Q4 budget: how to split it between Google and Meta

Every October I get the same message: “Google’s at 4.2 ROAS and Meta’s at 3.8, so I’m moving budget to Google.” It sounds reasonable. The trouble is those two numbers aren’t counting the same sales, and if you add up the revenue each dashboard reports you get a figure your payment gateway doesn’t recognise. I’ve audited accounts where that sum comfortably exceeded the month’s actual takings.

This post is about one decision: how to split a fixed Q4 budget between Google and Meta when both platforms take credit for the same conversions. It isn’t the Q4 calendar or the sizing of the total budget, which I’ve covered elsewhere. It’s the allocation, which moves the most money and gets decided worst.

In 30 seconds:

• Google and Meta each count their own conversions, with different windows and different reporting dates. Their ROAS figures don’t add up.
• Google admits discrepancies of up to 20% between Google Ads and Analytics, two systems owned by the same company. Between rivals the gap is wider.
• The only number that doesn’t overlap is the real revenue in your payment gateway. Anchor the split there, not in each platform’s ROAS.
• Incrementality tests belong before the peak. In November you have neither the time nor the stability to read one.
• Conversion Lift isn’t available on every Google Ads account. If you can’t measure it, phase the split and use geography instead.

Why can’t I compare Google’s ROAS with Meta’s?

Because each platform only sees its own stretch of the journey and claims the whole sale. Someone sees an Instagram ad on Tuesday, searches your brand on Google on Thursday, and buys. Meta records a conversion. Google records another. You record one order. All three are right inside their own system and none of them describes what happened.

The size of the gap isn’t a matter of opinion. Google documents that between Google Ads and Google Analytics, discrepancies, often up to 20%, are expected because of the different attribution models. That’s two products from the same company, with access to the same website. So do you really expect Google and Meta, who share nothing at all, to reconcile?

The detail almost nobody internalises is that they don’t even report on the same date. Google Ads reports conversions against the date and time of the click that led to the action, not against the date of the purchase. Someone who clicks on 20 November and buys on 2 December lands in your November report. Your accounts will see them in December. That detail alone tends to break any day-by-day comparison during Black Friday, which is exactly when everyone’s refreshing the dashboard hourly.

There’s one more layer. Of the three attribution models available in Analytics reports, paid and organic last click ignores direct traffic and gives 100% of the value to the last channel the customer clicked through. Under that model, the visitor who came from Instagram and returned via a brand search is handed entirely to Google. Meta never finds out, and keeps counting them as its own.

Where exactly does the double-counting happen?

In the attribution window, and in the type of interaction each system accepts as credit. Meta credits conversions not only after a click but after a view: its marketing API defines fields like 1d_view, 7d_click and 28d_click, values measured one day after seeing the ad, or seven and twenty-eight days after clicking it. Google doesn’t count those views. It counts its own click.

This is the table I pull up whenever someone asks me to add the two dashboards together:

Look at the last row. It’s the only one that overlaps with nothing and the only one that pays wages. The rest are estimates produced by an interested party. I’m not saying they lie: I’m saying they answer a different question from the one you’re asking.

If you want the fine detail of how Meta assigns conversions and which window suits your buying cycle, I worked through it in the guide to Meta Ads attribution models. Here I care about the consequence: no arithmetic turns two platform ROAS figures into a correct split.

So what do I base the split on instead?

On the real revenue of the business divided by total ad spend. MER, if you like acronyms. It’s a blunt number and it won’t tell you which campaign worked. It’s also the only one no dashboard can inflate.

The routine I run with my clients is boring and it works. Every week I write down three business figures: total revenue, total spend, total orders. No platforms. Then I move budget from one to the other and watch what the MER does two or three weeks later. If I push Meta up by 30% and total revenue doesn’t move, that 30% wasn’t buying anything new, however pretty its dashboard ROAS looked.

That means having the underlying business metrics straight, not just the campaign ones. If your CAC and LTV are a fog, any split is an opinion dressed as analysis; I left a walkthrough of the three ecommerce metrics that actually matter, and it’s worth separating them from the vanity metrics that fill seasonal reports.

One warning: MER also misleads when your seasonality is strong, which in Q4 it is by definition. Compare November against November, never against October.

How do I test which platform actually drives sales?

With an experiment, and before the peak. Incrementality measurement exists precisely because reported ROAS doesn’t answer the question. Google describes Conversion Lift as an incrementality tool that measures the conversions driven directly by people seeing your ads, splitting users into a group that sees the ads and a group that doesn’t.

Now the awkward part. The same page says Conversion Lift isn’t available for all Google Ads accounts and that you need to contact your Google account representative. If you’re managing £3,000 a month, you don’t have a representative. The geographic version carries the same limitation: the documentation for geography-based Conversion Lift describes separating customers into comparable geographic groups, but access stays restricted.

