec 2025 data includes 2,300+ page view conversions and 2.1M impressions from PMax Display/YouTube.

Phase-by-Phase Narrative

Phase 1: Pre-History (Jan–Jun 2020)

Knob products ran inside the general “AP | Shopping | Catch All” campaign (ID 11) alongside all other products. No product-level data survives from this period — Google’s API doesn’t retain product-level Shopping data from 2020. Campaign 11 shows £2,373 total spend at 3.32x ROAS across all products, but we can’t isolate knob performance.

Some product-level data does exist from Q2 2020 in google_ads_product_raw_daily for Shopify product IDs that appear to be hairpin leg products, not knobs. The knob product Shopify IDs (5150347xxxxxx series) don’t appear until Q4 2021 in the product data.

Key gap: We cannot confirm the “knobs launched well in 2020” narrative from the data. That claim relies on memory or agency reports.

Phase 2: The Gap (Jul 2020 – Oct 2021)

No Shopping campaigns running for ~16 months. Campaign 11 stopped in June 2020. Campaign 29 (the replacement general Shopping) started September 2021 but knobs weren’t in a dedicated campaign until November 2021.

This gap was likely caused by stock-outs. The consequence was severe: Google Shopping builds product-level quality scores over time (click history, conversion rates, relevance signals). A 16-month absence meant starting from zero when the products returned.

Phase 3: Dedicated Shopping Trial (Nov 2021 – Jan 2022)

Campaign 30: “AP | Standard | Trial | Knobs And Handles”

Month	Impressions	Clicks	Spend	Conv	Revenue	ROAS
Nov 2021	185,051	998	£866	27	£1,088	1.26x
Dec 2021	28,109	155	£152	1	£34	0.23x
Jan 2022	19,395	117	£104	4	£116	1.11x

November looked promising (1.26x ROAS, 27 conversions), but December collapsed to 0.23x. January limped along before the campaign was replaced by PMax.

The drop from November to December is notable: impressions fell 85%, suggesting the agency pulled budget after poor December performance. But this is also Black Friday → post-Christmas seasonality — November always outperforms December in home goods.

All conversions were “Transactions (The Hairpin Leg Co.)” — real purchases. No page views or add-to-carts polluting the data.

Phase 4: PMax Trial (Jan – Apr 2022)

Campaign 33: “AP | Knobs & Handles | Performance Max Trial”

Month	Impressions	Clicks	Spend	Conv	Revenue	ROAS
Jan 2022	55,184	416	£222	4	£199	0.90x
Feb 2022	90,143	584	£367	14	£601	1.64x
Mar 2022	71,184	477	£294	13	£373	1.27x
Apr 2022	57,950	399	£266	9	£783	2.94x

A slow improvement trend. PMax started poorly (0.90x) but reached 2.94x by April as the algorithm learned. Campaign 30 (Standard Shopping) overlapped in January, then was paused.

Again, all conversions were “Transactions” — real purchases. PMax was learning, but £1,149 total spend over 4 months is modest; the algorithm never had enough conversion volume to optimise effectively.

Phase 5: The ID Change Gap (May – Jul 2022)

May–June 2022: No dedicated knob campaigns running. The agency (Add People) was preparing the item ID change.

July 2022: Two new PMax campaigns launched simultaneously:

• Campaign 47: “NEW ID’S” — same products with (new) appended to every offer_id (e.g., shopify_gb_5150347165830_34746935017606(new))
• Campaign 49: “All Knobs and Handles” — contained the original knob IDs plus new handle products (Knurl Pull Handle, Spiral Pull Handle, and later Bar Trim Pull Handle)

The (new) suffix was literally appended to Shopify’s auto-generated shopify_gb_PRODUCT_VARIANT format IDs. The underlying products were identical — same Shopify product ID, same variant ID, just with (new) tacked on.

Phase 6: The Good Period (Aug – Nov 2022)

The only period where knobs performed genuinely well post-relaunch. Weekly data tells the story:

Week Starting	C47 Clicks	C47 ROAS	C49 Clicks	C49 ROAS	Combined ROAS
Aug 15	42	10.72x	92	6.11x	7.41x
Aug 22	57	9.82x	97	20.49x	17.10x
Aug 29	42	17.38x	95	4.80x	8.63x
Sep 12	50	17.85x	103	9.59x	12.39x
Sep 19	54	21.69x	73	3.17x	9.39x
Oct 10	82	8.87x	78	8.25x	8.53x
Oct 24	85	3.74x	101	6.61x	5.24x
Nov 7	63	8.52x	183	8.93x	8.84x
Nov 14	57	8.56x	189	4.93x	5.57x

Multiple weeks above 8x ROAS. Combined spend of £100–180/week yielding strong returns. This is the performance that justified continued investment.

