ROAS Optimization: Maximizing Return on Ad Spend Across All Channels

Google Ads reports a 3.52x median ROAS. Meta reports 1.86x. Put those two numbers in the same spreadsheet, add them up, divide by spend, and you get a blended figure that is wrong before you finish typing it. Not slightly wrong. Wrong at the foundation.

I have spent years optimising paid media across Google, Meta, and a long tail of smaller channels, and I will be blunt about the thing nobody wants to hear. Most ROAS optimisation is rearranging deck chairs. Teams tune bids, shift budgets, test creative, chase a target tROAS, and wonder why the blended number never quite reflects what hits the bank account. The reason is not the bidding strategy. The reason is the data feeding it.

This is not another ROAS tactics post. There are a hundred of those and they all assume your conversion data is clean. It is not. This is a post about why ROAS optimisation is a data-integrity problem first and a bidding problem a distant second, and why no amount of tROAS tuning fixes a corrupted conversion signal.

DataCops is the architectural fix. It is a first-party data layer that filters bots and junk at the point of collection, before any conversion event is sent to an ad platform via Meta CAPI or Google Ads CAPI. Clean signal in means honest ROAS out. Get that order wrong and everything downstream is optimisation theatre. For the calculator-side companion, see ROAS calculator tools and formulas.

Quick stuff people keep asking

What is a good ROAS across all channels in 2026? Benchmarks float around - Google near 3.5x median, Meta near 1.9x - but a blended target of 3x to 4x is common for e-commerce. Here is the catch: a "good" ROAS calculated on contaminated data is a good-looking number, not a good outcome. The benchmark only means something if the conversions underneath it are real.

How do I calculate blended ROAS across Google and Meta? Total revenue divided by total ad spend across every channel. Simple math. The hard part is the numerator. Platform-reported revenue double-counts conversions across channels and includes bot-driven events, so a naive blend inherits every distortion at once.

Why does my ROAS look different in each ad platform? Because each platform takes credit for the same sale. Google and Meta both claim a conversion if they both touched the buyer. Sum the platform numbers and you over-count. This attribution overlap is why platform-reported ROAS is structurally inflated above your real, verifiable revenue.

How do I optimize ROAS without increasing ad spend? Fix the conversion signal first. If 24 to 31% of the events feeding your campaigns are bots, cleaning that data improves ROAS without spending a cent more - because the algorithm finally optimises toward real buyers. That is the highest-leverage move available and almost nobody starts there.

What is the difference between ROAS and blended ROAS? Platform ROAS is what one ad platform reports for its own campaigns, scored by its own attribution. Blended ROAS is total revenue over total spend across everything. Blended is closer to the truth, but only if the revenue figure is verified rather than summed from platform claims.

How does bot traffic affect my ROAS calculations? Two ways. Bot-driven conversion events inflate the conversion count, so reported ROAS rises above reality. Worse, those bot events get sent to Meta and Google, whose models then optimise toward bot-shaped traffic - so bots corrupt both the number you read and the decisions the algorithm makes.

Should I use target ROAS bidding in Google Ads? tROAS works well when the conversion signal feeding it is clean. Feed it bot-contaminated data and you have automated bidding toward fake conversions at scale. The bidding strategy is fine. The input is the problem.

How do I allocate budget across channels to maximize ROAS? You cannot, not honestly, until your baseline is trustworthy. Allocating budget against inflated, double-counted, bot-contaminated ROAS just moves money toward whichever channel lies most flatteringly. Clean the data, then allocate.

The gap: ROAS is only as honest as the signal under it

Here is the structural problem, and it is Layer 5 - the layer where bad data does not just mislead you, it trains the machine to make itself worse.

ROAS is a ratio. Revenue over spend. Every optimisation guide treats the revenue number as solid ground and spends its energy on the spend side and the bidding logic. But the revenue number - the conversion signal - is built from events collected by third-party scripts and forwarded to ad platforms. And that signal is contaminated in three compounding ways.

One: attribution bias. Platforms over-credit themselves, especially bottom-funnel. Both Google and Meta will claim the same conversion. Your platform-reported ROAS is inflated above verifiable revenue before bots even enter the picture.

Two: bot contamination. Of the conversion-adjacent events that get collected, 24 to 31% are bots. Form fills, add-to-carts, signups, page events - automated traffic generates them, and your tracking counts them as human intent.