So most advertisers won’t run a clean test. What I do in those cases is a homemade version, imperfect and good enough:

1. Pick your window in September or early October. A test launched on 20 November doesn’t measure your advertising, it measures Black Friday.
2. Split the country into two comparable blocks of regions by historic order volume, not by population. Uneven blocks invalidate the test.
3. Halve Meta in one block for three weeks and leave the other untouched. Halving beats switching off: switching off wrecks the learning.
4. Measure total revenue per block from your backend, not from any ad dashboard.
5. Close it and decide before 1 November. If you reach the last week of October without a conclusion, you’re not going to get one.

It isn’t a publishable experiment. There’s contamination between regions and a sample size a statistician would pity. But it usually gives you a directional signal, and that already beats comparing two ROAS figures that don’t speak the same language.

Do the two channels change roles when demand spikes?

Yes, and in a way that distorts the numbers just when you’re staring at them hardest. During a spike, searches for your brand rise on their own, pushed by the season, by your competitors, and by the email you sent on Tuesday. Google tends to collect that traffic and book it to its own results. Meta, whose job is interrupting people who weren’t looking for you, gets no such gift.

That produces an expensive illusion: in Q4 Google’s ROAS almost always improves, and if you reallocate in response to that you end up paying for demand you already had. I set out each platform’s role in peak season in the post on summer sales across Google and Meta; in November the logic is the same, only the intensity changes.

Something else that gets overlooked: moving budget in one lump costs you learning on both sides. A 40% cut in mid-November resets the automated bidding calibration at the most expensive moment of the year. Move it in steps, with days of margin; the guide to Smart Bidding in Google Ads explains why abrupt changes get punished.

How do I split the budget across pre-peak, peak and post-peak?

With three different allocations, because the three phases are chasing different things. Here’s the skeleton I use.

Pre-peak, October and the first half of November. Meta carries more weight than its reported ROAS justifies, deliberately: you’re filling audiences, buying reach before the auction heats up, and seeding the demand Google will harvest three weeks later. Judge this phase on immediate return and you’ll cut it, and you won’t always get that volume back.

Peak, Black Friday week. Google carries more weight here, because there’s explicit demand to capture and the cost of missing that auction is a sale lost today. Keep Meta running on remarketing and catalogue, but the incremental money goes to capture. This isn’t the moment to discover new audiences.

Post-peak, December. The split drifts back towards balance, with one caveat: December brings a later buyer who cares more about delivery deadlines. It depends on the sector, but in most cases you want to pull spend down before conversion rate falls, not after.

The full Q4 calendar, with dates and the preparation timeline, lives in the Q4 2026 advertising strategy. And if what you’re missing is sizing the total budget before you split it, the search-data method is in calculating your budget with Google Trends.

Frequently asked questions

Can I add Google Ads conversions to Meta conversions?

No. Each platform credits conversions to its own interactions, with different windows and different reporting dates. Meta can count a sale after a one-day view while Google counts that same sale after a click. Adding them gives you a total higher than your real orders, and any split built on that sum starts from a figure that doesn’t exist.

What’s the right Google-to-Meta split for Q4?

There’s no universal number, and be wary of anyone who hands you one. It depends on your brand awareness, on the search volume in your category, and on whether your bottleneck is capturing existing demand or creating it. What stays constant is the method: anchor the decision in real business revenue and adjust in steps, not jumps.

Is it worth running an incrementality test in November?

Almost never. A test needs a stable behavioural period to isolate the effect of advertising, and November is the opposite of that. If you didn’t launch it in September or October, fall back on the phased split and save the experiment for January, when demand returns to normal and the results are readable.

What is MER and why do you prefer it to platform ROAS?

It’s total business revenue divided by total ad spend. I prefer it because no platform with an interest in looking good calculates it: it comes from your own books. It doesn’t replace ROAS for optimising inside an account, but for deciding how much goes to each platform it’s the only common denominator the two share.

What if my business is small and I can’t split the country in two?

Then phase the split and skip the test. At low volume, any geographic trial gives you a result that can’t separate signal from noise, and deciding on that is worse than not deciding. Focus on watching the weekly MER and moving budget in small increments you can reverse without drama.

There’s no perfect split, but there is a traceable one

After a good few Q4s, my conclusion is that nobody knows the optimal split, myself included. What separates a well-run account isn’t getting the percentage right: it’s knowing what the decision was based on, being able to revisit it with a number that doesn’t come from the platform that benefits, and having the reflex to correct in steps when the business doesn’t respond.

If you change one thing this year, make it the number you look at. Stop opening two tabs to compare ROAS and look at total revenue against total spend, week by week. It’s less satisfying than a pretty dashboard, but at least it measures something real.

If you’d like us to go over your Q4 split with your actual numbers before November starts, book 30 minutes of consulting and we’ll look at it together.