Shopping High Interiors Limited 24 Oct
Also capped 24 Aug to 6 Sep (shaded). Before 26 Sep to 23 Oct · after 24 Oct to 8 Nov.
- Conv/day
- 6.55 → 7.76+18%
- ROAS
- 5.79 → 3.06−47%
- CPC
- £1.81 → £2.70+49%
How we read it
- Lost to budget is the share of impressions Google says we were eligible for but didn't get because the daily budget ran out (Google's "search lost IS (budget)").
- Limited date is the first of 3 or more days in a row with 10% or more lost to budget. Shopping High: 24 Oct (also capped 24 Aug to 6 Sep). Search Sheepskin Rugs: 29 Oct. PMax Sheepskin: 26 Oct (also 27 Aug).
- Before is the 28 days before the limited date; after runs to 8 Nov. Campaigns that weren't limited use 21 Sep to 18 Oct vs 27 Oct to 8 Nov.
- Declined means ROAS or conversions per day down more than 15%.
Reading the numbers
- ROAS and conversions are Google's own (primary purchase action, last click). TW linear is Triple Whale's linear attribution, used as a second opinion.
- 9 Nov is left out: its conversions are still coming in.
- r is the correlation between daily lost to budget and daily ROAS, 1 Aug to 8 Nov. Near 0 means no link; −1 means a perfect negative link.
- All figures are campaign totals. No customer data is used.
Campaign by campaign
Three campaigns became limited in late October. Three others were barely or never limited, so they show what happened anyway. Bars are the weekly result; the red line is the share lost to budget.
Search Sheepskin Rugs Limited 29 Oct
Current episode. Before 1 to 28 Oct · after 29 Oct to 8 Nov.
- Conv/day
- 4.81 → 3.79−21%
- ROAS
- 5.50 → 2.24−59%
- CPC
- £1.62 → £2.16+33%
A short 3-day limited run from 4 Sep is ignored here; the campaign then ran unlimited until late October.
PMax Sheepskin Limited 26 Oct
Also capped from 27 Aug (shaded). Before 28 Sep to 25 Oct · after 26 Oct to 8 Nov.
- Conv/day
- 1.95 → 2.05+5%
- ROAS
- 8.90 → 3.87−57%
- CPC
- £1.27 → £1.58+24%
PMax Cowhide Control · barely limited
Lost to budget 9% at most in a week since September. Before 21 Sep to 18 Oct · after 27 Oct to 8 Nov.
- Conv/day
- 0.76 → 1.44+90%
- ROAS
- 7.51 → 4.65−38%
- CPC
- £1.33 → £1.60+20%
PMax Zombie SKUs Not a clean control
Reads 0% lost to budget every day, but 99% of its daily spend is gone by 20:00 (84% before). Before 21 Sep to 18 Oct · after 27 Oct to 8 Nov.
- Conv/day
- 1.42 → 1.75+23%
- ROAS
- 8.42 → 3.46−59%
- CPC
- £0.76 → £1.84+142%
Search Brand UK & US Control
Never limited (21% on a single day, 8 Nov). Before 21 Sep to 18 Oct · after 27 Oct to 8 Nov.
- Conv/day
- 6.68 → 10.12+52%
- ROAS
- 65.6 → 44.2−33%
- CPC
- £0.27 → £0.35+30%
Weeks run Monday to Sunday. "Declined" uses a 15% threshold, so small campaigns (about 2 conversions a day) can tip either way on a few orders.
The evidence that the cap is hurting
These are the strongest signs that the budget limit, rather than Q4 alone, is costing us. Each has limits, noted alongside.