Conversion audit confirms these were real purchases. The only conversion action firing was “Transactions (The Hairpin Leg Co.)” — a PURCHASE category action using data-driven attribution. No page views, no add-to-carts, no begin-checkouts in the mix.

Phase 7: The Collapse (Late Nov 2022 – Apr 2023)

The collapse was sudden and affected both campaigns simultaneously:

Week Starting	C47 Clicks	C47 ROAS	C49 Clicks	C49 ROAS	Combined ROAS
Nov 21	51	5.60x	152	1.49x	2.44x
Nov 28	60	0.24x	171	0.48x	0.43x
Dec 5	49	1.33x	168	3.33x	2.95x
Dec 12	44	0.62x	117	9.42x	7.32x
Dec 19	30	0.16x	97	2.75x	2.42x
Dec 26	28	0.00x	57	1.98x	1.58x
Jan 2	27	12.88x	70	8.41x	9.39x
Jan 9	34	5.42x	133	1.50x	2.04x

The week of November 28 was the inflection point: ROAS crashed to 0.43x across both campaigns. From here, performance became volatile and never sustainably recovered.

What happened to campaign 47 (NEW ID’S):

• Google progressively reduced its impressions from December onwards
• By January 2023, it was getting 18–34 clicks/week (down from 50–80)
• The algorithm effectively killed it — new products with no history building
• Last data: February 23, 2023

What happened to campaign 49 (All Knobs):

• Continued running but volatile. Some recovery weeks (Jan 2, Jan 16) mixed with poor ones
• February 20–March 27: The agency scaled budget dramatically, pushing from ~100 clicks/week to 350–537 clicks/week
• This was catastrophic: ROAS dropped to 0.03x–1.54x while burning £300–350/week
• The forced scaling sent Google to increasingly poor-quality traffic
• April 2023: Pulled back to ~100 clicks/week, still 0–1x ROAS
• May 2023: Effectively turned off (£7 spend)

Phase 8: Attribution Ghosts (Jun – Aug 2023)

Campaign 49 shows implausible numbers from June 2023:

Week Starting	Clicks	Spend	Conversions	ROAS
Jun 19	87	£87	92	14.32x
Jun 26	155	£67	176	111.90x
Jul 3	191	£75	215	53.84x
Jul 10	179	£65	249	157.31x
Jul 17	184	£69	239	87.28x

These are garbage numbers. The conversion action breakdown reveals why:

Month	Conversion Action	Category	Include in Conv?	Count
Jun 2023	Analyzify - Page View	PAGE_VIEW	No	222
Jun 2023	Analyzify - Purchase	PURCHASE	Yes	1
Jul 2023	Analyzify - Page View	PAGE_VIEW	No	787
Jul 2023	Analyzify - Purchase	PURCHASE	Yes	11
Jul 2023	Analyzify - Begin Checkout	BEGIN_CHECKOUT	Yes	7
Jul 2023	Analyzify - Add To Cart	ADD_TO_CART	Yes	4

The massive conversion counts are Page View conversions from Analyzify’s pixel, not purchases. The “conversions” column in google_ads_raw_daily appears to aggregate all conversion actions (including non-purchase ones). Only ~12 actual purchases across the two months.

August 2023: All AP-era campaigns were paused. The SC (self-managed) era began.

Phase 9: SC Era — General Campaigns (2024 – Sep 2025)

Knob products served passively through general Shopping campaigns (Brand Shopping, Shopping Catch All). No dedicated knob campaigns.

Product-level data from google_ads_product_raw_daily shows knob products getting steady but low-volume Shopping traffic:

Quarter	Products Served	Clicks	Spend	Conversions	Conv Value
Q2 2024	22	25	£12	0	£0
Q3 2024	22	8	£5	0	£0
Q1 2025	29	228	£216	22	£1,019
Q2 2025	29	400	£322	30	£2,861
Q3 2025	29	364	£360	27	£1,549

(Q4 2024 excluded — Brand Shopping campaign restructuring period with minimal knob traffic.)

In 2025, the general Shopping campaigns actually delivered reasonable knob/handle performance. Q2 2025 showed 30 conversions on £322 spend = 8.88x ROAS through Shopping Catch All and Brand Shopping combined. The products found their natural level of Shopping demand without forced budget.

Phase 10: Revival Attempt (Oct – Dec 2025)

Campaign 74: Knobs Only Shopping (SC) — Killed after 3 weeks.

• 85 clicks, £104 spend, 2 conversions, £61 revenue = 0.58x ROAS

Campaign 75: PMax: Knobs Only Shopping (SC) — Ran Oct–Dec 2025.

