Category: Positive Sparks News
What if your Google Ads account reports 10x ROAS, Meta reports 6x, and your total business revenue barely moves?
That is the question e-commerce businesses should answer before reallocating another pound of Q4 budget.
The holiday auction is too expensive for last-click attribution to decide your channel mix. In the latest Q4 benchmark data, US online holiday spending reached $257.8 billion, up 6.8% year over year. Google Search spend grew 13%, while Meta impressions increased 17% and spend rose 9%.
More demand is available, but competition is growing with it. The decision is no longer simply “Which platform reports the highest ROAS?” It is:
Which additional pound invested in Google or Meta creates the most revenue that would not have existed otherwise?
That is an incrementality question.
Why last-click ROAS is a weak Q4 allocation signal
Have you ever added together Google Ads revenue, Meta revenue, email revenue and direct traffic, only to discover that the platforms collectively claim more sales than your store generated?
That is not a reporting glitch. It is the predictable result of overlapping attribution systems.
Google may claim a sale after a branded search. Meta may claim the same customer after an impression or click. Your email platform may also claim the order. Last-click attribution then assigns the conversion to one final touchpoint, even though the customer may already have decided to buy.
This matters particularly in Q4 because branded demand and remarketing audiences expand rapidly. A shopper may:
- See a product on Meta.
- Search for the brand on Google several days later.
- Click a branded ad.
- Purchase after receiving an email reminder.
Which channel caused the purchase? Last click cannot answer that. It can only identify the final measurable interaction.
In our work with e-commerce accounts, we treat platform-reported ROAS as a useful optimisation signal, but not as sufficient evidence for capital allocation. High ROAS on branded search or short-window retargeting can indicate efficient demand capture rather than demand creation.

What does incrementality measure instead?
So, what would have happened if the ad had not been shown?
Incrementality testing estimates the answer by comparing an exposed test group with a statistically similar control group. The difference between the two represents the additional conversions or revenue caused by advertising.
The core calculation is:
Incremental revenue = revenue in the test group − estimated revenue in the control group
Then:
Incremental ROAS = incremental revenue ÷ incremental ad spend
Google’s Conversion Lift documentation uses the same principle. Its definition is:
Incremental conversion value divided by incremental cost equals iROAS.
If your test produces £60,000 in incremental revenue from £20,000 of additional spend, your incremental ROAS is 3.0x. The £60,000 is the value that would have been missed without the advertising intervention: not the total value of every order associated with the campaign.
That distinction changes how we judge Google and Meta.
Google may appear stronger because it captures shoppers already searching for a product. Meta may appear weaker because it introduces products before demand is explicit. Incrementality can reveal that Meta created more new demand, while Google efficiently harvested demand that was already forming.
Multimedia: Watch Google’s short explainer on incrementality testing for a visual explanation of test and control groups.
What does a controlled test reveal?
Would a high reported ROAS always survive a controlled test?
Consider a branded Google Search campaign that reports a remarkable ROAS. On the dashboard, the campaign can look untouchable.
The commercial question is whether these customers would have purchased without the branded ads.
A regional dark test can answer that. Keep the ads active in selected areas, and reduce or pause them in comparable control regions.
When paid branded search is reduced, organic sales often rise in the control areas to replace many of the paid conversions, so total company revenue may barely move.
A high reported ROAS can therefore translate into a much lower incremental return.
Paying for branded search can mean paying for customers who already knew what they wanted.
Budget can then be redirected toward campaigns designed to reach new audiences and generate additional demand. The result should be judged by total revenue, not by the platform report alone.
You can read the wider methodology in our article, Why Incrementality Testing Will Change the Way You Spend Your E-commerce Ad Budget.
This is why a commission-based model must be built around rigorous measurement. If an agency earns more when reported platform revenue increases, it can be tempting to optimise toward credit capture. If compensation is linked to agreed incremental growth above a baseline, the incentive is closer to the client’s actual commercial outcome.
Our approach to commission-based marketing starts with defining the baseline, data source, eligible channels and revenue rules before budget is scaled.
How should we test Google versus Meta before peak season?
What would a credible Q4 experiment look like if you need a decision before Black Friday?
The strongest design uses matched geographies and isolates the channel change. For larger e-commerce businesses, a four-cell framework can be useful:
| Test cell | Google Ads | Meta Ads |
|---|---|---|
| A | Reduced or paused | Reduced or paused |
| B | Active | Reduced or paused |
| C | Reduced or paused | Active |
| D | Active | Active |
This structure helps estimate:
- The incremental impact of Google alone.
- The incremental impact of Meta alone.
- The combined impact of both platforms.
- Potential overlap, synergy or cannibalisation.
In practice, fully pausing both channels may not be operationally appropriate. A sequential design is often safer:
- Test the marginal Google investment while Meta remains stable.
- Test the marginal Meta investment while Google remains stable.
- Compare incremental revenue, incremental orders and iROAS using the same backend source of truth.
We generally want 6–12 months of historical weekly revenue or order data by geography before selecting test markets. Test and control groups should track closely before launch; a pre-test difference of more than approximately 3–5% should trigger a redesign or rebalancing.
For high-volume brands, a 2–4 week user-level test may provide enough data. Geo-based studies often need longer, particularly when average order value is high or the purchase cycle extends beyond several days. A 10–20% holdout is a practical starting range, although the required size depends on baseline conversion volume and the minimum lift worth detecting.