A · Same week, cap on vs off
Shopping High, 2 to 3 Nov vs 4 to 7 Nov
On 2 to 3 Nov the campaign lost 0 to 2% to budget. On 4 to 7 Nov it lost 34 to 56%. Spend per day was similar, and both periods sit after the tracking drop.
| 2 to 3 Nov 0 to 2% lost | 4 to 7 Nov 34 to 56% lost | |
|---|---|---|
| ROAS (Google) | 5.98 | 1.91 |
| ROAS (TW linear) | 4.48 | 1.69 |
| CPC | £2.18 | £3.32 |
| Conversion rate | 4.9% | 2.5% |
| Spend per day | £479 | £546 |
Indicative only: 2 days vs 4 days, and Mon to Tue vs Wed to Sat.
B · Days grouped by how much was lost
Shopping High: the more lost to budget, the worse the day
Daily, 7 Sep to 7 Nov. Capped days are the expensive-click days: r(lost to budget, CPC) = +0.82 (n = 62). A cap that rationed spend toward the best auctions would show the opposite.
Every day over 20% falls on or after 26 Oct, so this mixes the cap with Q4 prices and the tracking drop.
C · Hour of day
The budget runs out before the evening
- Spent by 20:00
- 86% → 95%
- Evening share (18:00 to 23:59)
- 33.6% → 25.0%
Those lost evening hours are now Shopping High's best value: 1.26× the day's average value per £. Brand, which is barely capped, held steady (26.2% → 25.6% evening share), so this isn't people shopping at different times. PMax Zombie is the extreme: 99% of spend gone by 20:00.
D · Actual ROAS ÷ target
Shopping High went from beating its targets to missing them
From w/c 14 Sep to w/c 12 Oct, Shopping High ran at 1.25 to 1.52× its spend-weighted target. Since it became limited: 0.66× (w/c 26 Oct) and 0.64× (w/c 2 Nov).
About 0.94× in w/c 2 Nov if we correct for the tracking drop (an estimate). Targets before 12 Oct are taken from the first logged change.
E · Context
Google changed how capped campaigns bid in August 2026
From 17 August 2026, Google changed how target CPA and target ROAS campaigns behave when they are limited by budget: they now deliver closer to the target you set rather than beating it. In practice, a capped campaign no longer builds a cushion of ROAS above target, so where it lands depends more on the target we set. That fits what we see in card D, though we can't prove it is the cause. Google's help page on the change.
What this doesn't prove
- The controls fell too. Over the same windows, barely-limited Cowhide lost 38% of its ROAS and Brand lost 33%, against 56% to 63% for the three limited campaigns. Zombie, which reads 0% lost to budget, fell 59%.
- The extra spend bought little. Across the five non-brand campaigns, spend rose by £607 a day while Google's conversion value rose by only £126 a day (21 Sep to 18 Oct vs 27 Oct to 9 Nov). That looks like diminishing returns at higher spend as much as a cap.
- Day to day, there's almost no link. Pooling six campaigns and comparing each one only with itself, the correlation between lost to budget and ROAS is about −0.02 (n = 679 days).
- The August cap did no visible damage. From 24 Aug to 6 Sep Shopping High lost 49% to 67% of impressions to budget, and ROAS held at 4.82 and 3.87, close to its August norm of 4.5 to 4.95.
- Google is under-counting purchases since about 2 Nov. Google purchases per Shopify order fell from 0.58 to 0.39, which on its own knocks about a third off reported conversions. Add Q4 seasonality and the deliberate spend increase, and some of the ROAS fall would have happened anyway.
Next: test plan
A planned test, not a change made yet. The aim is to see whether clearing the budget limit with higher ROAS targets brings back efficient spend before Black Friday.
- Raise targets on limited campaignsStart with the campaign losing the smallest share to budget and work up, so each step is easy to read.
- Move Shopping High togetherIts ad groups share one budget, so all Shopping High targets move on the same day. Raising one just hands the budget to the others.
- Hold budgets, read twiceBudgets stay fixed. First read after 7 days, final read before Black Friday, judged on Triple Whale as well as Google.
- Clear stop rulesStep back if revenue drops beyond an agreed limit, and log other changes (tracking fixes, settings edits) so they don't muddy the result.