Month	Clicks	Spend	Purchase Conv	Purchase Value	Page Views (all_conv)
Oct 2025	—	—	2	£163	63
Nov 2025	—	—	68	£6,552	2,300
Dec 2025	—	—	10	£345	4,233
Total	9,703	£2,461	80	£7,060	6,596

The 80 “Analyzify - Purchase” conversions (which ARE include_in_conversions = true) at £7,060 give a 2.87x ROAS on paper. But November’s 68 purchases on a campaign with 912K impressions and 2.1M PMax reach raises the same attribution concerns as 2023. PMax claims conversions across all channels — any customer who saw a Display ad and later purchased organically gets attributed to PMax.

The conversion action breakdown also shows that “Analyzify - Add To Cart” (include_in_conversions = true) was racking up 182 conversions in November alone, suggesting the bidding algorithm was partially optimising toward add-to-carts rather than purchases.

---

Critical Finding: The Conversion Tracking Configuration

This is the most important finding in the audit.

The Problem

“Transactions (The Hairpin Leg Co.)” — the conversion action that tracked real purchases during the 2021–2023 era — was configured with:

Setting	Value	Impact
include_in_conversions	FALSE	Smart Bidding ignores this action
attribution_model	Data-Driven	Good model, but irrelevant if excluded
category	PURCHASE	Correctly categorised
status	REMOVED	No longer active

With include_in_conversions = false, Google’s Smart Bidding algorithms (tROAS, Maximise Conversions, etc.) did not optimise toward this conversion action. The campaigns were tracking purchases for reporting, but the bidding algorithm was not learning from them.

What Was the Algorithm Targeting?

During the 2021–2023 period, the purchase-category conversion actions with include_in_conversions = true were:

ID	Name	Type	Include?	Attribution
1	Transactions (The Hairpin Leg Co.)	UA Transaction	No	Data-Driven
2	The Hairpin leg co - hairpin leg conversion tracking from adwords	Webpage pixel	Yes	Last Click
8	Hairpin Leg Company	Webpage pixel	No	Last Click

Only conversion action 2 (“The Hairpin leg co - hairpin leg conversion tracking from adwords”) was included. This is an older Adwords-era pixel with Last Click attribution. It’s unclear whether this pixel was even firing correctly during 2022 — if it was a legacy tag on the old site, the algorithm may have been optimising toward nothing meaningful.

Why This Matters

If Smart Bidding had no reliable purchase signal to optimise toward during the 2022–2023 campaigns, then:

1. The “good” Aug–Nov 2022 period may have been coincidental — the algorithm stumbling into good traffic despite not knowing what a conversion looked like
2. The Nov 2022 collapse may be the algorithm drifting — without conversion signals, PMax gradually shifts to engagement/traffic metrics, serving to users who click but don’t buy
3. The agency’s budget scaling in Mar 2023 was doomed from the start — you can’t push an algorithm to find more purchases if it doesn’t know what a purchase is
4. The entire “knobs don’t work in Google Ads” narrative may be based on a broken setup

Current State (2025–2026)

The Analyzify conversion tracking (installed later) has Analyzify - Purchase with include_in_conversions = true. The 2025 knob campaigns (Oct–Dec) had purchase-only bidding — ATC and Begin Checkout were secondary (observation only) during that period, appearing only in the all_conversions column.

Analyzify - Add To Cart and Analyzify - Begin Checkout were switched to primary (included in bidding) around February 2026. Before that:

Month	ATC in Bidding	ATC Observed	BC in Bidding	Purchase in Bidding
Aug–Nov 2025	0	940–1,245	0	312–473
Dec 2025	1	983	1	645
Jan 2026	1	1,059	0	745
Feb 2026	14	308	5	184

This means the 2025 knob PMax campaign’s 2.87x ROAS was achieved with purchase-only bidding and correct conversion tracking — a materially fairer test than any of the 2021–2023 campaigns. The result (80 attributed purchases on £2,461 spend) still carries PMax attribution concerns, but the bidding setup was sound.

Micro-Conversion Strategy (Feb 12, 2026 Change)

Change date: February 12, 2026 at 14:06. Daniel at Snowball Creations switched “Analyzify - Begin Checkout” to primary (confirmed in Google Ads change history). “Analyzify - Add To Cart” was switched in the same session (not separately logged).

The ATC/BC switch to primary was applied account-wide, not limited to knob campaigns. In February 2026, ATC bidding conversions appeared across all major Shopping campaigns:

Campaign	ATC (bidding)	ATC (observed)
Shopping Top Performers	6.0	60
Shopping Catch All <£20	4.0	95
Shopping Catch All >£20	2.8	85
Brand Shopping	1.0	11
Brand Search	0.0	44
PMax: Table Tops	0.0	10

Why Account-Wide Might Be Justified

The per-SKU conversion data challenges the assumption that only low-volume categories need micro-conversions. In the last 3 months (Nov 2025 – Jan 2026):

Per-SKU Conversions	SKU Count	% of Total
0 conversions	541	72%
Fractional (0–1)	43	6%
1–2	104	14%
3–9	41	5%
10–29	19	2.5%
30+ (Smart Bidding threshold)	5	0.7%