Which metrics should decide the split?
Is the channel with the highest iROAS automatically the winner?
Not necessarily. We need to consider both efficiency and scale.
For each channel, measure:
- Incremental revenue.
- Incremental orders.
- Incremental customer acquisition cost.
- Incremental ROAS.
- New-to-file customer rate.
- Confidence interval around the result.
- Marginal iROAS as spend increases.
The final metric is especially important. A channel may produce a 4.0x iROAS at £20,000 of spend and only 2.0x at £100,000. Scaling changes the audience, auction pressure and quality of available inventory.
Your break-even iROAS should also reflect contribution margin. If your contribution margin before advertising is 40%, the revenue-based break-even ROAS is:
1 ÷ 0.40 = 2.5x
That means an incremental ROAS of 2.4x may look attractive in a platform report but lose money after product costs, fulfilment, returns and agency commission. Conversely, an incremental ROAS of 3.0x may be commercially viable even if the platform-reported ROAS is lower than your historical target.
Google’s reporting also includes confidence intervals. A point estimate of 2.8x with a range from 1.4x to 4.6x should not be treated with the same certainty as a result of 2.7x with a much narrower interval.
What should the Q4 reallocation rule be?
How do we turn the test into a budget decision without overreacting to noisy daily performance?
Use the test to establish a reallocation hierarchy:
- Protect spend where the lower confidence bound remains above your break-even iROAS.
- Increase the channel with the strongest marginal incremental return, not merely the highest reported ROAS.
- Cap branded search and hyper-short retargeting where control markets replace paid conversions with organic or direct sales.
- Reserve a defined test budget for new audiences, creative and landing pages.
- Recheck total business revenue, contribution profit and new-customer volume after each material budget change.
A possible Q4 planning structure is 50% toward proven incremental activity, 30% toward scalable experiments and 20% toward future bets. That is not a universal formula. It is a governance framework that prevents every pound from being committed to yesterday’s winners.
Meta prospecting may deserve more budget if it creates new-to-file customers at a profitable incremental CAC. Google Shopping or non-brand Search may deserve more if it captures high-intent demand with sustainable incremental returns. Branded Search and narrow retargeting should earn their budget through testing: not receive it automatically because the dashboard looks impressive.
For feed-dependent campaigns, review your Shopping feed infrastructure before the test begins. A product disapproval, incomplete feed or pricing mismatch can create an artificial channel failure that has nothing to do with incrementality.
Why does this matter in a commission-based model?
If the agency is paid a fixed fee, its commercial outcome may be disconnected from the client’s incremental profit. If the agency is paid on every attributed sale, it may be rewarded for harvesting existing demand.
A well-designed commission arrangement should define:
- The revenue baseline.
- The conversion source of truth.
- How returns, refunds and cancellations are treated.
- Which channels are under management.
- How incremental lift is validated.
- The commissionable revenue or profit definition.
- The testing calendar and decision rights.
That alignment forces better questions. Are we growing the business, or simply moving credit between dashboards? Are we creating new customers, or paying to intercept existing demand? Is the next £10,000 more productive on Google, Meta or neither?

The Q4 decision is causal, not cosmetic
Q4 does not give us much room for guesswork. Auction costs rise, delivery deadlines compress and promotional windows make daily results unusually volatile.
A reported 10x ROAS can be less valuable than a proven 2.5x iROAS if the first campaign mostly captures customers who were already going to buy. The right Google-versus-Meta split is therefore not a permanent percentage. It is the output of a measurement system that observes how each channel performs at the margin.
Start calibration early. Define your break-even economics. Match your markets. Keep the test clean. Use backend revenue. Report confidence intervals. Then reallocate budget toward the growth that survives the control group.
At Positive Sparks, our attribution analysis is built around that broader commercial view: not simply which platform claims a sale, but which activity moves total business revenue forward.
Are you currently allocating Q4 budget from last-click ROAS, blended MER or tested incremental return? Which channel would you expect to lose the most credit if you ran a regional holdout test: Google, Meta or both?