Three, and this is the one that turns a measurement error into a spiral: that contaminated signal gets sent server-side to Google and Meta via CAPI. Their machine learning models read every conversion event as a real human worth replicating. So they go find more traffic that looks like the events you sent.

If those events were bots, the model now hunts for more bot-shaped traffic. It reports that as conversions. ROAS looks fine. Then it does it again. Garbage in, garbage optimised, garbage out - and each cycle makes the next cycle worse, because the training data degrades every round.

Let me make it concrete. A team running a signup funnel at PillarlabAI set a honeypot - a clean funnel, real product, real tracking. 3,000 signups came through. 77% of them were fraud. 650 separate accounts traced to a single device fingerprint. One machine, presenting as 650 humans.

Now picture that funnel without the honeypot. Every one of those 3,000 signups fires a conversion event. Every one gets sent to Meta and Google via CAPI.

The platforms see 3,000 conversions, calculate a glorious ROAS, and their models go looking for 3,000 more people like that - except "people like that" means one device's fraud pattern. The algorithm was not optimising for customers. It was optimising for a fingerprint. And the ROAS dashboard called it a win the entire time.

That is why cross-channel ROAS optimisation consistently underdelivers. The whole exercise sits on a corrupted baseline. You can tune tROAS forever, A/B test creative forever, reallocate budget weekly forever - and none of it touches the fact that the conversion signal itself is part-fiction. You are optimising a number, not a business.

The root cause is the familiar one. Third-party scripts collect mixed data - human and bot, real and fake - with no isolation and no filtering before it leaves your infrastructure and hits the ad platforms. Once it is gone, you cannot un-send it. The model has already learned from it.

The fix is architectural, and it comes before bidding

If ROAS optimisation is a data-integrity problem, the fix is not a bidding tactic. It is an architecture that cleans the signal before it ships.

That means first-party collection on your own subdomain instead of third-party scripts scattered across the page - far more resilient, far less leaky. It means bot filtering at the point of ingestion, so automated traffic is identified and separated before any conversion event is forwarded. It means two tiers of data kept apart: anonymous session analytics, which are always lawful to collect, and identifiable conversion data, which is gated properly. And it means relaying only the cleaned, verified conversion signal onward to Meta, Google, TikTok, and LinkedIn via CAPI.

That is what DataCops is built to do. Bot filtering runs against a 361.8 billion-plus IP database that classifies residential, datacenter, VPN, proxy, and Tor traffic. The point is not to send the ad platforms more events.

It is to send them true ones. When the conversion signal is clean, the model trains on real buyers, ROAS reflects real revenue, and your tROAS targets and budget splits finally mean something. Bidding strategy becomes useful again - but only after the foundation is solid, never before.

Decision guide

Your blended ROAS never matches actual bank revenue: stop tuning bids - audit your conversion signal for bot contamination and attribution double-counting first.

You run tROAS or Advantage budget and results swing wildly: the algorithm is likely training on dirty data; clean the input before you touch the target.

Google ROAS looks far better than Meta ROAS: that gap is mostly attribution overlap - both platforms crediting the same sale - not a real channel-performance difference.

You want to improve ROAS without raising spend: filtering bots out of your conversion signal is the move; it lifts real ROAS at zero added budget.

You are about to reallocate budget across channels: do not, until your baseline is verified - allocating against inflated numbers just funds the best liar.

You send conversions server-side via CAPI: that is exactly the pipe that needs a filter in front of it, or you are training Meta and Google on whatever junk your scripts collected.

You are optimising a number, not a business

Here is the mistake, and it is nearly universal. Teams treat ROAS as the input to optimisation when it is the output of data quality. They open the dashboard, see the ratio, and start tuning - bids, budgets, creative, targets. All of it downstream. None of it touching the contaminated signal that produced the ratio in the first place.

A clean ROAS number is not a starting point you are handed. It is something you have to earn, by controlling what data reaches the algorithm. Skip that and every optimisation you run is just sophisticated guessing on top of fiction. The bidding engine is not broken. It is doing precisely what you asked - it is just doing it to the wrong data.

So go run the reconciliation. Take last month's total platform-reported revenue, take the actual money that landed in your accounts, and subtract. That gap is your contamination tax. Now ask the real question: how much of your ad budget is currently being optimised toward conversions that were never human at all?