Only 5 out of 753 SKUs clear the 30 conversions/quarter threshold. Those 5 are consolidated hairpin leg product groups (mkr-30-wa-01, etc.). Every other product category — including core ones — has the same per-SKU volume problem:

Category	SKUs	Total Conv	SKUs with 10+ Conv	Avg Conv (converting SKUs)
Hairpin Legs (mkr)	5	216	3	43.2
Box/Frame Legs	67	138	3	4.0
J-Hooks	18	121	5	10.1
Wall Shelves	10	53	1	6.6
Brackets/Shelf	49	50	1	2.8
Desk Legs	39	48	1	4.4
Wall Hooks	5	42	2	10.6
Knobs	12	8	0	2.0
Pull Handles	10	0	0	—

This is a long-tail problem: 753 SKUs generating 1,073 conversions = 1.4 conversions per SKU per quarter. At campaign level, Shopping Top Performers generates enough purchase volume for Smart Bidding. But 72% of the individual products the algorithm must bid on have zero conversion history to learn from.

The argument for ATC as primary: an add-to-cart gives the algorithm per-product intent signals it can’t get from purchase data alone. If 50 people add a specific desk leg variant to cart in a month (even if only 15 buy), that’s 50 data points the algorithm can use for bid optimisation on that SKU, vs 15 without ATC.

The Trade-Off

The algorithm optimises toward whatever it counts as conversions. With ATC included, tROAS bidding targets a blend of purchase ROAS and add-to-cart “ROAS”. Since ~70% of add-to-carts don’t convert to purchase, the algorithm may learn to find efficient add-to-cart traffic that doesn’t actually buy.

The risk is higher for PMax than Standard Shopping. In Standard Shopping, the algorithm can only serve Shopping ads to people with purchase intent (they searched for a product). In PMax, adding ATC to bidding gives the algorithm permission to count Display/YouTube “add-to-cart” events as success — potentially accelerating the exact Display expansion problem that killed the knob campaigns in 2022.

A value hierarchy would help mitigate this: assign Purchase = full transaction value, Begin Checkout = 50%, ATC = 25%. This tells the algorithm which signal matters more, rather than treating all three equally. Whether Snowball have set differentiated values should be verified.

ROAS Inflation: The Reporting Problem

The ATC/BC change has a serious side-effect: every ROAS figure in Google Ads is now significantly inflated.

ATC conversion events record the full cart value at the time of add-to-cart. If a customer adds £100 of products to cart and then purchases, that generates £100 of ATC conversion value AND £100 of Purchase conversion value — £200 of reported “conversion value” on £100 of actual revenue.

Feb 1–12, 2026 comparison (reported vs purchase-only):

Campaign	Spend	Reported Conv	Reported Value	Reported ROAS	Purchase Conv	Purchase Value	Real ROAS	Inflation
Brand Search	£4,137	54.1	£9,123	2.21x	35.1	£6,424	1.55x	30%
Catch All >£20	£3,349	106.9	£12,104	3.61x	62.0	£6,859	2.05x	43%
Top Performers	£2,384	72.1	£8,868	3.72x	37.0	£4,120	1.73x	54%
Catch All <£20	£668	52.0	£1,952	2.92x	27.0	£1,147	1.72x	41%
Brand Shopping	£276	21.4	£1,267	4.59x	12.6	£774	2.80x	39%

Three consequences:

1. All ROAS reporting is inflated 30–54%. Shopping Top Performers shows 3.72x when purchase ROAS is 1.73x. That’s the difference between “profitable campaign” and “barely breaking even after COGS.”

2. Historical data is retroactively inflated. Google applies include_in_conversions settings to all historical data. Reports for January, December 2025, etc. now show ATC in the Conversions column too, even though ATC was secondary during those periods. Before/after comparisons against pre-Feb 12 data are meaningless in the Google Ads UI.

3. Smart Bidding targets a blended number. When tROAS is set to 3.5x, the algorithm targets 3.5x on purchases + add-to-carts combined. A campaign could deliver 2x purchase ROAS while the algorithm believes it’s hitting 3.5x and doesn’t adjust. The bidding is effectively set ~40% looser than intended.

This is an account-wide issue, not knob-specific. Every campaign with ATC activity is affected. The inflation is highest on Shopping campaigns (40–54%) because Shopping drives the most add-to-cart events.

Recommendation: If the micro-conversion strategy is retained for its per-SKU signal benefits, the tROAS targets need to be recalibrated upward to account for the inflated denominator. A 3.5x tROAS with ATC included is roughly equivalent to a 2.0x purchase-only tROAS. Alternatively, assign ATC a nominal value (e.g. £1 or 25% of cart value) rather than the full transaction value, which preserves the signal volume benefit without destroying ROAS reporting accuracy.

---

The Item ID Change Investigation

What Happened

In July 2022, the agency created campaign 47 using the exact same Shopify product/variant IDs with (new) appended:

Original:  shopify_gb_5150347165830_34746935017606
New:       shopify_gb_5150347165830_34746935017606(new)

This was done for all ~28 knob variants across the 7 original knob products. The rationale was presumably to “reset” the products’ quality scores in Google’s system.

Did It Work?

No. Campaign 47 (new IDs) and Campaign 49 (original IDs + handles) were running simultaneously from July 2022 onwards. Comparing their performance:

Period	C47 (New IDs) ROAS	C49 (Original IDs) ROAS	Winner
Aug 2022	8.53x	7.51x	New IDs (marginally)
Sep 2022	11.57x	5.93x	New IDs
Oct 2022	4.17x	4.38x	Tie
Nov 2022	5.43x	3.78x	New IDs
Dec 2022	0.66x	4.68x	Original IDs
Jan 2023	4.07x	3.26x	New IDs
Feb 2023	4.46x	1.65x	New IDs

Campaign 47 (new IDs) had slightly better ROAS in most months, but both campaigns collapsed at the same time (late November 2022). If the ID change were the differentiating factor, one campaign should have performed differently from the other. Instead, they rose and fell together, suggesting the performance driver was external (seasonality, competition, algorithm behaviour) not product-ID-specific.

Google also progressively reduced impressions for campaign 47 from December onwards, despite its (marginally) better ROAS. By January 2023, it was getting 18–34 clicks/week while campaign 49 got 70–150. The algorithm preferred the campaign with more products and more history, regardless of the “new” IDs.

Conclusion: The item ID change was irrelevant. The (new) suffix didn’t meaningfully change performance, and Google treated the products similarly regardless.

The Handle Products Tell a Different Story

Campaign 49 contained new product types — Knurl Pull Handles, Spiral Pull Handles, and (from Feb 2023) Bar Trim Pull Handles. These weren’t just knobs with new IDs; they were genuinely different products:

Product	Lifetime Clicks	Lifetime Spend	Type
Knurl Pull Handle (6540915441798)	9,965	£14,176	Pull Handle
Bar Trim Pull Handle (6540915507334)	3,812	£4,093	Pull Handle
Spiral Pull Handle (6540915474566)	3,157	£3,238	Pull Handle

The Knurl Pull Handle alone accumulated more spend than all dedicated knob campaigns combined. These are higher-priced products (£10–20 per unit) that appear to have stronger Shopping demand. The “All Knobs and Handles” campaign’s lifetime ROAS was heavily carried by the handle products, not the original knob designs.

---

What Caused the Late November 2022 Collapse?

The data rules out several potential causes and points to a specific mechanism.

What It Was NOT

Not an account-wide problem. The rest of the account performed well through the exact same period:

Week	Knob ROAS	Rest of Account ROAS
Nov 7	8.84x	8.49x
Nov 14	5.57x	6.81x
Nov 21	2.44x	8.85x
Nov 28	0.43x	6.55x
Dec 5	2.95x	8.40x
Dec 12	7.32x	6.09x

The rest of the account maintained 6–8.5x ROAS throughout Black Friday and into December. Whatever killed knobs didn’t affect anything else.

Not a CPC spike (competition). CPCs on both knob campaigns actually decreased through the collapse — campaign 47 went from £0.74 (Nov 21) to £0.56 (Nov 28) to £0.33 by late December. Campaign 49 stayed in the £0.50–0.87 range. Competitors didn’t outbid Hairpin.

Not a conversion tracking change. The only conversion action firing throughout was “Transactions (The Hairpin Leg Co.)” — same before, during, and after the collapse. No new actions appeared, no existing ones stopped.

What Actually Happened: PMax Display/YouTube Expansion

The impression and CTR data tells a clear story:

Week	Impressions	CTR	Clicks	Conversions	Clicks per Conv
Aug 22 (peak)	8,572	1.80%	154	9.1	17
Sep 12	8,777	1.74%	153	9.9	15
Oct 24	14,218	1.31%	186	14.3	13
Oct 31	21,035	0.95%	199	5.5	36
Nov 7	25,292	0.97%	246	9.2	27
Nov 14	23,102	1.06%	246	10.0	25
Nov 21	18,350	1.11%	203	4.1	50
Nov 28	25,532	0.90%	231	3.0	76
Dec 5	24,482	0.89%	217	4.9	44

The progression:

1. Aug–Sep: 8–10K impressions/week, 1.7–1.8% CTR, 15–17 clicks per conversion. Tight, Search-heavy traffic that converts.
2. Late Oct: Impressions jumped to 21K. CTR dropped below 1%. PMax started expanding into Display/YouTube to spend the daily budget.
3. Nov: Impressions hit 25K. The extra 15K impressions came with 0.3–0.5% CTR (Display/YouTube typical). The algorithm was now generating 2–3x more impressions but the same number of clicks, and fewer conversions.
4. Nov 28: The ratio hit 76 clicks per conversion (vs 13–17 during the good period). For every converting click, the algorithm was burning 75 non-converting ones.

PMax reports all network types as “MIXED”, so we can’t see the exact Search/Display/YouTube split. But the pattern is unmistakable: impressions doubled while CTR halved and conversion rate collapsed. That only happens when the algorithm shifts budget from Search (high CTR, high intent) to Display/YouTube (low CTR, low intent).

Why PMax Expanded

PMax automatically expands to non-Search channels when it can’t fully spend the budget on Search alone. For niche products like knobs, there’s limited Search inventory — only so many people search for “kitchen knobs” each week. When the daily budget exceeds what Search can absorb, PMax fills the gap with Display and YouTube placements.

The expansion likely started around late October (first week above 20K impressions). ROAS was still acceptable because existing Search conversions masked the Display waste. But as PMax allocated more budget to non-Search channels, the conversion signal weakened:

1. Fewer Search conversions (budget diverted to Display)
2. PMax interprets Display impressions as “this audience works” (because there’s no conversion signal saying otherwise — remember, include_in_conversions = false on the purchase action)
3. Algorithm doubles down on Display/YouTube (it thinks the expanded audience is working)
4. Search allocation shrinks further (self-reinforcing decline)

The Conversion Tracking Made This Worse

With “Transactions” set to include_in_conversions = false, Smart Bidding had no reliable purchase signal to self-correct. In a properly configured account, the algorithm would see Display clicks producing zero purchases and shift back to Search. Without that signal, the feedback loop had no brake.

Product-Level Impact

Before the collapse (Oct 10–Nov 14), 12 different product variants were converting, including 5 knob variants. After (Nov 21–Dec 26), only the Knurl Pull Handle variants continued converting with any consistency. The original knob products (515034xxx series) almost entirely stopped converting — they were the first casualties of the traffic quality decline.

15 products also dropped from the feed around December 5 (product count fell from 58 to 43), which further reduced the campaign’s ability to match relevant searches.

The Agency’s Response Made It Worse

Add People responded to the decline by dramatically scaling campaign 49’s budget in late February – March 2023:

Week	Clicks	Spend	ROAS	Agency Action
Feb 13	107	£48	7.62x	Stable (small)
Feb 20	362	£220	0.23x	Scaled 4x
Feb 27	506	£325	1.52x	Still scaled
Mar 6	537	£356	1.54x	Still scaled
Mar 13	477	£342	0.29x	Still scaled
Mar 20	448	£329	0.03x	Still scaled
Mar 27	460	£347	0.39x	Still scaled

Four weeks at £300–350/week with sub-1x ROAS = £1,374 burned. The forced scaling sent Google to lower-quality traffic that couldn’t convert. The correct response would have been to reduce spend and investigate the conversion tracking setup.

---

Product-Level Insights

Order Economics: Both Categories Are Multi-Unit

The unit price difference between knobs (£7–9) and handles (£10–20) is misleading. Both categories are multi-unit purchases:

Metric	Knobs	Pull Handles
Orders containing product	1,482	1,281
Total line items	1,634	1,779
Average quantity per line	5.5	5.2
Average line value	£46	£76
Total category revenue	£74,412	£135,910

Quantity distribution (line items):

Qty Bucket	Knobs	Avg Knob Value	Handles	Avg Handle Value
1	338 (21%)	£8	493 (28%)	£15
2	354 (22%)	£16	338 (19%)	£29
3–4	341 (21%)	£30	343 (19%)	£54
5–6	213 (13%)	£47	197 (11%)	£86
7–10	187 (11%)	£69	178 (10%)	£120
11–20	146 (9%)	£120	156 (9%)	£209
21+	55 (3%)	£277	74 (4%)	£396

Both products show the same multi-unit buying pattern. The difference is price-per-unit (handles are ~1.7x more expensive), not buying behaviour. A customer buying 6 handles generates ~£86 vs ~£47 for 6 knobs — handles are higher-value but not a fundamentally different purchase pattern.

Handle-only orders: Average order total of £118, with £106 of that being handle products and an average of 7.2 handles per order.

Cross-buy: 219 orders (8% of either category) contain both knobs and handles. The categories are largely separate customer pools — people buying knobs are rarely buying handles in the same order, and vice versa.

Knobs vs Handles in Google Ads (2025)

The 2025 SC-era data, where both categories served passively through general Shopping campaigns, reveals something counterintuitive:

Quarter	Knob Clicks	Knob Spend	Knob ROAS	Handle Clicks	Handle Spend	Handle ROAS
Q1 2025	152	£127	3.67x	776	£1,228	1.81x
Q2 2025	239	£208	8.43x*	708	£874	3.30x
Q3 2025	228	£205	4.26x	595	£1,002	2.96x
Q4 2025	2,523	£421	1.48x	518	£970	1.98x

* Q2 2025 knob ROAS is skewed by three bulk orders (£453, £358, £184) that account for 75% of the quarter’s revenue. Excluding those weeks, the baseline is closer to 2.5–3x ROAS. Low-volume categories are inherently volatile — a single 50-knob order can swing a quarter’s numbers.

Handles absorbed 4–6x more budget than knobs. This isn’t because handles are “better” at converting — it’s because Google’s algorithm learned handle conversion patterns during the 2022–2024 PMax/Shopping campaigns and continues to serve them aggressively. Knobs were left passive and their ROAS looks healthy at natural demand levels, though it’s volatile week-to-week.

Realistic baseline: Knobs at passive Shopping volumes deliver ~2.5–4x ROAS, comparable to handles at similar spend levels. The headline quarterly numbers bounce around because any single bulk order (20+ units at £7–9 each = £140–180) can double a week’s revenue on £10–20 of spend.

The Knurl Pull Handle (product ID 6540915441798) alone accounts for £14,176 in lifetime Shopping spend — more than all dedicated knob campaigns combined. Its 2025 ROAS was 1.69–2.67x, which is in the same realistic range as knobs once you account for volume volatility. The handles just get more budget thrown at them.

Best-Performing Knob Products (2025, All Finishes)

Product	Clicks	Spend	Conv	Conv Value	ROAS
Knurl 40mm (kndisk2)	345	£275	25	£1,258	4.57x
Disk 30mm	197	£142	15	£1,003	7.05x
Knurl 18mm (knpip)	80	£85	11	£600	7.04x
Conical 18mm	73	£55	4	£145	2.62x
Hex 30mm	58	£34	5	£239	6.97x
Square 30mm	27	£23	0	£0	0.00x
Bowl 40mm	28	£30	4	£265	8.95x

The Knurl 40mm, Disk 30mm, and Knurl 18mm are reliable performers. Bowl 40mm shows strong ROAS but on very few clicks.

Best-Performing Handle Products (2025)

Product	Clicks	Spend	Conv	Conv Value	ROAS
Knurl Pull Handle	2,013	£3,659	94	£7,665	2.09x
Bar Trim Pull Handle	435	£286	15	£1,773	6.20x
Spiral Pull Handle	142	£123	6	£563	4.59x
Knurl 15mm Pull Handle	7	£6	0	£0	—

The Knurl Pull Handle consumes most of the handle budget but at only 2.09x ROAS. Bar Trim and Spiral handles perform better at lower volumes. The same pattern as knobs: the products that get less forced budget have better ROAS.

Low-Volume Variance: The Industrial Nickel Example (wh-kndisk2-40-sl)

The Knurl 40mm Knob in Industrial Nickel was the best-performing knob variant in 2025:

• Q2 2025: 45 clicks, £34 spend, 8 conversions, £644 value = 18.79x ROAS
• Q3 2025: 33 clicks, £34 spend, 2 conversions, £148 value = 4.34x ROAS

This isn’t an unexplained anomaly — it’s a direct consequence of low volume. At 30–45 clicks per quarter, a single multi-unit order (e.g. 10 knobs at £8.95 = £90) can double the quarter’s ROAS. The same applies in reverse: a quarter with zero large orders looks terrible. This variance is inherent to any product with under ~100 clicks/month and is exactly why quarterly ROAS figures for individual knob variants should be treated as noise rather than signal.

Handle Historical Performance by Quarter

The Knurl Pull Handle was by far the most-served handle product. Its trajectory shows the diminishing returns of scale:

Period	Knurl PH Clicks	Knurl PH Spend	Knurl PH ROAS	Campaign Context
Q3 2022	446	£378	3.92x	PMax “All Knobs and Handles”
Q4 2022	1,080	£1,027	5.00x	Peak period
Q1 2023	739	£459	1.60x	Post-collapse
Q2 2023	252	£126	32.80x*	Attribution artefact
Q3 2023	208	£78	68.44x*	Attribution artefact
Q4 2023	557	£1,011	3.61x	SC general campaigns
Q1 2024	1,800	£1,935	1.53x	Heavy spend era
Q2 2024	1,985	£3,282	1.45x	Peak spend, poor ROAS
Q3 2024	365	£898	1.14x	Budget cut
Q1 2025	706	£1,190	1.69x	Stabilised
Q2 2025	502	£720	2.67x	Reduced spend, better ROAS

* Q2-Q3 2023 are PMax attribution artefacts (page view conversions).

The pattern is clear: the Knurl Pull Handle performs best at moderate spend levels. When pushed to £1,000+/quarter, ROAS drops below 2x. At £700–900/quarter, it settles around 2–2.5x. This is the same over-scaling problem that killed the dedicated knob campaigns.

---

Should Knobs & Handles Be Retried?

Arguments For

1. The conversion tracking was broken. Every dedicated knob campaign (2021–2023) ran with include_in_conversions = false on the purchase action. Smart Bidding was flying blind. A retry with correctly configured conversion tracking could perform materially better.

2. Passive Shopping already works for both categories. In 2025, knobs generated 15–25 conversions per quarter at ~2.5–4x ROAS through general campaigns (volatile due to low volume, individual bulk orders swing quarterly numbers). Handles generated 20–38 conversions at 2–3x ROAS. The product-market fit exists for both at modest spend levels.

3. The AOV supports paid search for both. Knobs: avg line value £46, avg order total £86. Handles: avg line value £76, avg order total £118. Neither is the £7 single-unit sale that the unit price suggests. Both are viable at CPCs under £1.

4. Handles have never been tested with correct conversion tracking either. The Knurl Pull Handle’s £14K lifetime spend was almost entirely through PMax campaigns with broken or diluted conversion goals. Standard Shopping with purchase-only bidding has never been tried at scale.

Arguments Against

1. Four separate dedicated knob attempts all failed. Different structures, different times, same result. Though the broken conversion tracking casts doubt on all of them.

2. Handles already hit diminishing returns at current spend. The Knurl Pull Handle drops below 2x ROAS above ~£250/quarter spend. There may not be enough converting search demand to justify a dedicated campaign.

3. Generic search terms, low brand intent. People search for “copper kitchen handles” and “cabinet knobs” — generic terms where Hairpin competes on price against B&Q, Screwfix, and Dunelm.

4. PMax is the wrong format. Every PMax attempt dumped 25–30% of budget into Display/YouTube with zero conversions. Any retry must be Standard Shopping only.

5. Opportunity cost. £10,650 on knobs could have gone to table legs or shelving. But the same argument applies in reverse: the £14K spent on the Knurl Pull Handle at 1.5x ROAS was also poorly allocated.

Recommendation

Do not create a new dedicated campaign yet. Instead:

1. Evaluate the conversion tracking change. ATC and Begin Checkout were switched to primary (bidding) in Feb 2026. At the account level (300–745 purchases/month), this dilutes the bidding signal unnecessarily. Consider reverting them to secondary for general campaigns, and only using them as primary if a dedicated low-volume campaign needs the signal volume.

2. Monitor passive Shopping performance for both categories. The general Shopping campaigns are already generating reasonable volumes. Knobs: 15–25 conv/quarter at ~2.5–4x baseline ROAS (volatile due to low volume). Handles: 20–38 conv/quarter at 2–3x ROAS.

3. Investigate the Knurl Pull Handle over-spend. This single product absorbs £700–1,200/quarter in general Shopping campaigns at 1.7–2.7x ROAS. It may be worth capping its spend or ensuring it doesn’t crowd out better-performing products (Bar Trim at 6.2x, individual knobs at 4–9x).

4. If a dedicated campaign is warranted, use Standard Shopping only (no PMax), combining both knobs and handles:

   • £10–15/day maximum budget
   • TARGET_ROAS bidding at 3x minimum
   • Purchase-only conversion goal (correctly configured)
   • Run for 3 months minimum before evaluating

5. Track category-level, not unit-level, metrics. The right comparison isn’t “£7 knob vs £15 handle” — it’s “£46 avg knob line vs £76 avg handle line.” Both categories have viable economics; the question is whether dedicated budget can scale them without destroying ROAS. Historical evidence says the ceiling is modest for both.

---

Unanswered Questions

1. What was Aug–Dec 2020 performance? No product-level data. Would need agency reports or Google Ads UI export history from Add People.

2. Was the gap really 16 months? Were knobs in a campaign we can’t see, or genuinely absent from the feed?

3. Competitor landscape change? No competitive data from 2020–2022 to compare against.

4. Was conversion action 2 (“The Hairpin leg co”) actually firing during 2022? If this legacy Adwords pixel was broken, Smart Bidding had literally zero conversion signal.

5. Can correct conversion tracking change the outcome? This is the key open question. Every prior attempt had a broken setup. A clean retry with proper tracking would definitively answer whether the product works in paid search or not.

Data Sources

• google_ads_raw_daily — campaign-level daily data (complete from Jan 2020)
• google_ads_product_raw_daily — product-level Shopping data (sparse 2020, good from Sep 2021)
• google_ads_conversion_action_daily — conversion breakdown by action per campaign per day
• google_ads_conversion_actions — conversion action configuration (attribution model, include_in_conversions, lookback windows)
• google_ads_search_term_daily — search term data
• ad_campaigns — campaign metadata